add ipex tests

Author: echarlaix

optimum/intel/utils/import_utils.py

@@ -184,6 +184,15 @@
         _sentence_transformers_available = False
 
 
+_langchain_available = importlib.util.find_spec("langchain") is not None
+_langchain_version = "N/A"
+if _langchain_available:
+    try:
+        _langchain_version = importlib.metadata.version("langchain")
+    except importlib.metadata.PackageNotFoundError:
+        _langchain_available = False
+
+
 def is_transformers_available():
     return _transformers_available
 

tests/ipex/test_modeling.py

@@ -21,6 +21,7 @@
 import numpy as np
 import requests
 import torch
+from typing import Generator
 from parameterized import parameterized
 from PIL import Image
 from transformers import (
@@ -34,6 +35,7 @@
     pipeline,
     set_seed,
 )
+from packaging import version
 from optimum.intel import (
     IPEXModel,
     IPEXModelForAudioClassification,
@@ -47,7 +49,12 @@
     IPEXSentenceTransformer,
 )
 from optimum.utils.testing_utils import grid_parameters, require_sentence_transformers
-from optimum.intel.utils.import_utils import is_sentence_transformers_available, is_torch_version
+from optimum.intel.utils.import_utils import (
+    is_sentence_transformers_available,
+    is_torch_version,
+    _langchain_available,
+    _langchain_version,
+)
 
 if is_sentence_transformers_available():
     from sentence_transformers import SentenceTransformer
@@ -707,3 +714,35 @@ def test_sentence_transformers_save_and_infer(self, model_arch):
             model = IPEXSentenceTransformer(tmpdirname, model_kwargs={"subfolder": "ipex"})
             sentences = ["This is an example sentence", "Each sentence is converted"]
             model.encode(sentences)
+
+
+class IPEXLangchainTest(unittest.TestCase):
+    SUPPORTED_ARCHITECTURES = ("gpt2",)
+
+    @unittest.skipIf(
+        not _langchain_available or version.parse(_langchain_version) <= version.parse("0.3.30"),
+        reason="Unsupported langchain",
+    )
+    @parameterized.expand(SUPPORTED_ARCHITECTURES)
+    def test_huggingface_pipeline_streaming(self, model_arch):
+        from langchain_huggingface import HuggingFacePipeline
+
+        model_id = MODEL_NAMES[model_arch]
+
+        hf_pipe = HuggingFacePipeline.from_model_id(
+            model_id=model_id,
+            task="text-generation",
+            pipeline_kwargs={"max_new_tokens": 10},
+            backend="ipex",
+        )
+
+        generator = hf_pipe.stream("Q: How do you say 'hello' in German? A:'", stop=["."])
+
+        self.assertIsInstance(generator, Generator)
+
+        stream_results_string = ""
+        for chunk in generator:
+            self.assertIsInstance(chunk, str)
+            stream_results_string = chunk
+
+        self.assertTrue(len(stream_results_string.strip()) > 1)

tests/openvino/test_modeling.py

@@ -14,7 +14,6 @@
 
 import copy
 import gc
-import importlib
 import os
 import platform
 import tempfile
@@ -109,7 +108,12 @@
     _print_compiled_model_properties,
 )
 from optimum.intel.pipelines import pipeline as optimum_pipeline
-from optimum.intel.utils.import_utils import is_openvino_version, is_transformers_version
+from optimum.intel.utils.import_utils import (
+    _langchain_available,
+    _langchain_version,
+    is_openvino_version,
+    is_transformers_version,
+)
 from optimum.intel.utils.modeling_utils import _find_files_matching_pattern
 from optimum.utils import (
     DIFFUSION_MODEL_TEXT_ENCODER_2_SUBFOLDER,
@@ -132,15 +136,6 @@
 F32_CONFIG = {"INFERENCE_PRECISION_HINT": "f32"}
 
 
-_langchain_available = importlib.util.find_spec("langchain") is not None
-_langchain_version = "N/A"
-if _langchain_available:
-    try:
-        _langchain_version = importlib.metadata.version("langchain")
-    except importlib.metadata.PackageNotFoundError:
-        _langchain_available = False
-
-
 class Timer(object):
     def __enter__(self):
         self.elapsed = time.perf_counter()