Removed stubs from runtime

mutablelogic · Aug 11, 2024 · abc2cc1 · abc2cc1
1 parent 1b42196
commit abc2cc1
Show file tree

Hide file tree

Showing 3 changed files with 3 additions and 5 deletions.
diff --git a/Dockerfile.cuda b/Dockerfile.cuda
@@ -52,11 +52,9 @@ RUN apt-get update \
   && apt-get update \
   && apt-get -y install cuda-libraries-${CUDA} \  
   && apt-get clean
-COPY --from=build /usr/local/cuda/lib64/stubs /usr/local/cuda/lib64/stubs
-COPY --from=build /usr/local/cuda/lib64/stubs/libcuda.so /usr/local/cuda/lib64/stubs/libcuda.so.1
 
 FROM ${TARGET}
-ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64:/usr/local/cuda/lib64/stubs
+ENV LD_LIBRARY_PATH=/usr/local/cuda/lib64
 ENV PATH=/usr/local/cuda/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin
 ENV NVIDIA_ARCH="${ARCH}"
 ENV NVIDIA_CUDA_VERSION="${CUDA}"

diff --git a/llama.cpp b/llama.cpp
diff --git a/onnxruntime b/onnxruntime
+8 −14		Makefile
+14 −1		common/common.cpp
+2 −0		common/common.h
+139 −0		convert_hf_to_gguf.py
+98 −42		examples/embedding/embedding.cpp
+0 −14		examples/export-lora/export-lora.cpp
+7 −10		examples/llava/clip.cpp
+4 −0		examples/rpc/README.md
+12 −1		examples/rpc/rpc-server.cpp
+1 −1		ggml/src/ggml-metal.m
+35 −1		ggml/src/ggml-rpc.cpp
+2 −1		ggml/src/ggml.c
+0 −1		gguf-py/examples/writer.py
+17 −0		gguf-py/gguf/constants.py
+3 −0		include/llama.h
+15 −0		src/llama-impl.h
+0 −14		src/llama-vocab.cpp
+405 −251		src/llama.cpp
+59 −0		docs/ContribOperators.md
+1 −0		docs/OperatorKernels.md
+1 −1		include/onnxruntime/core/providers/tensorrt/tensorrt_provider_options.h
+1 −1		js/build_jsep.bat
+1 −0		js/web/docs/webgpu-operators.md
+2 −0		js/web/lib/wasm/jsep/webgpu/op-resolve-rules.ts
+219 −0		js/web/lib/wasm/jsep/webgpu/ops/quantize-linear.ts
+385 −0		js/web/test/data/ops/dequantizelinear.jsonc
+3 −2		js/web/test/suite-test-list.jsonc
+9 −0		onnxruntime/contrib_ops/cpu/cpu_contrib_kernels.cc
+282 −0		onnxruntime/contrib_ops/cpu/quantization/gather_block_quantized.cc
+118 −0		onnxruntime/core/graph/contrib_ops/contrib_defs.cc
+1 −1		onnxruntime/core/graph/model.cc
+1 −1		onnxruntime/core/graph/model.h
+62 −31		onnxruntime/core/optimizer/pad_fusion.cc
+25 −0		onnxruntime/core/providers/js/js_execution_provider.cc
+54 −0		onnxruntime/core/providers/js/operators/quantize_linear.cc
+31 −0		onnxruntime/core/providers/js/operators/quantize_linear.h
+17 −16		onnxruntime/core/providers/openvino/qdq_transformations/qdq_stripping.cc
+17 −7		onnxruntime/core/providers/tensorrt/tensorrt_execution_provider.cc
+1 −2		onnxruntime/core/providers/tensorrt/tensorrt_execution_provider.h
+1 −1		onnxruntime/core/providers/tensorrt/tensorrt_execution_provider_info.h
+0 −2		onnxruntime/python/tools/quantization/base_quantizer.py
+14 −8		onnxruntime/python/tools/quantization/matmul_4bits_quantizer.py
+8 −5		onnxruntime/python/tools/quantization/operators/matmul.py
+1 −2		onnxruntime/python/tools/quantization/qdq_quantizer.py
+2 −2		onnxruntime/python/tools/quantization/tensor_quant_overrides.py
+468 −0		onnxruntime/test/contrib_ops/gather_block_quantized_op_test.cc
+3 −0		onnxruntime/test/python/quantization/test_op_matmul_4bits.py
+44 −0		onnxruntime/test/python/quantization/test_tensor_quant_overrides_option.py
+35 −0		orttraining/orttraining/test/training_api/core/training_capi_tests.cc
+11 −4		orttraining/orttraining/training_api/module.cc
+1 −0		orttraining/orttraining/training_api/module.h
+1 −0		tools/ci_build/github/azure-pipelines/templates/win-web-multi-browsers.yml