update

.github/workflows/llm_performance_tests.yml

@@ -331,7 +331,7 @@ jobs:
 
           pip install --pre --upgrade %whl_name%[xpu] --extra-index-url https://pytorch-extension.intel.com/release-whl/stable/xpu/cn/
           if %ERRORLEVEL% neq 0 (exit /b 1)
-          pip install transformers==4.36.2
+          # pip install transformers==4.36.2
           pip list
 
           call conda deactivate
@@ -359,7 +359,6 @@ jobs:
       #      echo "Did not install ipex-llm with excepted version %TEST_VERSION_DATE%"
       #      exit /b 1
       #    )
-      #    pip install transformers==4.36.2
       #    pip list
 
       #    call conda deactivate

.github/workflows/llm_unit_tests.yml

@@ -210,12 +210,7 @@ jobs:
       - name: Run LLM cli test (Windows)
         if: runner.os == 'Windows' 
         uses: ./.github/actions/llm/cli-test-windows
-      # - name: Run LLM inference test
-        # shell: bash
-        # run: |
-          # python -m pip install einops datasets librosa openai-whisper
-          # bash python/llm/test/run-llm-inference-tests.sh
-      - name: Run LLM inference test for 4.36
+      - name: Run LLM inference test
         shell: bash
         run: |
           python -m pip install einops datasets librosa openai-whisper
@@ -239,6 +234,7 @@ jobs:
         shell: bash
         run: |
           pip uninstall sentence-transformers -y || true
+
   llm-unit-test-on-arc:
     needs: [setup-python-version, llm-cpp-build]
     strategy:

python/llm/setup.py

@@ -52,7 +52,7 @@
 libs_dir = os.path.join(llm_home, "ipex_llm", "libs")
 CONVERT_DEP = ['numpy == 1.26.4', # lastet 2.0.0b1 will cause error
                'torch',
-               'transformers == 4.31.0', 'sentencepiece', 'tokenizers == 0.13.3',
+               'transformers == 4.36.2', 'sentencepiece', 'tokenizers == 0.13.3',
                # TODO: Support accelerate 0.22.0
                'accelerate == 0.21.0', 'tabulate']
 SERVING_DEP = ['fschat[model_worker, webui] == 0.2.36', 'protobuf']
@@ -277,10 +277,9 @@ def setup_package():
 
     # Add internal requires for llama-index
     llama_index_requires = copy.deepcopy(all_requires)
-    for exclude_require in ['torch', 'transformers == 4.31.0', 'tokenizers == 0.13.3']:
+    for exclude_require in ['torch', 'tokenizers == 0.13.3']:
         llama_index_requires.remove(exclude_require)
     llama_index_requires += ["torch<2.2.0",
-                             "transformers>=4.34.0,<4.39.0",
                              "sentence-transformers~=2.6.1"]