Fixed import issue for seq2seq examples (intel#191)

* Fixed import issue for seq2seq examples

Signed-off-by: Cheng, Penghui <[email protected]>

Author: PenghuiCheng

examples/huggingface/pytorch/language-modeling/inference/run_clm_no_trainer.py

@@ -83,7 +83,7 @@
 if args.accuracy:
     user_model.eval()
     def eval_func(user_model):
-        from intel_extension_for_transformers.evaluation.lm_eval import evaluate
+        from intel_extension_for_transformers.llm.evaluation.lm_eval import evaluate
         results = evaluate(
             model="hf-causal",
             model_args='pretrained='+args.model+',tokenizer='+args.model+',dtype=float32',

examples/huggingface/pytorch/language-modeling/quantization/run_clm_no_trainer.py

@@ -339,7 +339,7 @@ def eval_func(model):
 
 if args.accuracy:
     user_model.eval()
-    from intel_extension_for_transformers.evaluation.lm_eval import evaluate
+    from intel_extension_for_transformers.llm.evaluation.lm_eval import evaluate
     results = evaluate(
         model="hf-causal",
         model_args='pretrained='+args.model+',tokenizer='+args.model+',dtype=float32',

examples/huggingface/pytorch/text2text-generation/run_seq2seq_generation.py

@@ -36,7 +36,7 @@
     AutoTokenizer,
 )
 
-from intel_extension_for_transformers.transformers.modeling import INCModelForSeq2SeqLM
+from intel_extension_for_transformers.transformers.modeling.modeling_seq2seq import INCModelForSeq2SeqLM
 
 
 prompt_texts = ["Translate to German: My name is Arthur",
@@ -431,7 +431,7 @@ def decoder_calib_func(prepared_model):
         print("Throughput: {} tokens/sec".format(throughput))
 
     if args.accuracy:
-        from intel_extension_for_transformers.evaluation.hf_eval import summarization_evaluate
+        from intel_extension_for_transformers.llm.evaluation.hf_eval import summarization_evaluate
         results = summarization_evaluate(
            model=model,
            tokenizer_name=args.model_name_or_path,

examples/huggingface/pytorch/text2text-generation/run_seq2seq_generation_bart.py

@@ -36,7 +36,7 @@
     AutoTokenizer,
 )
 
-from intel_extension_for_transformers.transformers.modeling import INCModelForSeq2SeqLM
+from intel_extension_for_transformers.transformers.modeling.modeling_seq2seq import INCModelForSeq2SeqLM
 
 
 prompt_texts = ["Translate to German: My name is Arthur",
@@ -430,7 +430,7 @@ def decoder_calib_func(prepared_model):
         print("Throughput: {} tokens/sec".format(throughput))
 
     if args.accuracy:
-        from intel_extension_for_transformers.evaluation.hf_eval import summarization_evaluate
+        from intel_extension_for_transformers.llm.evaluation.hf_eval import summarization_evaluate
         results = summarization_evaluate(
            model=model,
            tokenizer_name=args.model_name_or_path,

workflows/chatbot/fine_tuning/instruction_tuning_pipeline/finetune_clm.py

@@ -683,8 +683,7 @@ def concatenate_data(dataset, max_seq_length):
 
         if finetune_args.do_lm_eval and finetune_args.task != "summarization":
             unwrapped_model.eval()
-            from intel_extension_for_transformers.evaluation.lm_eval import evaluate
-
+            from intel_extension_for_transformers.llm.evaluation.lm_eval import evaluate
             with training_args.main_process_first(desc="lm_eval"):
                 if is_main_process(training_args.local_rank):
                     with torch.no_grad():