├── .gitignore ├── .gitmodules ├── LICENSE ├── README.md ├── compress.py ├── img └── palu_idea.png ├── kernel ├── __init__.py ├── abx_rope.py ├── palu_attention.py ├── pytorch_reference.py └── test_palu_attention.py ├── longbench_utils ├── __init__.py ├── config │ ├── dataset2maxlen.json │ ├── dataset2prompt.json │ ├── model2maxlen.json │ └── model2path.json ├── metrics.py └── scorer.py ├── palu ├── __init__.py ├── data_utils.py ├── decomposition.py ├── model │ ├── __init__.py │ ├── modules │ │ ├── __init__.py │ │ ├── hadamard_utils.py │ │ ├── quant.py │ │ └── svd_linear.py │ ├── svd_llama │ │ ├── __init__.py │ │ ├── configuration_palu_llama.py │ │ └── modeling_palu_llama.py │ ├── svd_mistral │ │ ├── __init__.py │ │ ├── configuration_palu_mistral.py │ │ └── modeling_palu_mistral.py │ └── svd_qwen │ │ ├── __init__.py │ │ ├── configuration_palu_qwen.py │ │ └── modeling_palu_qwen.py ├── quant_utils.py └── rank_search.py ├── requirements.txt ├── run_latency_attention.py ├── run_latency_kernel.py ├── run_lm_eval.py ├── run_long_bench.py ├── run_ppl_eval.py └── utils.py /.gitignore: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/.gitignore -------------------------------------------------------------------------------- /.gitmodules: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/.gitmodules -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/LICENSE -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/README.md -------------------------------------------------------------------------------- /compress.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/compress.py -------------------------------------------------------------------------------- /img/palu_idea.png: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/img/palu_idea.png -------------------------------------------------------------------------------- /kernel/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /kernel/abx_rope.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/kernel/abx_rope.py -------------------------------------------------------------------------------- /kernel/palu_attention.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/kernel/palu_attention.py -------------------------------------------------------------------------------- /kernel/pytorch_reference.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/kernel/pytorch_reference.py -------------------------------------------------------------------------------- /kernel/test_palu_attention.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/kernel/test_palu_attention.py -------------------------------------------------------------------------------- /longbench_utils/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/longbench_utils/__init__.py -------------------------------------------------------------------------------- /longbench_utils/config/dataset2maxlen.json: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/longbench_utils/config/dataset2maxlen.json -------------------------------------------------------------------------------- /longbench_utils/config/dataset2prompt.json: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/longbench_utils/config/dataset2prompt.json -------------------------------------------------------------------------------- /longbench_utils/config/model2maxlen.json: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/longbench_utils/config/model2maxlen.json -------------------------------------------------------------------------------- /longbench_utils/config/model2path.json: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/longbench_utils/config/model2path.json -------------------------------------------------------------------------------- /longbench_utils/metrics.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/longbench_utils/metrics.py -------------------------------------------------------------------------------- /longbench_utils/scorer.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/longbench_utils/scorer.py -------------------------------------------------------------------------------- /palu/__init__.py: -------------------------------------------------------------------------------- 1 | -------------------------------------------------------------------------------- /palu/data_utils.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/data_utils.py -------------------------------------------------------------------------------- /palu/decomposition.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/decomposition.py -------------------------------------------------------------------------------- /palu/model/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/model/__init__.py -------------------------------------------------------------------------------- /palu/model/modules/__init__.py: -------------------------------------------------------------------------------- 1 | from .svd_linear import HeadwiseLowRankModule -------------------------------------------------------------------------------- /palu/model/modules/hadamard_utils.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/model/modules/hadamard_utils.py -------------------------------------------------------------------------------- /palu/model/modules/quant.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/model/modules/quant.py -------------------------------------------------------------------------------- /palu/model/modules/svd_linear.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/model/modules/svd_linear.py -------------------------------------------------------------------------------- /palu/model/svd_llama/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/model/svd_llama/__init__.py -------------------------------------------------------------------------------- /palu/model/svd_llama/configuration_palu_llama.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/model/svd_llama/configuration_palu_llama.py -------------------------------------------------------------------------------- /palu/model/svd_llama/modeling_palu_llama.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/model/svd_llama/modeling_palu_llama.py -------------------------------------------------------------------------------- /palu/model/svd_mistral/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/model/svd_mistral/__init__.py -------------------------------------------------------------------------------- /palu/model/svd_mistral/configuration_palu_mistral.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/model/svd_mistral/configuration_palu_mistral.py -------------------------------------------------------------------------------- /palu/model/svd_mistral/modeling_palu_mistral.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/model/svd_mistral/modeling_palu_mistral.py -------------------------------------------------------------------------------- /palu/model/svd_qwen/__init__.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/model/svd_qwen/__init__.py -------------------------------------------------------------------------------- /palu/model/svd_qwen/configuration_palu_qwen.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/model/svd_qwen/configuration_palu_qwen.py -------------------------------------------------------------------------------- /palu/model/svd_qwen/modeling_palu_qwen.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/model/svd_qwen/modeling_palu_qwen.py -------------------------------------------------------------------------------- /palu/quant_utils.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/quant_utils.py -------------------------------------------------------------------------------- /palu/rank_search.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/palu/rank_search.py -------------------------------------------------------------------------------- /requirements.txt: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/requirements.txt -------------------------------------------------------------------------------- /run_latency_attention.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/run_latency_attention.py -------------------------------------------------------------------------------- /run_latency_kernel.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/run_latency_kernel.py -------------------------------------------------------------------------------- /run_lm_eval.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/run_lm_eval.py -------------------------------------------------------------------------------- /run_long_bench.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/run_long_bench.py -------------------------------------------------------------------------------- /run_ppl_eval.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/run_ppl_eval.py -------------------------------------------------------------------------------- /utils.py: -------------------------------------------------------------------------------- https://raw.githubusercontent.com/shadowpa0327/Palu/HEAD/utils.py --------------------------------------------------------------------------------