Add eval method for OVModels (#591)

Author: echarlaix

optimum/intel/openvino/modeling_base.py

@@ -448,6 +448,9 @@ def half(self):
         self.request = None
         return self
 
+    def eval(self):
+        return self
+
     def forward(self, *args, **kwargs):
         raise NotImplementedError
 

tests/openvino/test_modeling.py

@@ -272,6 +272,7 @@ def test_compare_to_transformers(self, model_arch):
     def test_pipeline(self, model_arch):
         model_id = MODEL_NAMES[model_arch]
         model = OVModelForSequenceClassification.from_pretrained(model_id, export=True, compile=False)
+        model.eval()
         tokenizer = AutoTokenizer.from_pretrained(model_id)
         pipe = pipeline("text-classification", model=model, tokenizer=tokenizer)
         text = "This restaurant is awesome"
@@ -345,6 +346,7 @@ def test_compare_to_transformers(self, model_arch):
     def test_pipeline(self, model_arch):
         model_id = MODEL_NAMES[model_arch]
         model = OVModelForQuestionAnswering.from_pretrained(model_id, export=True)
+        model.eval()
         tokenizer = AutoTokenizer.from_pretrained(model_id)
         pipe = pipeline("question-answering", model=model, tokenizer=tokenizer)
         question = "What's my name?"
@@ -411,6 +413,7 @@ def test_compare_to_transformers(self, model_arch):
     def test_pipeline(self, model_arch):
         model_id = MODEL_NAMES[model_arch]
         model = OVModelForTokenClassification.from_pretrained(model_id, export=True)
+        model.eval()
         tokenizer = AutoTokenizer.from_pretrained(model_id)
         pipe = pipeline("token-classification", model=model, tokenizer=tokenizer)
         outputs = pipe("My Name is Arthur and I live in Lyon.")
@@ -460,6 +463,7 @@ def test_compare_to_transformers(self, model_arch):
     def test_pipeline(self, model_arch):
         model_id = MODEL_NAMES[model_arch]
         model = OVModelForFeatureExtraction.from_pretrained(model_id, export=True)
+        model.eval()
         tokenizer = AutoTokenizer.from_pretrained(model_id)
         pipe = pipeline("feature-extraction", model=model, tokenizer=tokenizer)
         outputs = pipe("My Name is Arthur and I live in Lyon.")
@@ -568,6 +572,7 @@ def test_pipeline(self, model_arch):
         model = OVModelForCausalLM.from_pretrained(
             model_id, export=True, use_cache=False, compile=False, **model_kwargs
         )
+        model.eval()
         model.config.encoder_no_repeat_ngram_size = 0
         model.to("cpu")
         model.half()
@@ -758,6 +763,7 @@ def test_compare_to_transformers(self, model_arch):
     def test_pipeline(self, model_arch):
         model_id = MODEL_NAMES[model_arch]
         model = OVModelForMaskedLM.from_pretrained(model_id, export=True)
+        model.eval()
         tokenizer = AutoTokenizer.from_pretrained(model_id)
         pipe = pipeline("fill-mask", model=model, tokenizer=tokenizer)
         outputs = pipe(f"This is a {tokenizer.mask_token}.")
@@ -815,6 +821,7 @@ def test_compare_to_transformers(self, model_arch):
     def test_pipeline(self, model_arch):
         model_id = MODEL_NAMES[model_arch]
         model = OVModelForImageClassification.from_pretrained(model_id, export=True)
+        model.eval()
         preprocessor = AutoFeatureExtractor.from_pretrained(model_id)
         pipe = pipeline("image-classification", model=model, feature_extractor=preprocessor)
         outputs = pipe("http://images.cocodataset.org/val2017/000000039769.jpg")
@@ -911,6 +918,7 @@ def test_pipeline(self, model_arch):
         model_id = MODEL_NAMES[model_arch]
         tokenizer = AutoTokenizer.from_pretrained(model_id)
         model = OVModelForSeq2SeqLM.from_pretrained(model_id, export=True, compile=False)
+        model.eval()
         model.half()
         model.to("cpu")
         model.compile()
@@ -1044,6 +1052,7 @@ def test_compare_to_transformers(self, model_arch):
     def test_pipeline(self, model_arch):
         model_id = MODEL_NAMES[model_arch]
         model = OVModelForAudioClassification.from_pretrained(model_id, export=True)
+        model.eval()
         preprocessor = AutoFeatureExtractor.from_pretrained(model_id)
         pipe = pipeline("audio-classification", model=model, feature_extractor=preprocessor)
         outputs = pipe([np.random.random(16000)])
@@ -1354,6 +1363,7 @@ def test_compare_to_transformers(self, model_arch):
     def test_pipeline(self, model_arch):
         model_id = MODEL_NAMES[model_arch]
         model = OVModelForSpeechSeq2Seq.from_pretrained(model_id, export=True)
+        model.eval()
         processor = get_preprocessor(model_id)
         GenerationConfig.from_pretrained(model_id)
         pipe = pipeline(