Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[Dev] BUG Fix for bitnet integration #141

Merged
merged 189 commits into from
Aug 13, 2024

Conversation

LeiWang1999
Copy link
Contributor

This pull request includes several changes across multiple files to improve functionality, fix issues, and update dependencies. The most important changes include updating a submodule, modifying conditions for specific operations, and updating script behavior for model generation.

Submodule Update:

  • 3rdparty/tvm: Updated submodule commit to the latest version. ([3rdparty/tvmL1-R1](https://github.com/microsoft/BitBLAS/pull/141/files#diff-fa909c93fe94e9aa04c9e7f19e5754a2bb274678ad5c6275ee4bf54c6f9b1066L1-R1))

Code Enhancements:

  • bitblas/gpu/matmul_analysis.py: Modified the condition to check for dequantize_info to ensure it only supports float16 data type. ([bitblas/gpu/matmul_analysis.pyL626-R627](https://github.com/microsoft/BitBLAS/pull/141/files#diff-6f289ae4794dea10c7718ec09a9e100a564894c8a93c36559c9fa6d0688553e3L626-R627))
  • bitblas/ops/general_matmul/__init__.py: Added a temporary solution to legalize the dynamic symbolic, with a note to potentially remove it in the future. ([bitblas/ops/general_matmul/__init__.pyR88-R89](https://github.com/microsoft/BitBLAS/pull/141/files#diff-74fe5dd2824cb03a0fb2b0a913a2fc5caeb9c08e5368c318cd32b3af7e6f52edR88-R89))

Script Updates:

  • integration/BitNet/eval_correctness.py: Disabled flash_attention_2 and added functionality to quantize the model and compare generated text before and after quantization. ([integration/BitNet/eval_correctness.pyL75-R88](https://github.com/microsoft/BitBLAS/pull/141/files#diff-24610ce0fe8c0977dec03f95f17f709d384085cdb8ff49f86f563dba7dfa5ec4L75-R88))
  • integration/BitNet/maint/create_bitblas_ckpt.py: Disabled flash_attention_2 in the model configuration. ([integration/BitNet/maint/create_bitblas_ckpt.pyL71-R71](https://github.com/microsoft/BitBLAS/pull/141/files#diff-8b1a01ecd38f6ed3a87e4974faf672137b4a044a6137bc3ff7854a0d515689efL71-R71))
  • integration/BitNet/maint/generate_bitnet_model_bitblas_format.sh: Added a check to create the SAVED_MODEL_DIR directory if it does not exist. ([integration/BitNet/maint/generate_bitnet_model_bitblas_format.shR21-R23](https://github.com/microsoft/BitBLAS/pull/141/files#diff-d6095b66d1b0492e71833fb5a0f99593bf2261720a96ab43811a1b32f136b6c1R21-R23))

Dependency Update:

  • integration/BitNet/requirements.txt: Added transformers==4.40 to the requirements. ([integration/BitNet/requirements.txtR3](https://github.com/microsoft/BitBLAS/pull/141/files#diff-c274b4a63b1dc11eb5c71badfa85865c12befc3563dfa52d57f5c4655809c155R3))

@LeiWang1999 LeiWang1999 merged commit af697de into microsoft:main Aug 13, 2024
6 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

1 participant