Fix UT env

.azure-pipelines/scripts/fwk_version.sh

@@ -2,9 +2,9 @@
 
 echo "export FWs version..."
 export tensorflow_version='2.15.0-official'
-export pytorch_version='2.3.0+cpu'
-export torchvision_version='0.18.0+cpu'
-export ipex_version='2.3.0+cpu'
+export pytorch_version='2.4.0+cpu'
+export torchvision_version='0.19.0'
+export ipex_version='2.4.0+cpu'
 export onnx_version='1.16.0'
 export onnxruntime_version='1.18.0'
 export mxnet_version='1.9.1'

.azure-pipelines/scripts/install_nc.sh

@@ -5,14 +5,16 @@ cd /neural-compressor
 if [[ $1 = *"3x_pt"* ]]; then
     if [[ $1 != *"3x_pt_fp8"* ]]; then
         echo -e "\n Install torch CPU ... "
-        pip install torch==2.3.0 --index-url https://download.pytorch.org/whl/cpu
+        pip install torch==2.4.0 --index-url https://download.pytorch.org/whl/cpu
     fi
     python -m pip install --no-cache-dir -r requirements_pt.txt
-    python setup.py pt bdist_wheel
+    # python setup.py pt bdist_wheel
+    python setup.py bdist_wheel
     pip install --no-deps dist/neural_compressor*.whl --force-reinstall
 elif [[ $1 = *"3x_tf"* ]]; then
     python -m pip install --no-cache-dir -r requirements_tf.txt
-    python setup.py tf bdist_wheel
+    # python setup.py tf bdist_wheel
+    python setup.py bdist_wheel
     pip install dist/neural_compressor*.whl --force-reinstall
 else
     python -m pip install --no-cache-dir -r requirements.txt

.azure-pipelines/scripts/models/env_setup.sh

@@ -51,6 +51,10 @@ SCRIPTS_PATH="/neural-compressor/.azure-pipelines/scripts/models"
 log_dir="/neural-compressor/.azure-pipelines/scripts/models"
 if [[ "${inc_new_api}" == "3x"* ]]; then
     WORK_SOURCE_DIR="/neural-compressor/examples/3.x_api/${framework}"
+    git clone https://github.com/intel/intel-extension-for-transformers.git /itrex
+    cd /itrex
+    pip install -r requirements.txt
+    pip install -v .
 else
     WORK_SOURCE_DIR="/neural-compressor/examples/${framework}"
 fi
@@ -95,8 +99,8 @@ if [[ "${fwk_ver}" != "latest" ]]; then
             pip install intel-tensorflow==${fwk_ver}
         fi
     elif [[ "${framework}" == "pytorch" ]]; then
-        pip install torch==${fwk_ver} -f https://download.pytorch.org/whl/torch_stable.html
-        pip install torchvision==${torch_vision_ver} -f https://download.pytorch.org/whl/torch_stable.html
+        pip install torch==${fwk_ver} --index-url https://download.pytorch.org/whl/cpu
+        pip install torchvision==${torch_vision_ver} --index-url https://download.pytorch.org/whl/cpu
     elif [[ "${framework}" == "onnxrt" ]]; then
         pip install onnx==1.15.0
         pip install onnxruntime==${fwk_ver}

.azure-pipelines/scripts/ut/run_basic_pt_pruning.sh

@@ -4,9 +4,9 @@ test_case="run basic pt pruning"
 echo "${test_case}"
 
 echo "specify fwk version..."
-export pytorch_version='2.3.0+cpu'
+export pytorch_version='2.4.0+cpu'
 export torchvision_version='0.18.0+cpu'
-export ipex_version='2.3.0+cpu'
+export ipex_version='2.4.0+cpu'
 
 echo "set up UT env..."
 bash /neural-compressor/.azure-pipelines/scripts/ut/env_setup.sh "${test_case}"

.azure-pipelines/template/docker-template.yml

@@ -42,6 +42,7 @@ steps:
       - checkout: self
         clean: true
         displayName: "Checkout out Repo"
+        fetchDepth: 0
 
   - ${{ if eq(parameters.dockerConfigName, 'gitCloneDockerConfig') }}:
       - script: |

examples/3.x_api/pytorch/nlp/huggingface_models/language-modeling/quantization/weight_only/requirements.txt

@@ -9,7 +9,7 @@ wandb
 einops
 neural-compressor
 intel-extension-for-transformers
-lm_eval==0.4.2
+lm_eval==0.4.3
 peft
 auto_round
 intel_extension_for_pytorch

neural_compressor/torch/algorithms/pt2e_quant/utility.py

@@ -21,7 +21,7 @@
 from torch.ao.quantization.quantizer import QuantizationSpec
 from torch.ao.quantization.quantizer.x86_inductor_quantizer import QuantizationConfig, X86InductorQuantizer
 
-from neural_compressor.torch.utils import GT_TORCH_VERSION_2_3_2
+from neural_compressor.torch.utils import GT_OR_EQUAL_TORCH_VERSION_2_5
 
 
 def create_quant_spec_from_config(dtype, sym, granularity, algo, is_dynamic=False) -> QuantizationSpec:
@@ -102,8 +102,8 @@ def create_xiq_quantizer_from_pt2e_config(config, is_dynamic=False) -> X86Induct
     # set global
     global_config = _map_inc_config_to_torch_quant_config(config, is_dynamic)
     quantizer.set_global(global_config)
-    # need torch >= 2.3.2
-    if GT_TORCH_VERSION_2_3_2:  # pragma: no cover
+    # need torch >= 2.5
+    if GT_OR_EQUAL_TORCH_VERSION_2_5:  # pragma: no cover
         op_type_config_dict, op_name_config_dict = config._get_op_name_op_type_config()
         if op_type_config_dict:
             for op_type, config in op_type_config_dict.items():

neural_compressor/torch/utils/environ.py

@@ -104,7 +104,7 @@ def get_torch_version():
     return version
 
 
-GT_TORCH_VERSION_2_3_2 = get_torch_version() > Version("2.3.2")
+GT_OR_EQUAL_TORCH_VERSION_2_5 = get_torch_version() >= Version("2.5")
 
 
 def get_accelerator(device_name="auto"):

test/3x/torch/quantization/test_pt2e_quant.py

@@ -15,7 +15,7 @@
     prepare,
     quantize,
 )
-from neural_compressor.torch.utils import GT_TORCH_VERSION_2_3_2, TORCH_VERSION_2_2_2, get_torch_version
+from neural_compressor.torch.utils import GT_OR_EQUAL_TORCH_VERSION_2_5, TORCH_VERSION_2_2_2, get_torch_version
 
 torch.manual_seed(0)
 
@@ -131,7 +131,7 @@ def calib_fn(model):
         logger.warning("out shape is %s", out.shape)
         assert out is not None
 
-    @pytest.mark.skipif(not GT_TORCH_VERSION_2_3_2, reason="Requires torch>=2.3.2")
+    @pytest.mark.skipif(not GT_OR_EQUAL_TORCH_VERSION_2_5, reason="Requires torch>=2.5")
     def test_quantize_simple_model_with_set_local(self, force_not_import_ipex):
         model, example_inputs = self.build_simple_torch_model_and_example_inputs()
         float_model_output = model(*example_inputs)
@@ -243,7 +243,7 @@ def get_node_in_graph(graph_module):
                     nodes_in_graph[n] = 1
         return nodes_in_graph
 
-    @pytest.mark.skipif(not GT_TORCH_VERSION_2_3_2, reason="Requires torch>=2.3.0")
+    @pytest.mark.skipif(not GT_OR_EQUAL_TORCH_VERSION_2_5, reason="Requires torch>=2.5")
     def test_mixed_fp16_and_int8(self, force_not_import_ipex):
         model, example_inputs = self.build_model_include_conv_and_linear()
         model = export(model, example_inputs=example_inputs)