update setup to build auto-gptq cuda extensions

Dockerfile

@@ -49,10 +49,9 @@ RUN git clone https://github.com/smallcloudai/refact-lsp.git /tmp/refact-lsp \
     && cargo install --path . \
     && rm -rf /tmp/refact-lsp
 
-ENV INSTALL_OPTIONAL=TRUE
 ENV BUILD_CUDA_EXT=1
-ENV GITHUB_ACTIONS=true
 ENV TORCH_CUDA_ARCH_LIST="6.0;6.1;7.0;7.5;8.0;8.6;8.9;9.0+PTX"
+ENV INSTALL_OPTIONAL=TRUE
 ENV FLASH_ATTENTION_FORCE_BUILD=TRUE
 ENV MAX_JOBS=8
 COPY . /tmp/app

setup.py

@@ -43,9 +43,10 @@ class PyPackage:
     ),
     "self_hosting_machinery": PyPackage(
         requires=["aiohttp", "aiofiles", "cryptography", "fastapi==0.100.0", "giturlparse", "pydantic==1.10.13",
-                  "starlette==0.27.0", "uvicorn", "uvloop", "python-multipart", "auto-gptq==0.6.0", "accelerate",
+                  "starlette==0.27.0", "uvicorn", "uvloop", "python-multipart", "accelerate",
                   "termcolor", "torch", "transformers==4.36.2", "bitsandbytes", "safetensors", "peft", "triton",
-                  "torchinfo", "mpi4py", "deepspeed==0.11.1", "cassandra_driver==3.28.0", "pandas>=2.0.3", "litellm"],
+                  "torchinfo", "mpi4py", "deepspeed==0.11.1", "cassandra_driver==3.28.0", "pandas>=2.0.3", "litellm",
+                  "auto-gptq @ git+https://github.com/PanQiWei/[email protected]"],
         optional=["ninja", "flash_attn @ git+https://github.com/smallcloudai/flash-attention@feat/alibi"],
         requires_packages=["refact_scratchpads", "refact_scratchpads_no_gpu",
                            "known_models_db", "refact_data_pipeline"],