Merge CI Fix and Buffer Overrun Issue (#1013)

Co-authored-by: Abhishek Jindal <[email protected]>
Co-authored-by: Ryan Hill <[email protected]>

.github/workflows/linux-cpu-x64-build.yml

@@ -68,8 +68,9 @@ jobs:
       - name: Install the python wheel and test dependencies
         run: |
           python3 -m pip install -r test/python/requirements.txt --user
-          python3 -m pip install -r test/python/requirements-cpu.txt --user
-          python3 -m pip install build/cpu/wheel/onnxruntime_genai*.whl --user --no-deps
+          python3 -m pip install -r test/python/cpu/torch/requirements.txt --user
+          python3 -m pip install -r test/python/cpu/ort/requirements.txt --user
+          python3 -m pip install --user --no-index --find-links build/cpu/wheel onnxruntime_genai
 
       - name: Use Dummy HuggingFace Token
         run: |

.github/workflows/linux-cpu-x64-nightly-build.yml

@@ -51,7 +51,8 @@ jobs:
       - name: Install the python wheel and test dependencies
         run: |
           python3 -m pip install -r test/python/requirements.txt --user
-          python3 -m pip install -r test/python/requirements-cpu.txt --user
+          python3 -m pip install -r test/python/cpu/torch/requirements.txt --user
+          python3 -m pip install -r test/python/cpu/ort/requirements.txt --user
           python3 -m pip install build/cpu/wheel/onnxruntime_genai*.whl --no-deps
 
       - name: Use Dummy HuggingFace Token

.github/workflows/linux-gpu-x64-build.yml

@@ -124,7 +124,8 @@ jobs:
             -e HF_TOKEN=$HF_TOKEN \
             -w /ort_genai_src onnxruntimecudabuildx64 bash -c " \
               ${{ env.PYTHON_EXECUTABLE }} -m pip install -r test/python/requirements.txt --user && \
-              ${{ env.PYTHON_EXECUTABLE }} -m pip install -r test/python/requirements-cuda.txt --user && \
+              ${{ env.PYTHON_EXECUTABLE }} -m pip install -r test/python/cuda/torch/requirements.txt --user && \
+              ${{ env.PYTHON_EXECUTABLE }} -m pip install -r test/python/cuda/ort/requirements.txt --user && \
               ${{ env.PYTHON_EXECUTABLE }} -m pip install /ort_genai_src/build/cuda/wheel/onnxruntime_genai*manylinux*.whl --no-deps --user && \
               ${{ env.PYTHON_EXECUTABLE }} test/python/test_onnxruntime_genai.py --cwd test/python --test_models test/test_models --e2e"
 

.github/workflows/mac-cpu-arm64-build.yml

@@ -54,7 +54,8 @@ jobs:
           python3 -m venv genai-macos-venv
           source genai-macos-venv/bin/activate
           python3 -m pip install -r test/python/requirements.txt
-          python3 -m pip install -r test/python/requirements-macos.txt
+          python3 -m pip install -r test/python/macos/torch/requirements.txt
+          python3 -m pip install -r test/python/macos/ort/requirements.txt
           python3 -m pip install build/cpu/osx-arm64/wheel/onnxruntime_genai*.whl --no-deps
 
       - name: Remove the ort lib and header files

.github/workflows/win-cpu-x64-build.yml

@@ -75,7 +75,8 @@ jobs:
     - name: Install the python wheel and test dependencies
       run: |
         python3 -m pip install -r test\python\requirements.txt --user
-        python3 -m pip install -r test\python\requirements-cpu.txt --user
+        python3 -m pip install -r test\python\cpu\torch\requirements.txt --user
+        python3 -m pip install -r test\python\cpu\ort\requirements.txt --user
         python3 -m pip install (Get-ChildItem ("$env:binaryDir\wheel\*.whl")) --no-deps
 
     - name: Use Dummy HuggingFace Token

.github/workflows/win-cuda-x64-build.yml

@@ -81,7 +81,8 @@ jobs:
     - name: Install the Python Wheel and Test Dependencies
       run: |
         python -m pip install -r test\python\requirements.txt
-        python -m pip install -r test\python\requirements-cuda.txt
+        python -m pip install -r test\python\cuda\torch\requirements.txt
+        python -m pip install -r test\python\cuda\ort\requirements.txt
         python -m pip install (Get-ChildItem ("$env:binaryDir\wheel\*.whl")) --no-deps
 
     - name: Use Dummy HuggingFace Token

.github/workflows/win-directml-x64-build.yml

@@ -89,7 +89,8 @@ jobs:
     - name: Install the Python Wheel and Test Dependencies
       run: |
         python -m pip install -r test\python\requirements.txt
-        python -m pip install -r test\python\requirements-directml.txt
+        python -m pip install -r test\python\directml\torch\requirements.txt
+        python -m pip install -r test\python\directml\ort\requirements.txt
         python -m pip install (Get-ChildItem ("$env:binaryDir\wheel\*.whl")) --no-deps
 
     - name: Verify Build Artifacts

.pipelines/stages/jobs/py-validation-job.yml

@@ -188,7 +188,8 @@ jobs:
             bash -c " \
                 export ORTGENAI_LOG_ORT_LIB=1 && \
                 $python_exe -m pip install -r /ort_genai_src/test/python/requirements.txt && \
-                $python_exe -m pip install -r /ort_genai_src/test/python/requirements-cuda.txt && \
+                $python_exe -m pip install -r /ort_genai_src/test/python/cuda/torch/requirements.txt && \
+                $python_exe -m pip install -r /ort_genai_src/test/python/cuda/ort/requirements.txt && \
                 cd /ort_genai_src/examples/python && \
                 $python_exe -m pip install --no-index --find-links=/ort_genai_binary/wheel $(pip_package_name) && \
                 $python_exe model-generate.py -m ./models/$(prebuild_phi3_mini_model_folder) --min_length 25 --max_length 50 --verbose"
@@ -200,10 +201,12 @@ jobs:
           export ORTGENAI_LOG_ORT_LIB=1
           python -m pip install -r test/python/requirements.txt
           if [[ "$(os)" == "linux" ]]; then
-            python -m pip install -r test/python/requirements-cpu.txt
+            python -m pip install -r test/python/cpu/torch/requirements.txt
+            python -m pip install -r test/python/cpu/ort/requirements.txt
           fi
           if [[ "$(os)" == "osx" ]]; then
-            python -m pip install -r test/python/requirements-macos.txt
+            python -m pip install -r test/python/macos/torch/requirements.txt
+            python -m pip install -r test/python/macos/ort/requirements.txt
           fi
           cd examples/python
           python -m pip install --no-index --find-links=$(Build.BinariesDirectory)/wheel $(pip_package_name)
@@ -224,13 +227,16 @@ jobs:
           $env:CUDA_PATH = '$(Build.Repository.LocalPath)\cuda_sdk\v$(cuda_version)'
           $env:PATH = "$env:CUDA_PATH\bin;$env:CUDA_PATH\extras\CUPTI\lib64;$env:PATH"
           Write-Host $env:PATH
-          python -m pip install -r test/python/requirements-cuda.txt
+          python -m pip install -r test/python/cuda/torch/requirements.txt
+          python -m pip install -r test/python/cuda/ort/requirements.txt
         }
         elseif ("$(ep)" -eq "directml") {
-          python -m pip install -r test/python/requirements-directml.txt
+          python -m pip install -r test/python/requirements-torch-directml.txt
+          python -m pip install -r test/python/requirements-torch-directml.txt
         }
         else {
-          python -m pip install -r test/python/requirements-cpu.txt
+          python -m pip install -r test/python/cpu/torch/requirements.txt
+          python -m pip install -r test/python/cpu/ort/requirements.txt
         }
         cd examples\python
         python -m pip install --no-index --find-links=$(Build.BinariesDirectory)/wheel $(pip_package_name)

src/ort_genai_c.cpp

@@ -372,12 +372,7 @@ OgaResult* OGA_API_CALL OgaTokenizerDecode(const OgaTokenizer* p, const int32_t*
   auto string = tokenizer.Decode({tokens, token_count});
   auto length = string.length() + 1;
   auto cstr_buffer = std::make_unique<char[]>(length);
-#if _MSC_VER
-  strcpy_s(cstr_buffer.get(), length, string.c_str());
-#else
-  strncpy(cstr_buffer.get(), string.c_str(), length);
-  cstr_buffer[length] = 0;
-#endif
+  memcpy(cstr_buffer.get(), string.c_str(), length);
   *out_string = cstr_buffer.release();
   return nullptr;
   OGA_CATCH
@@ -390,12 +385,7 @@ OgaResult* OGA_API_CALL OgaProcessorDecode(const OgaMultiModalProcessor* p, cons
   auto string = processor.tokenizer_->Decode({tokens, token_count});
   auto length = string.length() + 1;
   auto cstr_buffer = std::make_unique<char[]>(length);
-#if _MSC_VER
-  strcpy_s(cstr_buffer.get(), length, string.c_str());
-#else
-  strncpy(cstr_buffer.get(), string.c_str(), length);
-  cstr_buffer[length] = 0;
-#endif
+  memcpy(cstr_buffer.get(), string.c_str(), length);
   *out_string = cstr_buffer.release();
   return nullptr;
   OGA_CATCH

test/python/cpu/ort/requirements.txt

@@ -0,0 +1,2 @@
+-i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/
+onnxruntime==1.20.0.dev20241017001

test/python/cpu/torch/requirements.txt

@@ -0,0 +1,2 @@
+-f https://download.pytorch.org/whl/torch_stable.html
+torch==2.2.1+cpu

test/python/requirements-cuda.txt → test/python/cuda/ort/requirements.txt

@@ -1,4 +1,2 @@
--f https://download.pytorch.org/whl/torch_stable.html
 -i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/
-torch==2.2.1+cu121
 ort-nightly-gpu==1.20.0.dev20241007001

test/python/cuda/torch/requirements.txt

@@ -0,0 +1,2 @@
+-f https://download.pytorch.org/whl/torch_stable.html
+torch==2.2.1+cu121

test/python/requirements-directml.txt → test/python/directml/ort/requirements.txt

@@ -1,4 +1,2 @@
--f https://download.pytorch.org/whl/torch_stable.html
 -i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/
-torch==2.2.1+cpu
 ort-nightly-directml==1.20.0.dev20241006001

test/python/directml/torch/requirements.txt

@@ -0,0 +1,2 @@
+-f https://download.pytorch.org/whl/torch_stable.html
+torch==2.2.1+cpu

test/python/macos/ort/requirements.txt

@@ -0,0 +1,2 @@
+-i https://aiinfra.pkgs.visualstudio.com/PublicPackages/_packaging/ORT-Nightly/pypi/simple/
+onnxruntime==1.20.0.dev20241017001

test/python/macos/torch/requirements.txt

@@ -0,0 +1,2 @@
+-f https://download.pytorch.org/whl/torch_stable.html
+torch==2.2.1