├── .gitattributes ├── .gitignore ├── .gitmodules ├── README.md ├── autodeco_megatron ├── nemo_converter │ ├── deepseek_v3 │ │ ├── fp8_cast_bf16.py │ │ ├── kernel.py │ │ └── nemo_import_deepseek_v3.py │ ├── gpt_oss │ │ ├── nemo_import_gpt_oss_120b.py │ │ └── nemo_import_gpt_oss_20b.py │ ├── nemo_export.py │ └── qwen3 │ │ ├── nemo_import_qwen3_235_a22b.py │ │ └── nemo_import_qwen3_30_a3b.py ├── nemo_trainer.py ├── py_packages │ ├── Megatron-LM-main.zip │ ├── NeMo-main.zip │ └── apex-25.09.zip ├── readme.md ├── scripts │ ├── deepseek │ │ └── end2end.sh │ ├── gpt_oss │ │ └── end2end_120b.sh │ └── qwen3 │ │ └── end2end_235b_a22b.sh └── src │ ├── __init__.py │ ├── data │ ├── __init__.py │ └── auto_deco_dataset.py │ ├── loss_utils.py │ ├── override_functions.py │ └── pl_models │ ├── __init__.py │ ├── common.py │ ├── deepseek_v3.py │ ├── gpt_oss.py │ ├── llama.py │ ├── qwen25.py │ └── qwen3.py ├── config └── deepspeed │ ├── __init__.py │ ├── deepspeed_zero3_gradaccu4.yaml │ └── templlm_gptoss.py ├── figure └── arch.png ├── model ├── __init__.py └── templlm_auto.py ├── requirements.txt ├── script ├── construct_autodeco.py ├── merge_autodeco.py ├── test_generation_tp4.sh ├── test_generation_tp8.sh └── trl_train.sh ├── trainer ├── __init__.py └── trl_autodeco.py ├── trl_train.py └── utils ├── boxed_extract.py ├── compute_acc_stats.py └── llm_eval.py /.gitattributes: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/.gitattributes -------------------------------------------------------------------------------- /.gitignore: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/.gitignore -------------------------------------------------------------------------------- /.gitmodules: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/.gitmodules -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/README.md -------------------------------------------------------------------------------- /autodeco_megatron/nemo_converter/deepseek_v3/fp8_cast_bf16.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/nemo_converter/deepseek_v3/fp8_cast_bf16.py -------------------------------------------------------------------------------- /autodeco_megatron/nemo_converter/deepseek_v3/kernel.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/nemo_converter/deepseek_v3/kernel.py -------------------------------------------------------------------------------- /autodeco_megatron/nemo_converter/deepseek_v3/nemo_import_deepseek_v3.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/nemo_converter/deepseek_v3/nemo_import_deepseek_v3.py -------------------------------------------------------------------------------- /autodeco_megatron/nemo_converter/gpt_oss/nemo_import_gpt_oss_120b.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/nemo_converter/gpt_oss/nemo_import_gpt_oss_120b.py -------------------------------------------------------------------------------- /autodeco_megatron/nemo_converter/gpt_oss/nemo_import_gpt_oss_20b.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/nemo_converter/gpt_oss/nemo_import_gpt_oss_20b.py -------------------------------------------------------------------------------- /autodeco_megatron/nemo_converter/nemo_export.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/nemo_converter/nemo_export.py -------------------------------------------------------------------------------- /autodeco_megatron/nemo_converter/qwen3/nemo_import_qwen3_235_a22b.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/nemo_converter/qwen3/nemo_import_qwen3_235_a22b.py -------------------------------------------------------------------------------- /autodeco_megatron/nemo_converter/qwen3/nemo_import_qwen3_30_a3b.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/nemo_converter/qwen3/nemo_import_qwen3_30_a3b.py -------------------------------------------------------------------------------- /autodeco_megatron/nemo_trainer.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/nemo_trainer.py -------------------------------------------------------------------------------- /autodeco_megatron/py_packages/Megatron-LM-main.zip: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/py_packages/Megatron-LM-main.zip -------------------------------------------------------------------------------- /autodeco_megatron/py_packages/NeMo-main.zip: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/py_packages/NeMo-main.zip -------------------------------------------------------------------------------- /autodeco_megatron/py_packages/apex-25.09.zip: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/py_packages/apex-25.09.zip -------------------------------------------------------------------------------- /autodeco_megatron/readme.md: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/readme.md -------------------------------------------------------------------------------- /autodeco_megatron/scripts/deepseek/end2end.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/scripts/deepseek/end2end.sh -------------------------------------------------------------------------------- /autodeco_megatron/scripts/gpt_oss/end2end_120b.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/scripts/gpt_oss/end2end_120b.sh -------------------------------------------------------------------------------- /autodeco_megatron/scripts/qwen3/end2end_235b_a22b.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/scripts/qwen3/end2end_235b_a22b.sh -------------------------------------------------------------------------------- /autodeco_megatron/src/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /autodeco_megatron/src/data/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /autodeco_megatron/src/data/auto_deco_dataset.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/src/data/auto_deco_dataset.py -------------------------------------------------------------------------------- /autodeco_megatron/src/loss_utils.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/src/loss_utils.py -------------------------------------------------------------------------------- /autodeco_megatron/src/override_functions.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/src/override_functions.py -------------------------------------------------------------------------------- /autodeco_megatron/src/pl_models/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /autodeco_megatron/src/pl_models/common.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/src/pl_models/common.py -------------------------------------------------------------------------------- /autodeco_megatron/src/pl_models/deepseek_v3.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/src/pl_models/deepseek_v3.py -------------------------------------------------------------------------------- /autodeco_megatron/src/pl_models/gpt_oss.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/src/pl_models/gpt_oss.py -------------------------------------------------------------------------------- /autodeco_megatron/src/pl_models/llama.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/src/pl_models/llama.py -------------------------------------------------------------------------------- /autodeco_megatron/src/pl_models/qwen25.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/src/pl_models/qwen25.py -------------------------------------------------------------------------------- /autodeco_megatron/src/pl_models/qwen3.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/autodeco_megatron/src/pl_models/qwen3.py -------------------------------------------------------------------------------- /config/deepspeed/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/config/deepspeed/__init__.py -------------------------------------------------------------------------------- /config/deepspeed/deepspeed_zero3_gradaccu4.yaml: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/config/deepspeed/deepspeed_zero3_gradaccu4.yaml -------------------------------------------------------------------------------- /config/deepspeed/templlm_gptoss.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/config/deepspeed/templlm_gptoss.py -------------------------------------------------------------------------------- /figure/arch.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/figure/arch.png -------------------------------------------------------------------------------- /model/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/model/__init__.py -------------------------------------------------------------------------------- /model/templlm_auto.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/model/templlm_auto.py -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- 1 | accelerate 2 | trl 3 | deepspeed 4 | orjson 5 | transformers>=4.55.0 -------------------------------------------------------------------------------- /script/construct_autodeco.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/script/construct_autodeco.py -------------------------------------------------------------------------------- /script/merge_autodeco.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/script/merge_autodeco.py -------------------------------------------------------------------------------- /script/test_generation_tp4.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/script/test_generation_tp4.sh -------------------------------------------------------------------------------- /script/test_generation_tp8.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/script/test_generation_tp8.sh -------------------------------------------------------------------------------- /script/trl_train.sh: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/script/trl_train.sh -------------------------------------------------------------------------------- /trainer/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/trainer/__init__.py -------------------------------------------------------------------------------- /trainer/trl_autodeco.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/trainer/trl_autodeco.py -------------------------------------------------------------------------------- /trl_train.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/trl_train.py -------------------------------------------------------------------------------- /utils/boxed_extract.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/utils/boxed_extract.py -------------------------------------------------------------------------------- /utils/compute_acc_stats.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/utils/compute_acc_stats.py -------------------------------------------------------------------------------- /utils/llm_eval.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/Zacks917/AutoDeco/HEAD/utils/llm_eval.py --------------------------------------------------------------------------------