├── .github ├── CODE_OF_CONDUCT.md └── workflows │ ├── docs.yml │ └── tests.yml ├── .gitignore ├── LICENSE ├── README.md ├── conf ├── archived_experiments │ ├── archived │ │ ├── finetune100k │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ │ ├── finetune100k_2 │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ │ ├── finetune100kpermute │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ │ ├── finetune300k │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ │ ├── finetune300kpermute │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ │ ├── finetune30k │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ │ ├── finetune30k_agai_old │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ │ ├── finetune30kpermute │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ │ ├── finetune_100k_atoms │ │ │ ├── atoms.yaml │ │ │ └── atoms_params.yaml │ │ ├── finetune_300k_atoms │ │ │ ├── atoms.yaml │ │ │ └── atoms_params.yaml │ │ ├── finetune_30k_again │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ │ ├── finetune_30k_atoms │ │ │ ├── atoms.yaml │ │ │ └── atoms_params.yaml │ │ ├── finetune_test │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ │ ├── finetune_zmatrix │ │ │ ├── zmatrix_100k.yaml │ │ │ ├── zmatrix_300k.yaml │ │ │ └── zmatrix_30k.yaml │ │ ├── pretrain100k_atoms │ │ │ ├── atoms.yaml │ │ │ └── atoms_params.yaml │ │ ├── pretrain2m │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ │ ├── pretrain300k_atoms │ │ │ ├── atoms.yaml │ │ │ └── atoms_params.yaml │ │ ├── pretrain30k │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ │ ├── pretrain30k_atoms │ │ │ ├── atoms.yaml │ │ │ └── atoms_params.yaml │ │ ├── pretrainzmatrix │ │ │ ├── zmatrix_100k.yaml │ │ │ ├── zmatrix_300k.yaml │ │ │ └── zmatrix_30k.yaml │ │ ├── testing_100k │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ │ └── testing_30k │ │ │ ├── cifp1.yaml │ │ │ ├── cifsymmetrized.yaml │ │ │ ├── composition.yaml │ │ │ ├── crystal_llm.yaml │ │ │ └── slice.yaml │ ├── config-hydra.yaml │ ├── config-potential.yaml │ ├── config-sft.yaml │ ├── config-smiles.yaml │ ├── config.yaml │ ├── finetune2m │ │ ├── atoms_params.yaml │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── finetune30k_rt │ │ ├── atoms_params.yaml │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── finetune30k_rt_2 │ │ ├── atoms.yaml │ │ ├── atoms_params.yaml │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── finetune_30k_spl_token │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ └── slice.yaml │ ├── llama_ft │ │ ├── cifp1.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── llama_sft │ │ ├── cifp1.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── llama_sft_10 │ │ ├── cifp1.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ └── slice.yaml │ ├── llama_sft_2 │ │ ├── cifp1.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── llama_sft_3 │ │ ├── cifp1.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── llama_sft_4 │ │ ├── cifp1.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── llama_sft_5 │ │ ├── cifp1.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── potential │ │ ├── cryst_0.yaml │ │ ├── cryst_0_2.yaml │ │ ├── cryst_0_4.yaml │ │ ├── cryst_0_5.yaml │ │ ├── cryst_0_6.yaml │ │ ├── cryst_0_8.yaml │ │ └── cryst_1.yaml │ ├── potential_2 │ │ ├── cryst_0.yaml │ │ ├── cryst_0_2.yaml │ │ ├── cryst_0_4.yaml │ │ ├── cryst_0_5.yaml │ │ ├── cryst_0_6.yaml │ │ ├── cryst_0_8.yaml │ │ └── cryst_1.yaml │ ├── potential_3 │ │ ├── cryst_0.yaml │ │ ├── cryst_0_2.yaml │ │ ├── cryst_0_4.yaml │ │ ├── cryst_0_5.yaml │ │ ├── cryst_0_6.yaml │ │ ├── cryst_0_8.yaml │ │ └── cryst_1.yaml │ ├── potential_4 │ │ ├── cryst_0.yaml │ │ ├── cryst_0_2.yaml │ │ ├── cryst_0_4.yaml │ │ ├── cryst_0_5.yaml │ │ ├── cryst_0_6.yaml │ │ ├── cryst_0_8.yaml │ │ └── cryst_1.yaml │ ├── potential_5 │ │ ├── cryst_0.yaml │ │ ├── cryst_0_2.yaml │ │ ├── cryst_0_4.yaml │ │ ├── cryst_0_5.yaml │ │ ├── cryst_0_6.yaml │ │ ├── cryst_0_8.yaml │ │ └── cryst_1.yaml │ ├── potential_6 │ │ ├── cryst_0.yaml │ │ ├── cryst_0_2.yaml │ │ ├── cryst_0_4.yaml │ │ ├── cryst_0_5.yaml │ │ ├── cryst_0_6.yaml │ │ ├── cryst_0_8.yaml │ │ └── cryst_1.yaml │ ├── potential_7 │ │ ├── cryst_0.yaml │ │ ├── cryst_0_2.yaml │ │ ├── cryst_0_4.yaml │ │ ├── cryst_0_5.yaml │ │ ├── cryst_0_6.yaml │ │ ├── cryst_0_8.yaml │ │ └── cryst_1.yaml │ ├── pretrain-rt │ │ ├── atoms.yaml │ │ ├── atoms_params.yaml │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── pretrain-test │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ └── slice.yaml │ ├── pretrain100k_spl_token │ │ ├── atoms_params.yaml │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── pretrain30k_spl_token │ │ ├── atoms_params.yaml │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── pretrain_2m │ │ ├── atoms.yaml │ │ ├── atoms_params.yaml │ │ └── zmatrix.yaml │ ├── pretrain_rt_100 │ │ ├── atoms.yaml │ │ ├── atoms_params.yaml │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── pretrain_rt_30 │ │ ├── atoms.yaml │ │ ├── atoms_params.yaml │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── pretrain_rt_300 │ │ ├── atoms.yaml │ │ ├── atoms_params.yaml │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── qmof_ft │ │ ├── atoms_params.yaml │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ ├── slice.yaml │ │ └── zmatrix.yaml │ ├── santiago │ │ └── cifp1.yaml │ ├── santiago_100k │ │ └── cifp1.yaml │ ├── santiago_2m │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ ├── composition.yaml │ │ ├── crystal_llm.yaml │ │ └── slice.yaml │ ├── smiles │ │ ├── slice_100.yaml │ │ ├── slice_2m.yaml │ │ ├── slice_30.yaml │ │ └── slice_300.yaml │ ├── testing_perturb │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ └── crystal_llm.yaml │ ├── testing_perturb_100 │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ └── crystal_llm.yaml │ ├── testing_perturb_300 │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ └── crystal_llm.yaml │ ├── testing_translate │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ └── crystal_llm.yaml │ ├── testing_translate_100 │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ └── crystal_llm.yaml │ └── testing_translate_300 │ │ ├── cifp1.yaml │ │ ├── cifsymmetrized.yaml │ │ └── crystal_llm.yaml ├── bandgap.yaml ├── benchmark.yaml ├── bg │ ├── atoms.yaml │ ├── atoms_params.yaml │ ├── cifp1.yaml │ ├── cifpsym.yaml │ ├── composition.yaml │ ├── crystal_llm.yaml │ ├── local_env.yaml │ ├── slices.yaml │ └── zmatrix.yaml ├── bg2m │ ├── atoms.yaml │ ├── atoms_params.yaml │ ├── cifp1.yaml │ ├── cifsymmetrized.yaml │ ├── composition.yaml │ ├── crystal_llm.yaml │ ├── local_env.yaml │ ├── slice.yaml │ └── zmatrix.yaml ├── classification.yaml ├── form │ ├── atoms.yaml │ ├── atoms_params.yaml │ ├── cifp1.yaml │ ├── cifpsym.yaml │ ├── composition.yaml │ ├── crystal_llm.yaml │ ├── local_env.yaml │ ├── slices.yaml │ └── zmatrix.yaml ├── form_energy.yaml ├── group-test │ ├── composition.yaml │ └── slices.yaml ├── is_metal │ ├── atoms.yaml │ ├── atoms_params.yaml │ ├── cifp1.yaml │ ├── cifpsym.yaml │ ├── composition.yaml │ ├── crystal_llm.yaml │ ├── local_env.yaml │ ├── slices.yaml │ └── zmatrix.yaml ├── llama_8b_bg │ ├── atoms.yaml │ ├── atoms_params.yaml │ ├── cifp1.yaml │ ├── cifpsym.yaml │ ├── composition.yaml │ ├── crystal_llm.yaml │ ├── local_env.yaml │ ├── slices.yaml │ └── zmatrix.yaml ├── llm_sft.yaml ├── model │ ├── archived_base │ │ ├── archived │ │ │ ├── finetune_template.yaml │ │ │ ├── finetune_template_dielectric.yaml │ │ │ ├── finetune_template_dielectric_atoms.yaml │ │ │ ├── finetune_template_dielectric_permuted.yaml │ │ │ ├── finetune_template_dielectric_test.yaml │ │ │ ├── finetune_template_dielectric_unfiltered.yaml │ │ │ ├── finetune_template_dielectric_zmatrix.yaml │ │ │ ├── finetune_template_gvrh.yaml │ │ │ ├── finetune_template_gvrh_atoms.yaml │ │ │ ├── finetune_template_gvrh_permuted copy.yaml │ │ │ ├── finetune_template_gvrh_permuted.yaml │ │ │ ├── finetune_template_gvrh_unfiltered.yaml │ │ │ ├── finetune_template_gvrh_zmatrix.yaml │ │ │ ├── finetune_template_jdft2d.yaml │ │ │ ├── finetune_template_kvrh.yaml │ │ │ ├── finetune_template_kvrh_atoms.yaml │ │ │ ├── finetune_template_kvrh_permuted copy.yaml │ │ │ ├── finetune_template_kvrh_permuted.yaml │ │ │ ├── finetune_template_kvrh_unfiltered.yaml │ │ │ ├── finetune_template_kvrh_zmatrix.yaml │ │ │ ├── finetune_template_perovskites.yaml │ │ │ ├── finetune_template_perovskites_atoms.yaml │ │ │ ├── finetune_template_perovskites_permuted copy.yaml │ │ │ ├── finetune_template_perovskites_permuted.yaml │ │ │ ├── finetune_template_perovskites_unfiltered.yaml │ │ │ ├── finetune_template_perovskites_zmatrix.yaml │ │ │ └── finetune_template_phonons.yaml │ │ ├── finetune_qmof.yaml │ │ ├── finetune_template_dielectric.yaml │ │ ├── finetune_template_dielectric_perturb.yaml │ │ ├── finetune_template_dielectric_potential.yaml │ │ ├── finetune_template_dielectric_smiles.yaml │ │ ├── finetune_template_dielectric_translate.yaml │ │ ├── finetune_template_gvrh.yaml │ │ ├── finetune_template_gvrh_perturb.yaml │ │ ├── finetune_template_gvrh_smiles.yaml │ │ ├── finetune_template_gvrh_translate.yaml │ │ ├── finetune_template_kvrh.yaml │ │ ├── finetune_template_kvrh_perturb.yaml │ │ ├── finetune_template_kvrh_potential.yaml │ │ ├── finetune_template_kvrh_smiles.yaml │ │ ├── finetune_template_kvrh_translate.yaml │ │ ├── finetune_template_perovskite_potential.yaml │ │ ├── finetune_template_perovskites.yaml │ │ ├── finetune_template_perovskites_perturb.yaml │ │ ├── finetune_template_perovskites_smiles.yaml │ │ ├── finetune_template_perovskites_translate.yaml │ │ ├── llama.yaml │ │ ├── llama3_gvrh_sft.yaml │ │ ├── llama_archived │ │ │ ├── llama_dielec.yaml │ │ │ ├── llama_gvrh.yaml │ │ │ ├── llama_kvrh.yaml │ │ │ └── llama_perov.yaml │ │ ├── llama_dielec.yaml │ │ ├── llama_dielectric_sft.yaml │ │ ├── llama_gvrh.yaml │ │ ├── llama_gvrh_sft.yaml │ │ ├── llama_gvrh_sft_nb_fold_2.yaml │ │ ├── llama_kvrh.yaml │ │ ├── llama_kvrh_sft.yaml │ │ ├── llama_perov.yaml │ │ ├── llama_perov_sft.yaml │ │ ├── pretrain_template.yaml │ │ └── pretrain_template_rt_token.yaml │ ├── benchmark_example.yaml │ ├── classification_example.yaml │ ├── formation_energy.yaml │ ├── llama_8b.yaml │ ├── llama_example.yaml │ ├── pretrain_example.yaml │ ├── pretrain_other_models.yaml │ └── pretrain_own_data_example.yaml └── pretrain.yaml ├── docs ├── api.md ├── benchmarking.md ├── getting_started.md ├── index.md ├── representations.md ├── static │ ├── logo.ai │ └── logo.png └── tokenizers.md ├── mkdocs.yml ├── notebooks ├── dataprep.ipynb ├── example_data │ ├── InCuS2_p1.cif │ ├── InCuS2_symmetrized.cif │ ├── N2_p1.cif │ ├── N2_symmetrized.cif │ ├── SrTiO3_p1.cif │ ├── SrTiO3_symmetrized.cif │ ├── TlCr5Se8_p1.cif │ └── TlCr5Se8_symmetrized.cif ├── example_mattext_representations.ipynb ├── linear_potential_dev.ipynb ├── tokens.ipynb └── tutorial.ipynb ├── pyproject.toml ├── revision-scripts ├── 5fold_split.py ├── matbench_is_metal.py ├── mp_classification.py ├── prep_json.py ├── prep_rep.py └── text_rep.py ├── scripts ├── filterdataset.py ├── linear_potential.py ├── llama-eval-responses │ ├── llama_evals_matbench_dielectric_cif_p1.json │ ├── llama_evals_matbench_dielectric_composition.json │ ├── llama_evals_matbench_dielectric_crystal_llm_rep.json │ ├── llama_evals_matbench_dielectric_slice.json │ ├── llama_evals_matbench_log_gvrh_cif_p1.json │ ├── llama_evals_matbench_log_gvrh_composition.json │ ├── llama_evals_matbench_log_gvrh_crystal_llm_rep.json │ ├── llama_evals_matbench_log_kvrh_cif_p1.json │ ├── llama_evals_matbench_log_kvrh_composition.json │ ├── llama_evals_matbench_log_kvrh_composition_old.json │ ├── llama_evals_matbench_log_kvrh_crystal_llm_rep.json │ ├── llama_evals_matbench_log_kvrh_slice.json │ ├── llama_evals_matbench_perovskites_cif_p1.json │ ├── llama_evals_matbench_perovskites_composition.json │ ├── llama_evals_matbench_perovskites_crystal_llm_rep.json │ └── llama_evals_matbench_perovskites_slice.json ├── llama_sft_evals.py ├── modify_mb_json.py ├── nomad_postprocess_parallel.py ├── qmof_prepare_data.py └── query_qmof.py ├── src └── mattext │ ├── __init__.py │ ├── analysis │ ├── __init__.py │ ├── attention.py │ └── xtal2pot.py │ ├── dataprep │ ├── __init__.py │ ├── download_matbench.py │ ├── matbench_prepare_data.py │ └── nomad_prepare_data.py │ ├── main.py │ ├── models │ ├── __init__.py │ ├── benchmark.py │ ├── finetune.py │ ├── helper.py │ ├── inference.py │ ├── llama.py │ ├── llama_sft.py │ ├── potential.py │ ├── predict.py │ ├── pretrain.py │ ├── score.py │ └── utils.py │ ├── representations │ ├── __init__.py │ ├── analysis.py │ ├── decoder.py │ └── transformations.py │ ├── tokenizer │ └── __init__.py │ └── utils.py └── tests ├── test_imports.py └── test_xtal2pot.py /.github/workflows/docs.yml: -------------------------------------------------------------------------------- 1 | name: Docs 2 | 3 | on: 4 | push: 5 | branches: [main] 6 | pull_request: 7 | workflow_dispatch: 8 | 9 | jobs: 10 | docs: 11 | runs-on: ubuntu-latest 12 | permissions: "write-all" 13 | 14 | steps: 15 | - uses: actions/checkout@v4 16 | 17 | - uses: actions/setup-python@v5 18 | with: 19 | python-version: "3.10" 20 | cache: pip 21 | cache-dependency-path: pyproject.toml 22 | 23 | - name: Install dependencies 24 | run: | 25 | python -m pip install --upgrade pip 26 | pip install wheel setuptools 27 | pip install -e . 28 | pip install mkdocs mkdocs-material "mkdocstrings[python]" mkdocs-autorefs 29 | 30 | - name: Build 31 | run: mkdocs build 32 | - run: mkdocs gh-deploy --force 33 | -------------------------------------------------------------------------------- /.github/workflows/tests.yml: -------------------------------------------------------------------------------- 1 | name: Tests 2 | 3 | on: 4 | push: 5 | branches: [ main ] 6 | pull_request: 7 | branches: [ main ] 8 | workflow_dispatch: 9 | 10 | jobs: 11 | tests: 12 | runs-on: ubuntu-latest 13 | strategy: 14 | fail-fast: true 15 | matrix: 16 | python-version: ["3.9"] 17 | timeout-minutes: 30 18 | defaults: 19 | run: 20 | shell: bash -l {0} 21 | steps: 22 | - name: Check out repository 23 | uses: actions/checkout@v4 24 | 25 | # - uses: pdm-project/setup-pdm@v3 26 | # name: Set up PDM 27 | # with: 28 | # python-version: ${{ matrix.python-version }} 29 | # cache: true 30 | 31 | - name: Setup Mambaforge 32 | uses: conda-incubator/setup-miniconda@v3 33 | with: 34 | miniforge-variant: Mambaforge 35 | miniforge-version: latest 36 | use-mamba: true 37 | python-version: ${{ matrix.python-version }} 38 | conda-channels: anaconda, conda-forge 39 | activate-environment: test 40 | 41 | - name: Install dependencies 42 | run: | 43 | mamba install -c conda-forge openbabel fftw -y 44 | pip install -e ".[dev]" 45 | pip install pyxtal 46 | pip install "numpy<2.0" 47 | 48 | - name: Test 49 | run: pytest tests -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | MIT License 2 | 3 | Copyright (c) 2024 LamaLab 4 | 5 | Permission is hereby granted, free of charge, to any person obtaining a copy 6 | of this software and associated documentation files (the "Software"), to deal 7 | in the Software without restriction, including without limitation the rights 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 9 | copies of the Software, and to permit persons to whom the Software is 10 | furnished to do so, subject to the following conditions: 11 | 12 | The above copyright notice and this permission notice shall be included in all 13 | copies or substantial portions of the Software. 14 | 15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 | SOFTWARE. -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100k/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: ft_100k_mb_small 6 | finetune: 7 | model_name: ft_100k_mb_small 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/pretrain100k/cif_p1/checkpoint-39000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100k/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: ft_100k_mb_small 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_symmetrized_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100k/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: ft_100k_mb_small 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 512 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/pretrain100k/composition/checkpoint-2000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100k/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: ft_100k_mb_small 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/pretrain100k/crystal_llm/checkpoint-156000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100k/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: ft_100k_mb_small 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/pretrain100k/slice/checkpoint-39000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100k_2/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: ft_100k_mb_small_test 6 | finetune: 7 | model_name: ft_100k_mb_small 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100k_2/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: ft_100k_mb_small_test 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_symmetrized_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100k_2/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: ft_100k_mb_small_test 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 512 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/composition_pt_30k_wes/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100k_2/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: ft_100k_mb_small_test 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100k_2/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: ft_100k_mb_small_test 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100kpermute/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: permute_100k_7seed 6 | finetune: 7 | model_name: ft_100k_mb_small 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100kpermute/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: permute_100k_7seed 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_symmetrized_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100kpermute/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: permute_100k_7seed 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 512 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/composition_pt_30k_wes/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100kpermute/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: permute_100k_7seed 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune100kpermute/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: permute_100k_7seed 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune300k/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: ft_300k_mb_small 6 | finetune: 7 | model_name: ft_300k_mb_small 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/pretrain300k/cif_p1_pt_300k_wes_2/checkpoint-58000 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune300k/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: ft_300k_mb_small 6 | 7 | finetune: 8 | model_name: ft_300k_mb_small 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/pretrain300k/cif_symmetrized_pt_300k_wes_2/checkpoint-58000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune300k/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: ft_300k_mb_small 6 | 7 | finetune: 8 | model_name: ft_300k_mb_small 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 512 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/pretrain300k/composition_pt_300/checkpoint-7000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune300k/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: ft_300k_mb_small 6 | 7 | finetune: 8 | model_name: ft_300k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/pretrain300k/crystal_llm_rep_pt_300/checkpoint-57000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune300k/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: ft_300k_mb_small 6 | 7 | finetune: 8 | model_name: ft_300k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/pretrain300k/slice_pt_300/checkpoint-117000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune300kpermute/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: permute_300k_7seed 6 | finetune: 7 | model_name: ft_300k_mb_small 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/pretrain300k/cif_p1_pt_300k_wes_2/checkpoint-58000 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune300kpermute/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: permute_300k_7seed 6 | finetune: 7 | model_name: ft_300k_mb_small 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/pretrain300k/cif_symmetrized_pt_300k_wes_2/checkpoint-58000 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune300kpermute/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: permute_300k_7seed 6 | finetune: 7 | model_name: ft_300k_mb_small 8 | context_length: 32 9 | training_arguments: 10 | per_device_train_batch_size: 512 11 | path: 12 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/pretrain300k/composition_pt_300/checkpoint-7000 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune300kpermute/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: permute_300k_7seed 6 | finetune: 7 | model_name: ft_300k_mb_small 8 | context_length: 512 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/pretrain300k/crystal_llm_rep_pt_300/checkpoint-57000 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune300kpermute/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: permute_300k_7seed 6 | finetune: 7 | model_name: ft_300k_mb_small 8 | context_length: 512 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/pretrain300k/slice_pt_300/checkpoint-117000 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30k/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: pt_30k_mb_small 6 | finetune: 7 | model_name: pt_30k_mb_small 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30k/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: pt_30k_mb_small 6 | 7 | finetune: 8 | model_name: pt_30k_mb_small 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_symmetrized_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30k/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: pt_30k_mb_small 6 | 7 | finetune: 8 | model_name: pt_30k_mb_small 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 512 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/composition_pt_30k_wes/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30k/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: pt_30k_mb_small 6 | 7 | finetune: 8 | model_name: pt_30k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30k/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: pt_30k_mb_small 6 | 7 | finetune: 8 | model_name: pt_30k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30k_agai_old/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: ft_30k_unfiltered_again 6 | finetune: 7 | model_name: ft_30k_unfiltered_again 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30k_agai_old/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: ft_30k_unfiltered_again 6 | 7 | finetune: 8 | model_name: ft_30k_unfiltered_again 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_symmetrized_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30k_agai_old/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: ft_30k_unfiltered_again 6 | 7 | finetune: 8 | model_name: ft_30k_unfiltered_again 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 512 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/composition_pt_30k_wes/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30k_agai_old/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: ft_30k_unfiltered_again 6 | 7 | finetune: 8 | model_name: ft_30k_unfiltered_again 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30k_agai_old/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: ft_30k_unfiltered_again 6 | 7 | finetune: 8 | model_name: ft_30k_unfiltered_again 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30kpermute/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: permute_30k_7seed 6 | finetune: 7 | model_name: pt_30k_mb_small 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30kpermute/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: permute_30k_7seed 6 | 7 | finetune: 8 | model_name: pt_30k_mb_small 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_symmetrized_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30kpermute/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: permute_30k_7seed 6 | 7 | finetune: 8 | model_name: pt_30k_mb_small 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 512 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/composition_pt_30k_wes/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30kpermute/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: permute_30k_7seed 6 | 7 | finetune: 8 | model_name: pt_30k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune30kpermute/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: permute_30k_7seed 6 | 7 | finetune: 8 | model_name: pt_30k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_100k_atoms/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms 4 | logging: 5 | wandb_project: ft_100k_atoms 6 | 7 | finetune: 8 | model_name: ft_100k_atoms 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_pt_100k_atoms/checkpoint-4000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_100k_atoms/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | logging: 5 | wandb_project: ft_100k_atoms_params 6 | 7 | finetune: 8 | model_name: ft_100k_atoms_params 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_params_pt_100k_atoms/checkpoint-4000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_300k_atoms/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms 4 | logging: 5 | wandb_project: ft_300k_atoms 6 | 7 | finetune: 8 | model_name: ft_300k_atoms 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_pt_300k_atoms/checkpoint-14000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_300k_atoms/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | logging: 5 | wandb_project: ft_300k_atoms_params 6 | 7 | finetune: 8 | model_name: ft_300k_atoms_params 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_params_pt_300k_atoms/checkpoint-14000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_30k_again/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: ft_30k_unfiltered_again 6 | finetune: 7 | model_name: ft_30k_unfiltered_again 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_30k_again/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: ft_30k_unfiltered_again 6 | 7 | finetune: 8 | model_name: ft_30k_unfiltered_again 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_symmetrized_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_30k_again/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: ft_30k_unfiltered_again 6 | 7 | finetune: 8 | model_name: ft_30k_unfiltered_again 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 512 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/composition_pt_30k_wes/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_30k_again/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: ft_30k_unfiltered_again 6 | 7 | finetune: 8 | model_name: ft_30k_unfiltered_again 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_30k_again/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: ft_30k_unfiltered_again 6 | 7 | finetune: 8 | model_name: ft_30k_unfiltered_again 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_30k_atoms/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms 4 | logging: 5 | wandb_project: ft_30k_atoms 6 | 7 | finetune: 8 | model_name: ft_30k_atoms 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_pt_30k_atoms/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_30k_atoms/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | logging: 5 | wandb_project: ft_30k_atoms_params 6 | 7 | finetune: 8 | model_name: ft_30k_atoms_params 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_params_pt_30k_atoms/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_test/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | finetune: 5 | context_length: 1024 6 | training_arguments: 7 | per_device_train_batch_size: 64 8 | path: 9 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 10 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_test/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | finetune: 5 | context_length: 1024 6 | training_arguments: 7 | per_device_train_batch_size: 64 8 | path: 9 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_symmetrized_pt_30k_wes/checkpoint-46000" 10 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_test/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | finetune: 5 | context_length: 32 6 | training_arguments: 7 | per_device_train_batch_size: 512 8 | path: 9 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/composition_pt_30k_wes/checkpoint-1000 10 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_test/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | finetune: 5 | context_length: 512 6 | training_arguments: 7 | per_device_train_batch_size: 64 8 | path: 9 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 10 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_test/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | finetune: 5 | context_length: 512 6 | training_arguments: 7 | per_device_train_batch_size: 64 8 | path: 9 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 10 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_zmatrix/zmatrix_100k.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: zmatrix_finetune 6 | 7 | finetune: 8 | model_name: ft_100k_zmatrix 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_100k_zmatrix/checkpoint-4000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_zmatrix/zmatrix_300k.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: zmatrix_finetune 6 | 7 | finetune: 8 | model_name: ft_300k_zmatrix 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_300k_zmatrix/checkpoint-14000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/finetune_zmatrix/zmatrix_30k.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: zmatrix_finetune 6 | 7 | finetune: 8 | model_name: ft_30k_zmatrix 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_30k_zmatrix/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain100k_atoms/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_atoms 5 | 6 | representation: atoms 7 | pretrain: 8 | name: pt_100k_atoms 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain100k_atoms/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_atoms 5 | 6 | representation: atoms_params 7 | pretrain: 8 | name: pt_100k_atoms 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain2m/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_2m_wes 5 | 6 | representation: cif_p1 7 | pretrain: 8 | name: pt_2m_wes 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/2m 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain2m/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_2m_wes 5 | 6 | representation: cif_symmetrized 7 | pretrain: 8 | name: pt_2m_wes 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/2m 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain2m/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_2m_wes 5 | 6 | representation: composition 7 | pretrain: 8 | name: pt_2m_wes 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/2m 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain2m/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_2m_wes 5 | 6 | representation: crystal_llm_rep 7 | pretrain: 8 | name: pt_2m_wes 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/2m 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain2m/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_2m_wes 5 | 6 | representation: slice 7 | pretrain: 8 | name: pt_2m_wes 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/2m 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain300k_atoms/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_300k_atoms 5 | 6 | representation: atoms 7 | pretrain: 8 | name: pt_300k_atoms 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/300k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain300k_atoms/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_300k_atoms 5 | 6 | representation: atoms_params 7 | pretrain: 8 | name: pt_300k_atoms 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/300k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain30k/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | pretrain: 5 | context_length: 1024 6 | training_arguments: 7 | per_device_train_batch_size: 32 8 | path: 9 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k 10 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain30k/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | pretrain: 5 | context_length: 1024 6 | training_arguments: 7 | per_device_train_batch_size: 32 8 | path: 9 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k 10 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain30k/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | pretrain: 5 | context_length: 32 6 | training_arguments: 7 | per_device_train_batch_size: 1024 8 | path: 9 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k 10 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain30k/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | pretrain: 5 | context_length: 512 6 | training_arguments: 7 | per_device_train_batch_size: 32 8 | path: 9 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k 10 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain30k/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | pretrain: 5 | context_length: 512 6 | training_arguments: 7 | per_device_train_batch_size: 32 8 | path: 9 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k 10 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain30k_atoms/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_30k_atoms 5 | 6 | representation: atoms 7 | pretrain: 8 | name: pt_30k_atoms 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrain30k_atoms/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_30k_atoms 5 | 6 | representation: atoms_params 7 | pretrain: 8 | name: pt_30k_atoms 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrainzmatrix/zmatrix_100k.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_zmatrix 5 | 6 | representation: zmatrix 7 | pretrain: 8 | name: pt_100k_zmatrix 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k/zmatrix 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrainzmatrix/zmatrix_300k.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_zmatrix 5 | 6 | representation: zmatrix 7 | pretrain: 8 | name: pt_300k_zmatrix 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/300k/zmatrix 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/pretrainzmatrix/zmatrix_30k.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_zmatrix 5 | 6 | representation: zmatrix 7 | pretrain: 8 | name: pt_30k_zmatrix 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k/zmatrix 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/testing_100k/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: ft_100k_mb_small_test 6 | finetune: 7 | model_name: ft_100k_mb_small 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/testing_100k/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: ft_100k_mb_small_test 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_symmetrized_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/testing_100k/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: ft_100k_mb_small_test 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 512 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/composition_pt_30k_wes/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/testing_100k/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: ft_100k_mb_small_test 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/testing_100k/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: ft_100k_mb_small_test 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/testing_30k/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: pt_30k_mb_test 6 | finetune: 7 | model_name: finetune_30k_wes_3 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/testing_30k/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: pt_30k_mb_test 6 | 7 | finetune: 8 | model_name: finetune_30k_wes_3 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_symmetrized_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/testing_30k/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: pt_30k_mb_test 6 | 7 | finetune: 8 | model_name: finetune_30k_wes_3 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 512 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/composition_pt_30k_wes/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/testing_30k/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: pt_30k_mb_test 6 | 7 | finetune: 8 | model_name: finetune_30k_wes_3 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/archived/testing_30k/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: pt_30k_mb_test 6 | 7 | finetune: 8 | model_name: finetune_30k_wes_3 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/config-hydra.yaml: -------------------------------------------------------------------------------- 1 | hydra: 2 | job: 3 | name: finetune 4 | run: 5 | dir: ${hydra:runtime.cwd}/outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} 6 | sweep: 7 | dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} 8 | subdir: ${hydra.job.override_dirname} 9 | 10 | launcher: 11 | _target_: hydra_plugins.hydra_submitit_launcher.submitit_launcher.SlurmLauncher 12 | submitit_folder: ${hydra.sweep.dir}/.submitit/%j 13 | timeout_min: 3600 14 | mem_gb: 160 15 | nodes: 1 16 | #gpus_per_task: 1 17 | gres: gpu:1 18 | #gpus_per_node: 2 19 | name: ${hydra.job.name} 20 | partition: 'gpu' 21 | additional_parameters: 22 | nodelist: 'gpu[005-007,013-014]' 23 | tasks_per_node: 1 24 | 25 | 26 | defaults: 27 | - model: none 28 | - override hydra/launcher: submitit_slurm 29 | 30 | runs: 31 | 32 | # - name: pretrain_run 33 | # tasks: [pretrain] 34 | 35 | - name: benchmark_run 36 | tasks: [benchmark] 37 | 38 | # - name: test_run 39 | # tasks: [inference] 40 | -------------------------------------------------------------------------------- /conf/archived_experiments/config-potential.yaml: -------------------------------------------------------------------------------- 1 | hydra: 2 | job: 3 | name: llama_instruct 4 | run: 5 | dir: ${hydra:runtime.cwd}/outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} 6 | sweep: 7 | dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} 8 | subdir: ${hydra.job.override_dirname} 9 | 10 | launcher: 11 | _target_: hydra_plugins.hydra_submitit_launcher.submitit_launcher.SlurmLauncher 12 | submitit_folder: ${hydra.sweep.dir}/.submitit/%j 13 | timeout_min: 3600 14 | mem_gb: 160 15 | nodes: 1 16 | #gpus_per_task: 1 17 | gres: gpu:1 18 | #gpus_per_node: 2 19 | name: ${hydra.job.name} 20 | partition: 'gpu' 21 | additional_parameters: 22 | nodelist: 'gpu[005,006,007,013-014]' 23 | tasks_per_node: 1 24 | 25 | 26 | defaults: 27 | - model: none 28 | - override hydra/launcher: submitit_slurm 29 | 30 | 31 | 32 | runs: 33 | 34 | # - name: pretrain_run 35 | # tasks: [pretrain] 36 | 37 | # - name: benchmark_run 38 | # tasks: [benchmark] 39 | 40 | # - name: test_run 41 | # tasks: [inference] 42 | 43 | # - name: qmof_run 44 | # tasks: [qmof] 45 | 46 | # - name: llama_run 47 | # tasks: [llama] 48 | 49 | # - name: llama_sft_run 50 | # tasks: [llama_sft] 51 | 52 | - name: potential_run 53 | tasks: [potential] 54 | 55 | -------------------------------------------------------------------------------- /conf/archived_experiments/config-sft.yaml: -------------------------------------------------------------------------------- 1 | hydra: 2 | job: 3 | name: llama_instruct 4 | run: 5 | dir: ${hydra:runtime.cwd}/outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} 6 | sweep: 7 | dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} 8 | subdir: ${hydra.job.override_dirname} 9 | 10 | launcher: 11 | _target_: hydra_plugins.hydra_submitit_launcher.submitit_launcher.SlurmLauncher 12 | submitit_folder: ${hydra.sweep.dir}/.submitit/%j 13 | timeout_min: 3600 14 | mem_gb: 160 15 | nodes: 1 16 | #gpus_per_task: 1 17 | gres: gpu:1 18 | #gpus_per_node: 2 19 | name: ${hydra.job.name} 20 | partition: 'gpu' 21 | additional_parameters: 22 | nodelist: 'gpu[005,007,013-014]' 23 | tasks_per_node: 1 24 | 25 | 26 | defaults: 27 | - model: none 28 | - override hydra/launcher: submitit_slurm 29 | 30 | 31 | 32 | runs: 33 | 34 | # - name: pretrain_run 35 | # tasks: [pretrain] 36 | 37 | # - name: benchmark_run 38 | # tasks: [benchmark] 39 | 40 | # - name: test_run 41 | # tasks: [inference] 42 | 43 | # - name: qmof_run 44 | # tasks: [qmof] 45 | 46 | # - name: llama_run 47 | # tasks: [llama] 48 | 49 | - name: llama_sft_run 50 | tasks: [llama_sft] 51 | 52 | # - name: potential_run 53 | # tasks: [potential] 54 | 55 | -------------------------------------------------------------------------------- /conf/archived_experiments/config-smiles.yaml: -------------------------------------------------------------------------------- 1 | hydra: 2 | job: 3 | name: llama_instruct 4 | run: 5 | dir: ${hydra:runtime.cwd}/outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} 6 | sweep: 7 | dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} 8 | subdir: ${hydra.job.override_dirname} 9 | 10 | launcher: 11 | _target_: hydra_plugins.hydra_submitit_launcher.submitit_launcher.SlurmLauncher 12 | submitit_folder: ${hydra.sweep.dir}/.submitit/%j 13 | timeout_min: 3600 14 | mem_gb: 160 15 | nodes: 1 16 | #gpus_per_task: 1 17 | gres: gpu:1 18 | #gpus_per_node: 2 19 | name: ${hydra.job.name} 20 | partition: 'gpu' 21 | additional_parameters: 22 | nodelist: 'gpu[005,006,007,013-014]' 23 | tasks_per_node: 1 24 | 25 | 26 | defaults: 27 | - model: none 28 | - override hydra/launcher: submitit_slurm 29 | 30 | 31 | 32 | runs: 33 | 34 | # - name: pretrain_run 35 | # tasks: [pretrain] 36 | 37 | - name: benchmark_run 38 | tasks: [benchmark] 39 | 40 | # - name: test_run 41 | # tasks: [inference] 42 | 43 | # - name: qmof_run 44 | # tasks: [qmof] 45 | 46 | # - name: llama_run 47 | # tasks: [llama] 48 | 49 | # - name: llama_sft_run 50 | # tasks: [llama_sft] 51 | 52 | # - name: potential_run 53 | # tasks: [potential] 54 | 55 | -------------------------------------------------------------------------------- /conf/archived_experiments/config.yaml: -------------------------------------------------------------------------------- 1 | hydra: 2 | job: 3 | name: llama_sft 4 | run: 5 | dir: ${hydra:runtime.cwd}/outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} 6 | sweep: 7 | dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} 8 | subdir: ${hydra.job.override_dirname} 9 | 10 | launcher: 11 | _target_: hydra_plugins.hydra_submitit_launcher.submitit_launcher.SlurmLauncher 12 | submitit_folder: ${hydra.sweep.dir}/.submitit/%j 13 | timeout_min: 3600 14 | mem_gb: 160 15 | nodes: 1 16 | #gpus_per_task: 1 17 | gres: gpu:1 18 | #gpus_per_node: 2 19 | name: ${hydra.job.name} 20 | partition: 'gpu' 21 | additional_parameters: 22 | nodelist: 'gpu[005,007,013-014]' 23 | tasks_per_node: 1 24 | 25 | 26 | defaults: 27 | - model: none 28 | - override hydra/launcher: submitit_slurm 29 | 30 | 31 | 32 | runs: 33 | 34 | # - name: pretrain_run 35 | # tasks: [pretrain] 36 | 37 | # - name: benchmark_run 38 | # tasks: [benchmark] 39 | 40 | # - name: test_run 41 | # tasks: [inference] 42 | 43 | # - name: qmof_run 44 | # tasks: [qmof] 45 | 46 | # - name: llama_run 47 | # tasks: [llama] 48 | 49 | - name: llama_sft_run 50 | tasks: [llama_sft] 51 | 52 | # - name: potential_run 53 | # tasks: [potential] 54 | 55 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune2m/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_params_pt_30k_atoms/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune2m/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/cif_p1_pt_30k_rt_2/checkpoint-46000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune2m/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/cif_symmetrized_pt_30k_rt/checkpoint-45000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune2m/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/composition_pt_30k_rt/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune2m/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/crystal_llm_rep_pt_30k_rt/checkpoint-11000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune2m/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/slice_pt_30k_rt/checkpoint-23000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune2m/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_params_pt_30k_atoms/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | logging: 5 | wandb_project: 30k_ft 6 | 7 | finetune: 8 | model_name: 30k_ft 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_params_pt_30k_atoms/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: 30k_ft 6 | 7 | finetune: 8 | model_name: 30k_ft 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/cif_p1_pt_30k_rt_2/checkpoint-46000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: 30k_ft 6 | 7 | finetune: 8 | model_name: 30k_ft 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/cif_symmetrized_pt_30k_rt/checkpoint-45000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: 30k_ft 6 | 7 | finetune: 8 | model_name: 30k_ft 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/composition_pt_30k_rt/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: 30k_ft 6 | 7 | finetune: 8 | model_name: 30k_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/crystal_llm_rep_pt_30k_rt/checkpoint-11000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: 30k_ft 6 | 7 | finetune: 8 | model_name: 30k_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/slice_pt_30k_rt/checkpoint-23000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: 30k_ft 6 | 7 | finetune: 8 | model_name: 30k_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_params_pt_30k_atoms/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt_2/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms 4 | special_num_token: True 5 | logging: 6 | wandb_project: 30k_ft_rt2 7 | 8 | finetune: 9 | model_name: 30k_ft_rt2 10 | context_length: 32 11 | training_arguments: 12 | per_device_train_batch_size: 1024 13 | path: 14 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/atoms_pt_30k_rt/checkpoint-1000 15 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt_2/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | special_num_token: True 5 | logging: 6 | wandb_project: 30k_ft_rt2 7 | 8 | finetune: 9 | model_name: 30k_ft_rt2 10 | context_length: 32 11 | training_arguments: 12 | per_device_train_batch_size: 1024 13 | path: 14 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/atoms_params_pt_30k_rt/checkpoint-1000 15 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt_2/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: 30k_ft_rt2 6 | 7 | finetune: 8 | model_name: 30k_ft_rt2 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/cif_p1_pt_30k_rt_2/checkpoint-46000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt_2/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: 30k_ft_rt2 6 | 7 | finetune: 8 | model_name: 30k_ft_rt2 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/cif_symmetrized_pt_30k_rt/checkpoint-46000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt_2/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: 30k_ft_rt2 6 | 7 | finetune: 8 | model_name: 30k_ft_rt2 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/composition_pt_30k_rt/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt_2/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: 30k_ft_rt2 6 | 7 | finetune: 8 | model_name: 30k_ft_rt2 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/crystal_llm_rep_pt_30k_rt/checkpoint-11000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt_2/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: 30k_ft_rt2 6 | 7 | finetune: 8 | model_name: 30k_ft_rt2 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/slice_pt_30k_rt/checkpoint-23000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune30k_rt_2/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | special_num_token: True 5 | logging: 6 | wandb_project: 30k_ft_rt2 7 | 8 | finetune: 9 | model_name: 30k_ft_rt2 10 | context_length: 512 11 | training_arguments: 12 | per_device_train_batch_size: 64 13 | path: 14 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/zmatrix_pt_30k_rt/checkpoint-46000 15 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune_30k_spl_token/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: ft_30k_spl 6 | finetune: 7 | model_name: ft_30k_spl 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/cif_p1_30k_ft/checkpoint-46000 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune_30k_spl_token/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: ft_30k_spl 6 | 7 | finetune: 8 | model_name: ft_30k_spl 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/cif_symmetrized_30k_ft/checkpoint-46000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune_30k_spl_token/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: ft_30k_spl 6 | 7 | finetune: 8 | model_name: ft_30k_spl 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 512 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/composition_30k_ft/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune_30k_spl_token/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: ft_30k_spl 6 | 7 | finetune: 8 | model_name: ft_30k_spl 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/crystal_llm_rep_30k_ft/checkpoint-11000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/finetune_30k_spl_token/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: ft_30k_spl 6 | 7 | finetune: 8 | model_name: ft_30k_spl 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/slice_30k_ft/checkpoint-11000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_ft/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: lama_2 6 | finetune: 7 | model_name: lama_2 8 | # context_length: 1024 9 | # training_arguments: 10 | # per_device_train_batch_size: 64 11 | # path: 12 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_ft/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: lama_2 6 | 7 | finetune: 8 | model_name: lama_2 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 64 12 | # path: 13 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_ft/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: lama_2 6 | 7 | finetune: 8 | model_name: lama_2 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 64 12 | # path: 13 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_ft/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: lama_2 6 | 7 | finetune: 8 | model_name: lama_2 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 128 12 | # path: 13 | # pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_30k_zmatrix/checkpoint-23000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: llama_instruct_2 6 | finetune: 7 | model_name: llama_instruct_2 8 | # context_length: 1024 9 | # training_arguments: 10 | # per_device_train_batch_size: 64 11 | # path: 12 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: llama_instruct_2 6 | finetune: 7 | model_name: llama_instruct_3 8 | # context_length: 1024 9 | # training_arguments: 10 | # per_device_train_batch_size: 64 11 | # path: 12 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: llama_instruct_2 6 | 7 | finetune: 8 | model_name: llama_instruct_2 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 64 12 | # path: 13 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: llama_instruct_2 6 | 7 | finetune: 8 | model_name: llama_instruct_2 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 64 12 | # path: 13 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: llama_instruct 6 | 7 | finetune: 8 | model_name: llama_instruct 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 128 12 | # path: 13 | # pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_30k_zmatrix/checkpoint-23000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_10/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: llama3_8_gen 6 | finetune: 7 | path: 8 | pretrained_checkpoint: "meta-llama/Meta-Llama-3-8B" 9 | model_name: llama3_8_gen 10 | # context_length: 1024 11 | training_arguments: 12 | per_device_train_batch_size: 16 13 | # path: 14 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 15 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_10/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: llama3_8_gen 6 | finetune: 7 | model_name: llama3_8_gen 8 | path: 9 | pretrained_checkpoint: "meta-llama/Meta-Llama-3-8B" 10 | # context_length: 1024 11 | # training_arguments: 12 | # per_device_train_batch_size: 64 13 | # path: 14 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 15 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_10/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: llama3_8_gen 6 | 7 | finetune: 8 | model_name: llama3_8_gen 9 | path: 10 | pretrained_checkpoint: "meta-llama/Meta-Llama-3-8B" 11 | # context_length: 512 12 | # training_arguments: 13 | # per_device_train_batch_size: 64 14 | # path: 15 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_10/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: llama3_8_gen 6 | 7 | finetune: 8 | model_name: llama3_8_gen 9 | path: 10 | pretrained_checkpoint: "meta-llama/Meta-Llama-3-8B" 11 | # context_length: 512 12 | # training_arguments: 13 | # per_device_train_batch_size: 64 14 | # path: 15 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_2/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: llama_collator_2 6 | finetune: 7 | model_name: llama_collator_2 8 | # context_length: 1024 9 | # training_arguments: 10 | # per_device_train_batch_size: 64 11 | # path: 12 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_2/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: llama_collator_2 6 | finetune: 7 | model_name: llama_collator_2 8 | # context_length: 1024 9 | # training_arguments: 10 | # per_device_train_batch_size: 64 11 | # path: 12 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_2/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: llama_collator_2 6 | 7 | finetune: 8 | model_name: llama_collator_2 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 64 12 | # path: 13 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_2/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: llama_collator_2 6 | 7 | finetune: 8 | model_name: llama_collator_2 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 64 12 | # path: 13 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_2/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: llama_instruct 6 | 7 | finetune: 8 | model_name: llama_instruct 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 128 12 | # path: 13 | # pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_30k_zmatrix/checkpoint-23000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_3/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: llama_collator_3 6 | finetune: 7 | model_name: llama_collator_3 8 | # context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 16 11 | # path: 12 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_3/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: llama_collator_3 6 | finetune: 7 | model_name: llama_collator_3 8 | # context_length: 1024 9 | # training_arguments: 10 | # per_device_train_batch_size: 64 11 | # path: 12 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_3/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: llama_collator_3 6 | 7 | finetune: 8 | model_name: llama_collator_3 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 64 12 | # path: 13 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_3/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: llama_collator_3 6 | 7 | finetune: 8 | model_name: llama_collator_3 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 64 12 | # path: 13 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_3/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: llama_collator_3 6 | 7 | finetune: 8 | model_name: llama_collator_3 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 128 12 | # path: 13 | # pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_30k_zmatrix/checkpoint-23000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_4/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: llama_collator_4 6 | finetune: 7 | model_name: llama_collator_4 8 | # context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 16 11 | # path: 12 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_4/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: llama_collator_4 6 | finetune: 7 | model_name: llama_collator_4 8 | # context_length: 1024 9 | # training_arguments: 10 | # per_device_train_batch_size: 64 11 | # path: 12 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_4/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: llama_collator_4 6 | 7 | finetune: 8 | model_name: llama_collator_4 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 64 12 | # path: 13 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_4/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: llama_collator_4 6 | 7 | finetune: 8 | model_name: llama_collator_4 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 64 12 | # path: 13 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_4/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: llama_collator_4 6 | 7 | finetune: 8 | model_name: llama_collator_4 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 128 12 | # path: 13 | # pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_30k_zmatrix/checkpoint-23000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_5/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: llama_collator_5 6 | finetune: 7 | model_name: llama_collator_5 8 | # context_length: 1024 9 | # training_arguments: 10 | # per_device_train_batch_size: 16 11 | # path: 12 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_5/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: llama_collator_5 6 | finetune: 7 | model_name: llama_collator_5 8 | # context_length: 1024 9 | # training_arguments: 10 | # per_device_train_batch_size: 64 11 | # path: 12 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_5/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: llama_collator_5 6 | 7 | finetune: 8 | model_name: llama_collator_5 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 64 12 | # path: 13 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_5/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: llama_collator_5 6 | 7 | finetune: 8 | model_name: llama_collator_5 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 64 12 | # path: 13 | # pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/llama_sft_5/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: llama_collator_5 6 | 7 | finetune: 8 | model_name: llama_collator_5 9 | # context_length: 512 10 | # training_arguments: 11 | # per_device_train_batch_size: 128 12 | # path: 13 | # pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_30k_zmatrix/checkpoint-23000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential/cryst_0.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 0 5 | logging: 6 | wandb_project: potential_lj 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: potential_0 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential/cryst_0_2.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 0.2 5 | logging: 6 | wandb_project: potential_lj 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: potential_0_2 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential/cryst_0_4.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 0.4 5 | logging: 6 | wandb_project: potential_lj 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: potential_0_4 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential/cryst_0_5.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 0.5 5 | logging: 6 | wandb_project: potential_lj 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: potential_0_5 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential/cryst_0_6.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 0.6 5 | logging: 6 | wandb_project: potential_lj 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: potential_0_6 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential/cryst_0_8.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 0.8 5 | logging: 6 | wandb_project: potential_lj 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: potential_0_8 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential/cryst_1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 1 5 | logging: 6 | wandb_project: potential_lj 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: potential_1 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_2/cryst_0.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 0 5 | logging: 6 | wandb_project: potential_lj_standard 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: standard_potential_0 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_2/cryst_0_2.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 0.2 5 | logging: 6 | wandb_project: potential_lj_standard 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: standard_potential_0_2 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_2/cryst_0_4.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 0.4 5 | logging: 6 | wandb_project: potential_lj_standard 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: standard_potential_0_4 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_2/cryst_0_5.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 0.5 5 | logging: 6 | wandb_project: potential_lj_standard 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: standard_potential_0_5 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_2/cryst_0_6.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 0.6 5 | logging: 6 | wandb_project: potential_lj_standard 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: standard_potential_0_6 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_2/cryst_0_8.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 0.8 5 | logging: 6 | wandb_project: potential_lj_standard 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: standard_potential_0_8 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_2/cryst_1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | alpha: 1 5 | logging: 6 | wandb_project: potential_lj_standard 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: standard_potential_1 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/crystal_llm_rep_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_3/cryst_0.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | alpha: 0 5 | logging: 6 | wandb_project: potential_lj_composition 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: composition_potential_0 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/composition_pt_2m/checkpoint-12000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_3/cryst_0_2.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | alpha: 0.2 5 | logging: 6 | wandb_project: potential_lj_composition 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: composition_potential_0_2 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/composition_pt_2m/checkpoint-12000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_3/cryst_0_4.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | alpha: 0.4 5 | logging: 6 | wandb_project: potential_lj_composition 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: composition_potential_0_4 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/composition_pt_2m/checkpoint-12000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_3/cryst_0_5.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | alpha: 0.5 5 | logging: 6 | wandb_project: potential_lj_composition 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: composition_potential_0_5 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/composition_pt_2m/checkpoint-12000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_3/cryst_0_6.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | alpha: 0.6 5 | logging: 6 | wandb_project: potential_lj_composition 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: composition_potential_0_6 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/composition_pt_2m/checkpoint-12000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_3/cryst_0_8.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | alpha: 0.8 5 | logging: 6 | wandb_project: potential_lj_composition 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: composition_potential_0_8 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/composition_pt_2m/checkpoint-12000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_3/cryst_1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | alpha: 1 5 | logging: 6 | wandb_project: potential_lj_composition 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: composition_potential_1 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/composition_pt_2m/checkpoint-12000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_4/cryst_0.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | alpha: 0 5 | logging: 6 | wandb_project: potential_lj_slice 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: slice_potential_0 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/slice_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_4/cryst_0_2.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | alpha: 0.2 5 | logging: 6 | wandb_project: potential_lj_slice 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: slice_potential_0_2 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/slice_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_4/cryst_0_4.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | alpha: 0.4 5 | logging: 6 | wandb_project: potential_lj_slice 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: slice_potential_0_4 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/slice_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_4/cryst_0_5.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | alpha: 0.5 5 | logging: 6 | wandb_project: potential_lj_slice 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: slice_potential_0_5 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/slice_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_4/cryst_0_6.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | alpha: 0.6 5 | logging: 6 | wandb_project: potential_lj_slice 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: slice_potential_0_6 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/slice_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_4/cryst_0_8.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | alpha: 0.8 5 | logging: 6 | wandb_project: potential_lj_slice 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: slice_potential_0_8 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/slice_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_4/cryst_1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | alpha: 1 5 | logging: 6 | wandb_project: potential_lj_slice 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: slice_potential_1 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/slice_pt_2m/checkpoint-393000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_5/cryst_0.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | alpha: 0 5 | logging: 6 | wandb_project: potential_lj_zmatrix 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: zmatrix_potential_0 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_300k_zmatrix/checkpoint-85000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_5/cryst_0_2.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | alpha: 0.2 5 | logging: 6 | wandb_project: potential_lj_zmatrix 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: zmatrix_potential_0_2 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_300k_zmatrix/checkpoint-85000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_5/cryst_0_4.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | alpha: 0.4 5 | logging: 6 | wandb_project: potential_lj_zmatrix 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: zmatrix_potential_0_4 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_300k_zmatrix/checkpoint-85000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_5/cryst_0_5.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | alpha: 0.5 5 | logging: 6 | wandb_project: potential_lj_zmatrix 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: zmatrix_potential_0_5 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_300k_zmatrix/checkpoint-85000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_5/cryst_0_6.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | alpha: 0.6 5 | logging: 6 | wandb_project: potential_lj_zmatrix 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: zmatrix_potential_0_6 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_300k_zmatrix/checkpoint-85000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_5/cryst_0_8.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | alpha: 0.8 5 | logging: 6 | wandb_project: potential_lj_zmatrix 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: zmatrix_potential_0_8 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_300k_zmatrix/checkpoint-85000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_5/cryst_1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | alpha: 1 5 | logging: 6 | wandb_project: potential_lj_zmatrix 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: zmatrix_potential_1 11 | context_length: 512 12 | training_arguments: 13 | per_device_train_batch_size: 128 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_300k_zmatrix/checkpoint-85000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_6/cryst_0.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | alpha: 0 5 | logging: 6 | wandb_project: potential_lj_cif_p1 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: cif_p1_potential_0 11 | context_length: 1024 12 | training_arguments: 13 | per_device_train_batch_size: 64 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/cif_p1_pt_2m/checkpoint-524000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_6/cryst_0_2.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | alpha: 0.2 5 | logging: 6 | wandb_project: potential_lj_cif_p1 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: cif_p1_potential_0_2 11 | context_length: 1024 12 | training_arguments: 13 | per_device_train_batch_size: 64 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/cif_p1_pt_2m/checkpoint-524000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_6/cryst_0_4.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | alpha: 0.4 5 | logging: 6 | wandb_project: potential_lj_cif_p1 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: cif_p1_potential_0_4 11 | context_length: 1024 12 | training_arguments: 13 | per_device_train_batch_size: 64 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/cif_p1_pt_2m/checkpoint-524000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_6/cryst_0_5.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | alpha: 0.5 5 | logging: 6 | wandb_project: potential_lj_cif_p1 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: cif_p1_potential_0_5 11 | context_length: 1024 12 | training_arguments: 13 | per_device_train_batch_size: 64 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/cif_p1_pt_2m/checkpoint-524000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_6/cryst_0_6.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | alpha: 0.6 5 | logging: 6 | wandb_project: potential_lj_cif_p1 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: cif_p1_potential_0_6 11 | context_length: 1024 12 | training_arguments: 13 | per_device_train_batch_size: 64 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/cif_p1_pt_2m/checkpoint-524000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_6/cryst_0_8.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | alpha: 0.8 5 | logging: 6 | wandb_project: potential_lj_cif_p1 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: cif_p1_potential_0_8 11 | context_length: 1024 12 | training_arguments: 13 | per_device_train_batch_size: 64 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/cif_p1_pt_2m/checkpoint-524000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_6/cryst_1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | alpha: 1 5 | logging: 6 | wandb_project: potential_lj_cif_p1 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: cif_p1_potential_1 11 | context_length: 1024 12 | training_arguments: 13 | per_device_train_batch_size: 64 14 | path: 15 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_normal/cif_p1_pt_2m/checkpoint-524000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_7/cryst_0.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | alpha: 0 5 | logging: 6 | wandb_project: potential_lj_atoms_params 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: atoms_params_potential_0 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/atoms_params_pt_2m_atoms_params/checkpoint-24000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_7/cryst_0_2.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | alpha: 0.2 5 | logging: 6 | wandb_project: potential_lj_atoms_params 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: atoms_params_potential_0_2 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/atoms_params_pt_2m_atoms_params/checkpoint-24000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_7/cryst_0_4.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | alpha: 0.4 5 | logging: 6 | wandb_project: potential_lj_atoms_params 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: atoms_params_potential_0_4 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/atoms_params_pt_2m_atoms_params/checkpoint-24000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_7/cryst_0_5.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | alpha: 0.5 5 | logging: 6 | wandb_project: potential_lj_atoms_params 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: atoms_params_potential_0_5 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/atoms_params_pt_2m_atoms_params/checkpoint-24000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_7/cryst_0_6.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | alpha: 0.6 5 | logging: 6 | wandb_project: potential_lj_atoms_params 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: atoms_params_potential_0_6 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/atoms_params_pt_2m_atoms_params/checkpoint-24000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_7/cryst_0_8.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | alpha: 0.8 5 | logging: 6 | wandb_project: potential_lj_atoms_params 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: atoms_params_potential_0_8 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/atoms_params_pt_2m_atoms_params/checkpoint-24000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/potential_7/cryst_1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | alpha: 1 5 | logging: 6 | wandb_project: potential_lj_atoms_params 7 | special_num_token: False 8 | 9 | finetune: 10 | model_name: atoms_params_potential_1 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/atoms_params_pt_2m_atoms_params/checkpoint-24000 16 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain-rt/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_30k_rt 5 | 6 | representation: atoms 7 | pretrain: 8 | name: pt_30k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain-rt/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_30k_rt 5 | 6 | representation: atoms_params 7 | pretrain: 8 | name: pt_30k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain-rt/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_30k_rt 5 | 6 | representation: cif_p1 7 | pretrain: 8 | name: pt_30k_rt 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain-rt/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_30k_rt 5 | 6 | representation: cif_symmetrized 7 | pretrain: 8 | name: pt_30k_rt 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain-rt/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_30k_rt 5 | 6 | representation: composition 7 | pretrain: 8 | name: pt_30k_rt 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain-rt/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_30k_rt 5 | 6 | representation: crystal_llm_rep 7 | pretrain: 8 | name: pt_30k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain-rt/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_30k_rt 5 | 6 | representation: slice 7 | pretrain: 8 | name: pt_30k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain-rt/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_30k_rt 5 | 6 | representation: zmatrix 7 | pretrain: 8 | name: pt_30k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k/zmatrix 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain-test/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pretrain-test 5 | 6 | representation: cif_p1 7 | pretrain: 8 | name: pt_30k 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k #--> Change this to folder continaing 30k dataset train.json and test.json 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain-test/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pretrain-test 5 | 6 | representation: cif_symmetrized 7 | pretrain: 8 | name: pt_30k 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k #--> Change this to folder continaing 30k dataset train.json and test.json 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain-test/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pretrain-test 5 | 6 | representation: composition 7 | pretrain: 8 | name: pt_30k 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k #--> Change this to folder continaing 30k dataset train.json and test.json 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain-test/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pretrain-test 5 | 6 | representation: crystal_llm_rep 7 | pretrain: 8 | name: pt_30k 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k #--> Change this to folder continaing 30k dataset train.json and test.json 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain-test/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pretrain-test 5 | 6 | representation: slice 7 | pretrain: 8 | name: pt_30k 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k #--> Change this to folder continaing 30k dataset train.json and test.json 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain100k_spl_token/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: 100k_ft 5 | 6 | representation: atoms_params 7 | pretrain: 8 | name: 100k_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain100k_spl_token/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: 100k_ft 5 | 6 | representation: cif_p1 7 | pretrain: 8 | name: 100k_ft 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain100k_spl_token/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: 100k_ft 5 | 6 | representation: cif_symmetrized 7 | pretrain: 8 | name: 100k_ft 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain100k_spl_token/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: 100k_ft 5 | 6 | representation: composition 7 | pretrain: 8 | name: 100k_ft 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain100k_spl_token/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: 100k_ft 5 | 6 | representation: crystal_llm_rep 7 | pretrain: 8 | name: 100k_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain100k_spl_token/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: 100k_ft 5 | 6 | representation: slice 7 | pretrain: 8 | name: 100k_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain100k_spl_token/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: 100k_ft 5 | 6 | representation: zmatrix 7 | pretrain: 8 | name: 100k_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain30k_spl_token/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | logging: 5 | wandb_project: 30k_ft 6 | 7 | finetune: 8 | model_name: 30k_ft 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_params_pt_30k_atoms/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain30k_spl_token/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: 30k_ft 5 | 6 | representation: cif_p1 7 | pretrain: 8 | name: 30k_ft 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain30k_spl_token/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: 30k_ft 5 | 6 | representation: cif_symmetrized 7 | pretrain: 8 | name: 30k_ft 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain30k_spl_token/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: 30k_ft 5 | 6 | representation: composition 7 | pretrain: 8 | name: 30k_ft 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain30k_spl_token/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: 30k_ft 5 | 6 | representation: crystal_llm_rep 7 | pretrain: 8 | name: 30k_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain30k_spl_token/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: 30k_ft 5 | 6 | representation: slice 7 | pretrain: 8 | name: 30k_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain30k_spl_token/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: 30k_ft 6 | 7 | finetune: 8 | model_name: 30k_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | pretrained_checkpoint: /work/so87pot/material_db/mattext_dataset/30k/zmatrix 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_2m/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_2m 5 | 6 | representation: atoms 7 | pretrain: 8 | name: pt_2m 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/2m/newreps 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_2m/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_2m 5 | 6 | representation: atoms_params 7 | pretrain: 8 | name: pt_2m 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/2m/newreps 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_2m/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: zmatrix 5 | 6 | representation: zmatrix 7 | pretrain: 8 | name: zmatrix 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/2m/zmatrix 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_100/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_rt 5 | 6 | representation: atoms 7 | pretrain: 8 | name: pt_100k_rt 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_100/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_rt 5 | 6 | representation: atoms_params 7 | pretrain: 8 | name: pt_100k_rt 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_100/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_rt 5 | 6 | representation: cif_p1 7 | pretrain: 8 | name: pt_100k_rt 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_100/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_rt 5 | 6 | representation: cif_symmetrized 7 | pretrain: 8 | name: pt_100k_rt 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_100/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_rt 5 | 6 | representation: composition 7 | pretrain: 8 | name: pt_100k_rt 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_100/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_rt 5 | 6 | representation: crystal_llm_rep 7 | pretrain: 8 | name: pt_100k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_100/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_rt 5 | 6 | representation: slice 7 | pretrain: 8 | name: pt_100k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_100/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_rt 5 | 6 | representation: zmatrix 7 | pretrain: 8 | name: pt_100k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k/zmatrix 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_30/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_30k_rt 5 | 6 | representation: atoms 7 | pretrain: 8 | name: pt_30k_rt 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_30/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_30k_rt 5 | 6 | representation: atoms_params 7 | pretrain: 8 | name: pt_30k_rt 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_30/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_rt 5 | 6 | representation: cif_p1 7 | pretrain: 8 | name: pt_100k_rt 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_30/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_rt 5 | 6 | representation: cif_symmetrized 7 | pretrain: 8 | name: pt_100k_rt 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_30/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_rt 5 | 6 | representation: composition 7 | pretrain: 8 | name: pt_100k_rt 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_30/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_rt 5 | 6 | representation: crystal_llm_rep 7 | pretrain: 8 | name: pt_100k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_30/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_100k_rt 5 | 6 | representation: slice 7 | pretrain: 8 | name: pt_100k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/100k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_30/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_30k_rt 5 | 6 | representation: zmatrix 7 | pretrain: 8 | name: pt_30k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/30k/zmatrix 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_300/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_300k_rt 5 | 6 | representation: atoms 7 | pretrain: 8 | name: pt_300k_rt 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/300k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_300/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_300k_rt 5 | 6 | representation: atoms_params 7 | pretrain: 8 | name: pt_300k_rt 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/300k/atoms 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_300/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_300k_rt 5 | 6 | representation: cif_p1 7 | pretrain: 8 | name: pt_300k_rt 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/300k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_300/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_300k_rt 5 | 6 | representation: cif_symmetrized 7 | pretrain: 8 | name: pt_300k_rt 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/300k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_300/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_300k_rt 5 | 6 | representation: composition 7 | pretrain: 8 | name: pt_300k_rt 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/300k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_300/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_300k_rt 5 | 6 | representation: crystal_llm_rep 7 | pretrain: 8 | name: pt_300k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/300k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_300/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_300k_rt 5 | 6 | representation: slice 7 | pretrain: 8 | name: pt_300k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/300k 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/pretrain_rt_300/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | logging: 4 | wandb_project: pt_300k_rt 5 | 6 | representation: zmatrix 7 | pretrain: 8 | name: pt_300k_rt 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | data_root_path: /work/so87pot/material_db/mattext_dataset/300k/zmatrix 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/qmof_ft/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | logging: 5 | wandb_project: qmof_bg 6 | 7 | finetune: 8 | model_name: qmof_bg 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_params_pt_30k_atoms/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/qmof_ft/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: qmof_bg 6 | finetune: 7 | model_name: qmof_bg 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/qmof_ft/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: qmof_bg 6 | 7 | finetune: 8 | model_name: qmof_bg 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_symmetrized_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/qmof_ft/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: qmof_bg 6 | 7 | finetune: 8 | model_name: qmof_bg 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 512 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/composition_pt_30k_wes/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/qmof_ft/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: qmof_bg 6 | 7 | finetune: 8 | model_name: qmof_bg 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/qmof_ft/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: qmof_bg 6 | 7 | finetune: 8 | model_name: qmof_bg 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/slice_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/qmof_ft/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: qmof_bg 6 | 7 | finetune: 8 | model_name: qmof_bg 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 128 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/zmatrix_pt_30k_zmatrix/checkpoint-23000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/santiago/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: santiago_30k 6 | finetune: 7 | model_name: ft_30k_santiago 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: /home/so87pot/n0w0f/santiago_ckpt/cif_p1_pt_30k/checkpoint-23000 13 | 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/santiago_100k/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: santiago_100 6 | finetune: 7 | model_name: ft_100k_santiago 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: /home/so87pot/n0w0f/santiago_ckpt/cif_p1_pt_100k/checkpoint-26000 -------------------------------------------------------------------------------- /conf/archived_experiments/santiago_2m/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: santiago_2m 6 | finetune: 7 | model_name: ft_2m_santiago 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: /home/so87pot/n0w0f/santiago_ckpt/cif_p1_pt_2m/checkpoint-524000 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/santiago_2m/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: santiago_2m 6 | 7 | finetune: 8 | model_name: ft_2m_santiago 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/cif_symmetrized_pt_30k_rt/checkpoint-45000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/santiago_2m/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: santiago_2m 6 | 7 | finetune: 8 | model_name: ft_2m_santiago 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/composition_pt_30k_rt/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/santiago_2m/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: santiago_2m 6 | 7 | finetune: 8 | model_name: ft_2m_santiago 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/crystal_llm_rep_pt_30k_rt/checkpoint-11000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/santiago_2m/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: santiago_2m 6 | 7 | finetune: 8 | model_name: ft_2m_santiago 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/slice_pt_30k_rt/checkpoint-23000 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/smiles/slice_100.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: local_env 4 | special_num_token: False 5 | logging: 6 | wandb_project: 100k_ft_smiles 7 | 8 | finetune: 9 | model_name: 100k_ft_smiles 10 | context_length: 512 11 | training_arguments: 12 | per_device_train_batch_size: 64 13 | path: 14 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/smiles/local_env_smiles_100k/checkpoint-38000 15 | -------------------------------------------------------------------------------- /conf/archived_experiments/smiles/slice_2m.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: local_env 4 | special_num_token: False 5 | logging: 6 | wandb_project: 2m_ft_smiles 7 | 8 | finetune: 9 | model_name: 2m_ft_smiles 10 | context_length: 512 11 | training_arguments: 12 | per_device_train_batch_size: 64 13 | path: 14 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/santiago_ckpt_rt/checkpoint-95000 15 | -------------------------------------------------------------------------------- /conf/archived_experiments/smiles/slice_30.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | special_num_token: False 4 | representation: local_env 5 | logging: 6 | wandb_project: 30k_ft_smiles 7 | 8 | finetune: 9 | model_name: 30k_ft_smiles 10 | context_length: 512 11 | training_arguments: 12 | per_device_train_batch_size: 64 13 | path: 14 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/smiles/local_env_smiles_30k/checkpoint-9000 15 | -------------------------------------------------------------------------------- /conf/archived_experiments/smiles/slice_300.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: local_env 4 | special_num_token: False 5 | logging: 6 | wandb_project: 300k_ft_smiles 7 | 8 | finetune: 9 | model_name: 300k_ft_smiles 10 | context_length: 512 11 | training_arguments: 12 | per_device_train_batch_size: 64 13 | path: 14 | pretrained_checkpoint: /home/so87pot/n0w0f/mattext_ckpt/smiles/local_env_smiles_300k/checkpoint-91000 15 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_perturb/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: perturb_1 6 | finetune: 7 | model_name: finetune_30k_wes_3 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_perturb/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: perturb_1 6 | 7 | finetune: 8 | model_name: finetune_30k_wes_3 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_symmetrized_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_perturb/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: perturb_1 6 | 7 | finetune: 8 | model_name: finetune_30k_wes_3 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_perturb_100/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: perturb_1 6 | finetune: 7 | model_name: ft_100k_mb_small 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: ft_100k_mb_small 13 | 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_perturb_100/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: perturb_1 6 | 7 | 8 | finetune: 9 | model_name: ft_100k_mb_small 10 | context_length: 1024 11 | training_arguments: 12 | per_device_train_batch_size: 64 13 | path: 14 | pretrained_checkpoint: ft_100k_mb_small -------------------------------------------------------------------------------- /conf/archived_experiments/testing_perturb_100/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: perturb_1 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: ft_100k_mb_small 14 | 15 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_perturb_300/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: perturb_1 6 | finetune: 7 | model_name: ft_300k_mb_small 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: ft_300k_mb_small 13 | 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_perturb_300/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: perturb_1 6 | 7 | 8 | finetune: 9 | model_name: ft_300k_mb_small 10 | context_length: 1024 11 | training_arguments: 12 | per_device_train_batch_size: 64 13 | path: 14 | pretrained_checkpoint: ft_300k_mb_small -------------------------------------------------------------------------------- /conf/archived_experiments/testing_perturb_300/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: perturb_1 6 | 7 | finetune: 8 | model_name: ft_300k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: ft_300k_mb_small 14 | 15 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_translate/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: translate_1 6 | finetune: 7 | model_name: finetune_30k_wes_3 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_p1_pt_30k_wes/checkpoint-46000" 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_translate/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: translate_1 6 | 7 | finetune: 8 | model_name: finetune_30k_wes_3 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/cif_symmetrized_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_translate/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: translate_1 6 | 7 | finetune: 8 | model_name: finetune_30k_wes_3 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: "/work/so87pot/mattext/megaloop/checkpoints/checkpoints/crystal_llm_rep_pt_30k_wes/checkpoint-46000" 14 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_translate_100/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: translate_1 6 | finetune: 7 | model_name: ft_100k_mb_small 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: ft_100k_mb_small 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_translate_100/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: translate_1 6 | 7 | 8 | finetune: 9 | model_name: ft_100k_mb_small 10 | context_length: 1024 11 | training_arguments: 12 | per_device_train_batch_size: 64 13 | path: 14 | pretrained_checkpoint: ft_100k_mb_small -------------------------------------------------------------------------------- /conf/archived_experiments/testing_translate_100/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: translate_1 6 | 7 | finetune: 8 | model_name: ft_100k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: ft_100k_mb_small 14 | 15 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_translate_300/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: translate_1 6 | finetune: 7 | model_name: ft_300k_mb_small 8 | context_length: 1024 9 | training_arguments: 10 | per_device_train_batch_size: 64 11 | path: 12 | pretrained_checkpoint: ft_300k_mb_small 13 | -------------------------------------------------------------------------------- /conf/archived_experiments/testing_translate_300/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: translate_1 6 | 7 | 8 | finetune: 9 | model_name: ft_300k_mb_small 10 | context_length: 1024 11 | training_arguments: 12 | per_device_train_batch_size: 64 13 | path: 14 | pretrained_checkpoint: ft_300k_mb_small -------------------------------------------------------------------------------- /conf/archived_experiments/testing_translate_300/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: translate_1 6 | 7 | finetune: 8 | model_name: ft_300k_mb_small 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: ft_300k_mb_small 14 | 15 | -------------------------------------------------------------------------------- /conf/bandgap.yaml: -------------------------------------------------------------------------------- 1 | 2 | 3 | hydra: 4 | job: 5 | name: bandgap 6 | run: 7 | dir: ${hydra:runtime.cwd}/outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} 8 | sweep: 9 | dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} 10 | subdir: ${hydra.job.override_dirname} 11 | 12 | # launcher: 13 | # _target_: hydra_plugins.hydra_submitit_launcher.submitit_launcher.SlurmLauncher 14 | # submitit_folder: ${hydra.sweep.dir}/.submitit/%j 15 | # timeout_min: 3600 16 | # mem_gb: 160 17 | # nodes: 1 18 | # #gpus_per_task: 1 19 | # gres: gpu:1 20 | # #gpus_per_node: 2 21 | # name: ${hydra.job.name} 22 | # partition: 'gpu' 23 | # additional_parameters: 24 | # nodelist: 'gpu[008,013-017]' 25 | # tasks_per_node: 1 26 | 27 | defaults: 28 | - model: none 29 | # - override hydra/launcher: submitit_slurm 30 | 31 | runs: 32 | - name: benchmark_run 33 | tasks: [benchmark] -------------------------------------------------------------------------------- /conf/benchmark.yaml: -------------------------------------------------------------------------------- 1 | 2 | 3 | hydra: 4 | job: 5 | name: benchmark 6 | run: 7 | dir: ${hydra:runtime.cwd}/outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} 8 | sweep: 9 | dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} 10 | subdir: ${hydra.job.override_dirname} 11 | 12 | # launcher: 13 | # _target_: hydra_plugins.hydra_submitit_launcher.submitit_launcher.SlurmLauncher 14 | # submitit_folder: ${hydra.sweep.dir}/.submitit/%j 15 | # timeout_min: 3600 16 | # mem_gb: 160 17 | # nodes: 1 18 | # #gpus_per_task: 1 19 | # gres: gpu:1 20 | # #gpus_per_node: 2 21 | # name: ${hydra.job.name} 22 | # partition: 'gpu' 23 | # additional_parameters: 24 | # nodelist: 'gpu[008,013-017]' 25 | # tasks_per_node: 1 26 | 27 | defaults: 28 | - model: none 29 | # - override hydra/launcher: submitit_slurm 30 | 31 | runs: 32 | - name: benchmark_run 33 | tasks: [benchmark] -------------------------------------------------------------------------------- /conf/bg/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atom_sequences 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-atom-seq-2m 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 32 14 | training_arguments: 15 | per_device_train_batch_size: 1024 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-atom-seq-2m 18 | 19 | -------------------------------------------------------------------------------- /conf/bg/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atom_sequences_plusplus 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-atom-seq-plusplus-2m 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 32 14 | training_arguments: 15 | per_device_train_batch_size: 1024 16 | 17 | -------------------------------------------------------------------------------- /conf/bg/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-cifp1-2m 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 1024 14 | training_arguments: 15 | per_device_train_batch_size: 128 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-cifp1-2m -------------------------------------------------------------------------------- /conf/bg/cifpsym.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-cifsymmetrized-2m 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 1024 14 | training_arguments: 15 | per_device_train_batch_size: 64 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-cifsymmetrized-2m -------------------------------------------------------------------------------- /conf/bg/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-composition-2m 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 32 14 | training_arguments: 15 | per_device_train_batch_size: 1024 16 | 17 | -------------------------------------------------------------------------------- /conf/bg/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_text_llm 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: /home/so87pot/n0w0f/structllm_ckpt/alpaca_ckpt/checkpoint-393000 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 512 14 | training_arguments: 15 | per_device_train_batch_size: 256 16 | -------------------------------------------------------------------------------- /conf/bg/local_env.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: local_env 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: /home/so87pot/n0w0f/structllm_ckpt/santiago_ckpt_rt/checkpoint-95000 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 512 14 | training_arguments: 15 | per_device_train_batch_size: 256 16 | path: 17 | pretrained_checkpoint: /home/so87pot/n0w0f/structllm_ckpt/santiago_ckpt_rt/checkpoint-95000 -------------------------------------------------------------------------------- /conf/bg/slices.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slices 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-slices-2m 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 512 14 | training_arguments: 15 | per_device_train_batch_size: 256 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-slices-2m -------------------------------------------------------------------------------- /conf/bg/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-zmatrix-2m 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 512 14 | training_arguments: 15 | per_device_train_batch_size: 256 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-zmatrix-2m -------------------------------------------------------------------------------- /conf/bg2m/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_params_pt_30k_atoms/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/bg2m/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atoms_params 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_params_pt_30k_atoms/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/bg2m/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/cif_p1_pt_30k_rt_2/checkpoint-46000 14 | -------------------------------------------------------------------------------- /conf/bg2m/cifsymmetrized.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 1024 10 | training_arguments: 11 | per_device_train_batch_size: 32 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/cif_symmetrized_pt_30k_rt/checkpoint-45000 14 | -------------------------------------------------------------------------------- /conf/bg2m/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 32 10 | training_arguments: 11 | per_device_train_batch_size: 1024 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/composition_pt_30k_rt/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/bg2m/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_llm_rep 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/crystal_llm_rep_pt_30k_rt/checkpoint-11000 14 | -------------------------------------------------------------------------------- /conf/bg2m/local_env.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_params_pt_30k_atoms/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/bg2m/slice.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slice 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop2/checkpoints/checkpoints/slice_pt_30k_rt/checkpoint-23000 14 | -------------------------------------------------------------------------------- /conf/bg2m/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | logging: 5 | wandb_project: 2m_intel_ft 6 | 7 | finetune: 8 | model_name: 2m_intel_ft 9 | context_length: 512 10 | training_arguments: 11 | per_device_train_batch_size: 64 12 | path: 13 | pretrained_checkpoint: /work/so87pot/mattext/megaloop/checkpoints/checkpoints/atoms_params_pt_30k_atoms/checkpoint-1000 14 | -------------------------------------------------------------------------------- /conf/classification.yaml: -------------------------------------------------------------------------------- 1 | 2 | 3 | hydra: 4 | job: 5 | name: is_metal 6 | run: 7 | dir: ${hydra:runtime.cwd}/outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} 8 | sweep: 9 | dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} 10 | subdir: ${hydra.job.override_dirname} 11 | 12 | # launcher: 13 | # _target_: hydra_plugins.hydra_submitit_launcher.submitit_launcher.SlurmLauncher 14 | # submitit_folder: ${hydra.sweep.dir}/.submitit/%j 15 | # timeout_min: 3600 16 | # mem_gb: 160 17 | # nodes: 1 18 | # #gpus_per_task: 1 19 | # gres: gpu:1 20 | # #gpus_per_node: 2 21 | # name: ${hydra.job.name} 22 | # partition: 'gpu' 23 | # additional_parameters: 24 | # nodelist: 'gpu[008,013-017]' 25 | # tasks_per_node: 1 26 | 27 | defaults: 28 | - model: none 29 | # - override hydra/launcher: submitit_slurm 30 | 31 | runs: 32 | - name: classification_run 33 | tasks: [classification] -------------------------------------------------------------------------------- /conf/form/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atom_sequences 4 | dataset: "form_energy" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-atom-seq-2m 8 | logging: 9 | wandb_project: revision-form 10 | 11 | finetune: 12 | model_name: revision-form 13 | context_length: 32 14 | training_arguments: 15 | per_device_train_batch_size: 1024 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-atom-seq-2m 18 | 19 | -------------------------------------------------------------------------------- /conf/form/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atom_sequences_plusplus 4 | dataset: "form_energy" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-atom-seq-plusplus-2m 8 | logging: 9 | wandb_project: revision-form 10 | 11 | finetune: 12 | model_name: revision-form 13 | context_length: 32 14 | training_arguments: 15 | per_device_train_batch_size: 2048 16 | 17 | -------------------------------------------------------------------------------- /conf/form/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | dataset: "form_energy" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-cifp1-2m 8 | logging: 9 | wandb_project: revision-form 10 | 11 | finetune: 12 | model_name: revision-form 13 | context_length: 1024 14 | training_arguments: 15 | per_device_train_batch_size: 64 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-cifp1-2m -------------------------------------------------------------------------------- /conf/form/cifpsym.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | dataset: "form_energy" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-cifsymmetrized-2m 8 | logging: 9 | wandb_project: revision-form 10 | 11 | finetune: 12 | model_name: revision-form 13 | context_length: 1024 14 | training_arguments: 15 | per_device_train_batch_size: 64 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-cifsymmetrized-2m -------------------------------------------------------------------------------- /conf/form/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | dataset: "form_energy" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-composition-2m 8 | logging: 9 | wandb_project: revision-form 10 | 11 | finetune: 12 | model_name: revision-form 13 | context_length: 32 14 | training_arguments: 15 | per_device_train_batch_size: 2048 16 | 17 | -------------------------------------------------------------------------------- /conf/form/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_text_llm 4 | dataset: "form_energy" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: /home/so87pot/n0w0f/structllm_ckpt/alpaca_ckpt/cllm/checkpoint-393000 8 | logging: 9 | wandb_project: revision-form 10 | 11 | finetune: 12 | model_name: revision-form 13 | context_length: 512 14 | training_arguments: 15 | per_device_train_batch_size: 256 16 | -------------------------------------------------------------------------------- /conf/form/local_env.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: local_env 4 | dataset: "form_energy" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: /home/so87pot/n0w0f/structllm_ckpt/alpaca_ckpt/local_env/checkpoint-381000 8 | logging: 9 | wandb_project: revision-form 10 | 11 | finetune: 12 | model_name: revision-form 13 | context_length: 512 14 | training_arguments: 15 | per_device_train_batch_size: 256 16 | path: 17 | pretrained_checkpoint: /home/so87pot/n0w0f/structllm_ckpt/alpaca_ckpt/local_env/checkpoint-381000 -------------------------------------------------------------------------------- /conf/form/slices.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slices 4 | dataset: "form_energy" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-slices-2m 8 | logging: 9 | wandb_project: revision-form 10 | 11 | finetune: 12 | model_name: revision-form 13 | context_length: 512 14 | training_arguments: 15 | per_device_train_batch_size: 128 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-slices-2m -------------------------------------------------------------------------------- /conf/form/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | dataset: "form_energy" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-zmatrix-2m 8 | logging: 9 | wandb_project: revision-form 10 | 11 | finetune: 12 | model_name: revision-form 13 | context_length: 512 14 | training_arguments: 15 | per_device_train_batch_size: 64 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-zmatrix-2m -------------------------------------------------------------------------------- /conf/form_energy.yaml: -------------------------------------------------------------------------------- 1 | 2 | 3 | hydra: 4 | job: 5 | name: formation_energy 6 | run: 7 | dir: ${hydra:runtime.cwd}/outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} 8 | sweep: 9 | dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} 10 | subdir: ${hydra.job.override_dirname} 11 | 12 | 13 | defaults: 14 | - model: none 15 | 16 | 17 | runs: 18 | - name: benchmark_run 19 | tasks: [benchmark] -------------------------------------------------------------------------------- /conf/group-test/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | dataset: "gvrh" 5 | special_num_token: False 6 | logging: 7 | wandb_project: test-benchmark 8 | 9 | finetune: 10 | model_name: test-benchmark 11 | context_length: 32 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /work/so87pot/structllm/megaloop2/checkpoints/checkpoints/composition_30k_ft/checkpoint-1000 16 | 17 | -------------------------------------------------------------------------------- /conf/group-test/slices.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slices 4 | dataset: "gvrh" 5 | special_num_token: False 6 | logging: 7 | wandb_project: test-benchmark 8 | 9 | finetune: 10 | model_name: test-benchmark 11 | context_length: 64 12 | training_arguments: 13 | per_device_train_batch_size: 1024 14 | path: 15 | pretrained_checkpoint: /work/so87pot/structllm/megaloop/checkpoints/checkpoints/slice_pretrain_30k_draco/checkpoint-23000 16 | -------------------------------------------------------------------------------- /conf/is_metal/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atom_sequences 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-atom-seq-2m 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 32 14 | training_arguments: 15 | per_device_train_batch_size: 1024 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-atom-seq-2m 18 | 19 | -------------------------------------------------------------------------------- /conf/is_metal/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atom_sequences_plusplus 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-atom-seq-plusplus-2m 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 32 14 | training_arguments: 15 | per_device_train_batch_size: 1024 16 | 17 | -------------------------------------------------------------------------------- /conf/is_metal/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-cifp1-2m 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 1024 14 | training_arguments: 15 | per_device_train_batch_size: 128 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-cifp1-2m -------------------------------------------------------------------------------- /conf/is_metal/cifpsym.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-cifsymmetrized-2m 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 1024 14 | training_arguments: 15 | per_device_train_batch_size: 64 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-cifsymmetrized-2m -------------------------------------------------------------------------------- /conf/is_metal/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | dataset: "is-metal" 5 | dataset_type: filtered 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-composition-2m 8 | logging: 9 | wandb_project: revision-bg-filtered 10 | 11 | finetune: 12 | model_name: revision-bg-filtered 13 | context_length: 32 14 | training_arguments: 15 | per_device_train_batch_size: 1024 16 | 17 | -------------------------------------------------------------------------------- /conf/is_metal/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_text_llm 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: /home/so87pot/n0w0f/structllm_ckpt/alpaca_ckpt/checkpoint-393000 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 512 14 | training_arguments: 15 | per_device_train_batch_size: 256 16 | -------------------------------------------------------------------------------- /conf/is_metal/local_env.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: local_env 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: /home/so87pot/n0w0f/structllm_ckpt/santiago_ckpt_rt/checkpoint-95000 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 512 14 | training_arguments: 15 | per_device_train_batch_size: 256 16 | path: 17 | pretrained_checkpoint: /home/so87pot/n0w0f/structllm_ckpt/santiago_ckpt_rt/checkpoint-95000 -------------------------------------------------------------------------------- /conf/is_metal/slices.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slices 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-slices-2m 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 512 14 | training_arguments: 15 | per_device_train_batch_size: 256 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-slices-2m -------------------------------------------------------------------------------- /conf/is_metal/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | dataset: "bandgap" 5 | dataset_type: matbench 6 | special_num_token: False 7 | checkpoint: n0w0f/MatText-zmatrix-2m 8 | logging: 9 | wandb_project: revision-bg 10 | 11 | finetune: 12 | model_name: revision-bg 13 | context_length: 512 14 | training_arguments: 15 | per_device_train_batch_size: 256 16 | path: 17 | pretrained_checkpoint: n0w0f/MatText-zmatrix-2m -------------------------------------------------------------------------------- /conf/llama_8b_bg/atoms.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atom_sequences 4 | dataset: "bandgap" 5 | dataset_type: filtered 6 | logging: 7 | wandb_project: llama-7B-ft 8 | 9 | -------------------------------------------------------------------------------- /conf/llama_8b_bg/atoms_params.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: atom_sequences_plusplus 4 | dataset: "bandgap" 5 | dataset_type: filtered 6 | logging: 7 | wandb_project: llama-7B-ft 8 | 9 | 10 | 11 | -------------------------------------------------------------------------------- /conf/llama_8b_bg/cifp1.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_p1 4 | dataset: "bandgap" 5 | dataset_type: filtered 6 | logging: 7 | wandb_project: llama-7B-ft 8 | 9 | -------------------------------------------------------------------------------- /conf/llama_8b_bg/cifpsym.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: cif_symmetrized 4 | dataset: "bandgap" 5 | dataset_type: filtered 6 | logging: 7 | wandb_project: llama-7B-ft 8 | -------------------------------------------------------------------------------- /conf/llama_8b_bg/composition.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: composition 4 | dataset: "bandgap" 5 | dataset_type: filtered 6 | logging: 7 | wandb_project: llama-7B-ft 8 | -------------------------------------------------------------------------------- /conf/llama_8b_bg/crystal_llm.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: crystal_text_llm 4 | dataset: "bandgap" 5 | dataset_type: filtered 6 | logging: 7 | wandb_project: llama-7B-ft -------------------------------------------------------------------------------- /conf/llama_8b_bg/local_env.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: local_env 4 | dataset: "bandgap" 5 | dataset_type: filtered 6 | logging: 7 | wandb_project: llama-7B-ft 8 | -------------------------------------------------------------------------------- /conf/llama_8b_bg/slices.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: slices 4 | dataset: "bandgap" 5 | dataset_type: filtered 6 | logging: 7 | wandb_project: llama-7B-ft 8 | -------------------------------------------------------------------------------- /conf/llama_8b_bg/zmatrix.yaml: -------------------------------------------------------------------------------- 1 | # @package _global_ 2 | model: 3 | representation: zmatrix 4 | dataset: "bandgap" 5 | dataset_type: filtered 6 | logging: 7 | wandb_project: llama-7B-ft 8 | -------------------------------------------------------------------------------- /conf/llm_sft.yaml: -------------------------------------------------------------------------------- 1 | hydra: 2 | job: 3 | name: llama_sft 4 | run: 5 | dir: ${hydra:runtime.cwd}/outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} 6 | sweep: 7 | dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} 8 | subdir: ${hydra.job.override_dirname} 9 | 10 | 11 | 12 | defaults: 13 | - model: none 14 | 15 | 16 | 17 | runs: 18 | - name: llama_sft_run 19 | tasks: [llama_sft] 20 | 21 | 22 | -------------------------------------------------------------------------------- /conf/pretrain.yaml: -------------------------------------------------------------------------------- 1 | hydra: 2 | job: 3 | name: llama_sft 4 | run: 5 | dir: ${hydra:runtime.cwd}/outputs/${now:%Y-%m-%d}/${now:%H-%M-%S} 6 | sweep: 7 | dir: multirun/${now:%Y-%m-%d}/${now:%H-%M-%S} 8 | subdir: ${hydra.job.override_dirname} 9 | 10 | 11 | 12 | defaults: 13 | - model: none 14 | 15 | 16 | 17 | runs: 18 | 19 | - name: pretrain_run 20 | tasks: [pretrain] 21 | 22 | 23 | -------------------------------------------------------------------------------- /docs/api.md: -------------------------------------------------------------------------------- 1 | # API Documentation 2 | 3 | 4 | ## Text representation 5 | 6 | ### Core class 7 | 8 | ::: mattext.representations.TextRep 9 | heading_level: 3 10 | 11 | 12 | ### Decoding 13 | 14 | ::: mattext.representations.decoder 15 | heading_level: 3 16 | 17 | ### Transformations 18 | 19 | ::: mattext.representations.transformations 20 | heading_level: 3 21 | 22 | ## Tokenizer 23 | 24 | ::: mattext.tokenizer 25 | heading_level: 3 26 | 27 | 28 | ## Models 29 | 30 | ::: mattext.models.benchmark 31 | heading_level: 3 32 | 33 | ::: mattext.models.finetune 34 | heading_level: 3 35 | 36 | ::: mattext.models.llama_sft 37 | heading_level: 3 38 | 39 | ::: mattext.models.llama 40 | heading_level: 3 41 | 42 | ::: mattext.models.potential 43 | heading_level: 3 44 | 45 | ::: mattext.models.predict 46 | heading_level: 3 47 | 48 | ::: mattext.models.pretrain 49 | heading_level: 3 50 | 51 | -------------------------------------------------------------------------------- /docs/getting_started.md: -------------------------------------------------------------------------------- 1 | # Installation 2 | 3 | 4 | The most recent code and data can be installed directly from GitHub with: 5 | 6 | 7 | ```shell 8 | $ pip install git+https://github.com/lamalab-org/mattext.git 9 | ``` 10 | 11 | To install in development mode, use the following: 12 | 13 | ```shell 14 | $ git clone git+https://github.com/lamalab-org/mattext.git 15 | $ cd mattext 16 | $ pip install -e . 17 | ``` 18 | 19 | 20 | If you want to use the Local Env representation, you will also need to install OpenBabel, e.g. using 21 | 22 | ```bash 23 | conda install openbabel -c conda-forge 24 | ``` 25 | -------------------------------------------------------------------------------- /docs/index.md: -------------------------------------------------------------------------------- 1 | # MatText documentation 2 | 3 |
4 |
5 | 6 |
7 |
8 | 9 | MatText is a framework for text-based materials modeling. It supports 10 | 11 | - conversion of crystal structures in to text representations 12 | - transformations of crystal structures for sensitivity analyses 13 | - decoding of text representations to crystal structures 14 | - tokenization of text-representation of crystal structures 15 | - pre-training, finetuning and testing of language models on text-representations of crystal structures 16 | - analysis of language models trained on text-representations of crystal structures -------------------------------------------------------------------------------- /docs/static/logo.ai: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/lamalab-org/MatText/866a0e24e723835f8fe799e326d730112fbc148a/docs/static/logo.ai -------------------------------------------------------------------------------- /docs/static/logo.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/lamalab-org/MatText/866a0e24e723835f8fe799e326d730112fbc148a/docs/static/logo.png -------------------------------------------------------------------------------- /notebooks/example_data/InCuS2_symmetrized.cif: -------------------------------------------------------------------------------- 1 | # generated using pymatgen 2 | data_InCuS2 3 | _symmetry_space_group_name_H-M I-42d 4 | _cell_length_a 5.52040491 5 | _cell_length_b 5.52040491 6 | _cell_length_c 11.12616975 7 | _cell_angle_alpha 90.00000000 8 | _cell_angle_beta 90.00000000 9 | _cell_angle_gamma 90.00000000 10 | _symmetry_Int_Tables_number 122 11 | _chemical_formula_structural InCuS2 12 | _chemical_formula_sum 'In4 Cu4 S8' 13 | _cell_volume 339.06858092 14 | _cell_formula_units_Z 4 15 | loop_ 16 | _symmetry_equiv_pos_site_id 17 | _symmetry_equiv_pos_as_xyz 18 | 1 'x, y, z' 19 | 2 'y, -x, -z' 20 | 3 '-x, -y, z' 21 | 4 '-y, x, -z' 22 | 5 'x, -y+1/2, -z+1/4' 23 | 6 'y, x+1/2, z+1/4' 24 | 7 '-x, y+1/2, -z+1/4' 25 | 8 '-y, -x+1/2, z+1/4' 26 | 9 'x+1/2, y+1/2, z+1/2' 27 | 10 'y+1/2, -x+1/2, -z+1/2' 28 | 11 '-x+1/2, -y+1/2, z+1/2' 29 | 12 '-y+1/2, x+1/2, -z+1/2' 30 | 13 'x+1/2, -y, -z+3/4' 31 | 14 'y+1/2, x, z+3/4' 32 | 15 '-x+1/2, y, -z+3/4' 33 | 16 '-y+1/2, -x, z+3/4' 34 | loop_ 35 | _atom_type_symbol 36 | _atom_type_oxidation_number 37 | In3+ 3.0 38 | Cu+ 1.0 39 | S2- -2.0 40 | loop_ 41 | _atom_site_type_symbol 42 | _atom_site_label 43 | _atom_site_symmetry_multiplicity 44 | _atom_site_fract_x 45 | _atom_site_fract_y 46 | _atom_site_fract_z 47 | _atom_site_occupancy 48 | In3+ In0 4 0.00000000 0.00000000 0.50000000 1 49 | Cu+ Cu1 4 0.00000000 0.00000000 0.00000000 1 50 | S2- S2 8 0.22119731 0.25000000 0.12500000 1 51 | -------------------------------------------------------------------------------- /notebooks/example_data/N2_p1.cif: -------------------------------------------------------------------------------- 1 | # generated using pymatgen 2 | data_N2 3 | _symmetry_space_group_name_H-M 'P 1' 4 | _cell_length_a 5.60540879 5 | _cell_length_b 5.60540879 6 | _cell_length_c 5.60540879 7 | _cell_angle_alpha 90.00000000 8 | _cell_angle_beta 90.00000000 9 | _cell_angle_gamma 90.00000000 10 | _symmetry_Int_Tables_number 1 11 | _chemical_formula_structural N2 12 | _chemical_formula_sum N4 13 | _cell_volume 176.12535061 14 | _cell_formula_units_Z 2 15 | loop_ 16 | _symmetry_equiv_pos_site_id 17 | _symmetry_equiv_pos_as_xyz 18 | 1 'x, y, z' 19 | loop_ 20 | _atom_type_symbol 21 | _atom_type_oxidation_number 22 | N0+ 0.0 23 | loop_ 24 | _atom_site_type_symbol 25 | _atom_site_label 26 | _atom_site_symmetry_multiplicity 27 | _atom_site_fract_x 28 | _atom_site_fract_y 29 | _atom_site_fract_z 30 | _atom_site_occupancy 31 | N0+ N0 1 0.47679036 0.97679036 0.52320964 1 32 | N0+ N1 1 0.97679036 0.52320964 0.47679036 1 33 | N0+ N2 1 0.02320964 0.02320964 0.02320964 1 34 | N0+ N3 1 0.52320964 0.47679036 0.97679036 1 35 | -------------------------------------------------------------------------------- /notebooks/example_data/N2_symmetrized.cif: -------------------------------------------------------------------------------- 1 | # generated using pymatgen 2 | data_N2 3 | _symmetry_space_group_name_H-M P2_13 4 | _cell_length_a 5.60540879 5 | _cell_length_b 5.60540879 6 | _cell_length_c 5.60540879 7 | _cell_angle_alpha 90.00000000 8 | _cell_angle_beta 90.00000000 9 | _cell_angle_gamma 90.00000000 10 | _symmetry_Int_Tables_number 198 11 | _chemical_formula_structural N2 12 | _chemical_formula_sum N4 13 | _cell_volume 176.12535061 14 | _cell_formula_units_Z 2 15 | loop_ 16 | _symmetry_equiv_pos_site_id 17 | _symmetry_equiv_pos_as_xyz 18 | 1 'x, y, z' 19 | 2 '-x+1/2, -y, z+1/2' 20 | 3 'x+1/2, -y+1/2, -z' 21 | 4 '-x, y+1/2, -z+1/2' 22 | 5 'z, x, y' 23 | 6 'z+1/2, -x+1/2, -y' 24 | 7 '-z, x+1/2, -y+1/2' 25 | 8 '-z+1/2, -x, y+1/2' 26 | 9 'y, z, x' 27 | 10 '-y, z+1/2, -x+1/2' 28 | 11 '-y+1/2, -z, x+1/2' 29 | 12 'y+1/2, -z+1/2, -x' 30 | loop_ 31 | _atom_type_symbol 32 | _atom_type_oxidation_number 33 | N0+ 0.0 34 | loop_ 35 | _atom_site_type_symbol 36 | _atom_site_label 37 | _atom_site_symmetry_multiplicity 38 | _atom_site_fract_x 39 | _atom_site_fract_y 40 | _atom_site_fract_z 41 | _atom_site_occupancy 42 | N0+ N0 4 0.02320964 0.02320964 0.02320964 1 43 | -------------------------------------------------------------------------------- /notebooks/example_data/SrTiO3_p1.cif: -------------------------------------------------------------------------------- 1 | # generated using pymatgen 2 | data_SrTiO3 3 | _symmetry_space_group_name_H-M 'P 1' 4 | _cell_length_a 3.91270131 5 | _cell_length_b 3.91270131 6 | _cell_length_c 3.91270131 7 | _cell_angle_alpha 90.00000000 8 | _cell_angle_beta 90.00000000 9 | _cell_angle_gamma 90.00000000 10 | _symmetry_Int_Tables_number 1 11 | _chemical_formula_structural SrTiO3 12 | _chemical_formula_sum 'Sr1 Ti1 O3' 13 | _cell_volume 59.90045031 14 | _cell_formula_units_Z 1 15 | loop_ 16 | _symmetry_equiv_pos_site_id 17 | _symmetry_equiv_pos_as_xyz 18 | 1 'x, y, z' 19 | loop_ 20 | _atom_type_symbol 21 | _atom_type_oxidation_number 22 | Sr2+ 2.0 23 | Ti4+ 4.0 24 | O2- -2.0 25 | loop_ 26 | _atom_site_type_symbol 27 | _atom_site_label 28 | _atom_site_symmetry_multiplicity 29 | _atom_site_fract_x 30 | _atom_site_fract_y 31 | _atom_site_fract_z 32 | _atom_site_occupancy 33 | Sr2+ Sr0 1 0.00000000 0.00000000 0.00000000 1 34 | Ti4+ Ti1 1 0.50000000 0.50000000 0.50000000 1 35 | O2- O2 1 0.50000000 0.00000000 0.50000000 1 36 | O2- O3 1 0.50000000 0.50000000 0.00000000 1 37 | O2- O4 1 0.00000000 0.50000000 0.50000000 1 38 | -------------------------------------------------------------------------------- /revision-scripts/5fold_split.py: -------------------------------------------------------------------------------- 1 | import json 2 | import os 3 | import random 4 | from sklearn.model_selection import KFold 5 | import fire 6 | 7 | def split_dataset(input_json, output_dir, n_splits=5, random_state=42): 8 | # Load the data 9 | with open(input_json, 'r') as f: 10 | data = json.load(f) 11 | 12 | # Create KFold object 13 | kf = KFold(n_splits=n_splits, shuffle=True, random_state=random_state) 14 | 15 | # Ensure output directory exists 16 | os.makedirs(output_dir, exist_ok=True) 17 | 18 | # Perform the split 19 | for fold, (train_index, test_index) in enumerate(kf.split(data), 1): 20 | train_data = [data[i] for i in train_index] 21 | test_data = [data[i] for i in test_index] 22 | 23 | # Save train data 24 | train_file = os.path.join(output_dir, f'train_mp_classification_fold_{fold}.json') 25 | with open(train_file, 'w') as f: 26 | json.dump(train_data, f, indent=2) 27 | 28 | # Save test data 29 | test_file = os.path.join(output_dir, f'test_mp_classification_fold_{fold}.json') 30 | with open(test_file, 'w') as f: 31 | json.dump(test_data, f, indent=2) 32 | 33 | print(f"Fold {fold} created: {train_file} and {test_file}") 34 | 35 | print("Dataset splitting completed.") 36 | 37 | if __name__ == "__main__": 38 | fire.Fire(split_dataset) -------------------------------------------------------------------------------- /src/mattext/__init__.py: -------------------------------------------------------------------------------- 1 | from importlib.metadata import version 2 | 3 | # Load the version 4 | __version__ = version("mattext") 5 | -------------------------------------------------------------------------------- /src/mattext/analysis/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/lamalab-org/MatText/866a0e24e723835f8fe799e326d730112fbc148a/src/mattext/analysis/__init__.py -------------------------------------------------------------------------------- /src/mattext/dataprep/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/lamalab-org/MatText/866a0e24e723835f8fe799e326d730112fbc148a/src/mattext/dataprep/__init__.py -------------------------------------------------------------------------------- /src/mattext/models/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/lamalab-org/MatText/866a0e24e723835f8fe799e326d730112fbc148a/src/mattext/models/__init__.py -------------------------------------------------------------------------------- /src/mattext/representations/__init__.py: -------------------------------------------------------------------------------- 1 | from xtal2txt.core import TextRep 2 | 3 | __all__ = ["TextRep"] 4 | -------------------------------------------------------------------------------- /src/mattext/representations/analysis.py: -------------------------------------------------------------------------------- 1 | from xtal2txt.analysis import ( 2 | ANALYSIS_MASK_TOKENS, 3 | ATOM_LIST_, 4 | CIF_ANALYSIS_DICT, 5 | COMPOSITION_ANALYSIS_DICT, 6 | CRYSTAL_LLM_ANALYSIS_DICT, 7 | NUMS_, 8 | SLICE_ANALYSIS_DICT, 9 | ) 10 | 11 | __all__ = [ 12 | "ATOM_LIST_", 13 | "NUMS_", 14 | "COMPOSITION_ANALYSIS_DICT", 15 | "SLICE_ANALYSIS_DICT", 16 | "CRYSTAL_LLM_ANALYSIS_DICT", 17 | "CIF_ANALYSIS_DICT", 18 | "ANALYSIS_MASK_TOKENS", 19 | ] 20 | -------------------------------------------------------------------------------- /src/mattext/representations/decoder.py: -------------------------------------------------------------------------------- 1 | from xtal2txt.decoder import DecodeTextRep, MatchRep 2 | 3 | __all__ = ["DecodeTextRep", "MatchRep"] 4 | -------------------------------------------------------------------------------- /src/mattext/representations/transformations.py: -------------------------------------------------------------------------------- 1 | from xtal2txt.transformations import TransformationCallback 2 | 3 | __all__ = ["TransformationCallback"] 4 | -------------------------------------------------------------------------------- /src/mattext/tokenizer/__init__.py: -------------------------------------------------------------------------------- 1 | from xtal2txt.tokenizer import ( 2 | CifTokenizer, 3 | CompositionTokenizer, 4 | CrysllmTokenizer, 5 | NumTokenizer, 6 | RobocrysTokenizer, 7 | SliceTokenizer, 8 | SmilesTokenizer, 9 | ) 10 | 11 | __all__ = [ 12 | "NumTokenizer", 13 | "SliceTokenizer", 14 | "CompositionTokenizer", 15 | "CifTokenizer", 16 | "CrysllmTokenizer", 17 | "SmilesTokenizer", 18 | "RobocrysTokenizer", 19 | ] 20 | -------------------------------------------------------------------------------- /src/mattext/utils.py: -------------------------------------------------------------------------------- 1 | import pystow 2 | 3 | mattext_storage = pystow.module("mattext") 4 | -------------------------------------------------------------------------------- /tests/test_imports.py: -------------------------------------------------------------------------------- 1 | from mattext.tokenizer import CifTokenizer, CompositionTokenizer, CrysllmTokenizer, NumTokenizer, RobocrysTokenizer, SliceTokenizer, SmilesTokenizer 2 | from mattext.representations import TextRep 3 | from pymatgen.core import Lattice, Structure, Molecule 4 | 5 | 6 | def test_textrep(): 7 | coords = [[0, 0, 0], [0.75,0.5,0.75]] 8 | lattice = Lattice.from_parameters(a=3.84, b=3.84, c=3.84, alpha=120, 9 | beta=90, gamma=60) 10 | struct = Structure(lattice, ["Si", "Si"], coords) 11 | 12 | textrep = TextRep.from_input(struct) 13 | 14 | slices = textrep.get_slices() 15 | assert isinstance(slices, str) -------------------------------------------------------------------------------- /tests/test_xtal2pot.py: -------------------------------------------------------------------------------- 1 | from mattext.analysis.xtal2pot import Xtal2Pot 2 | from pymatgen.core import Structure, Lattice 3 | 4 | 5 | def test_xtal2pot(): 6 | bcc_fe = Structure(Lattice.cubic(2.8), ["Fe", "Fe"], [[0, 0, 0], [0.5, 0.5, 0.5]]) 7 | xtal2pot = Xtal2Pot() 8 | potential = xtal2pot.get_potential(bcc_fe) 9 | assert isinstance(potential[0], float) 10 | assert isinstance(potential[1], float) 11 | assert len(potential) == 2 --------------------------------------------------------------------------------