update example dataloader for transformers 4.31.x (#1125)

Signed-off-by: Cheng, Zixuan <zixuan.cheng@intel.com>

examples/pytorch/nlp/huggingface_models/language-modeling/quantization/ptq_static/fx/run_clm.py

@@ -536,6 +536,22 @@ def compute_metrics(eval_preds):
         else None,
     )
 
+    eval_dataloader = trainer.get_eval_dataloader()
+    # transformer issue #1
+    # for transformers 4.31.0: accelerate dataloader
+    # *** ValueError: batch_size attribute should not be set 
+    # after DataLoaderShard is initialized
+    if eval_dataloader.batch_size is None:
+        def _build_inc_dataloader(dataloader):
+            class INCDataLoader:
+                __iter__ = dataloader.__iter__
+                def __init__(self) -> None:
+                    self.dataloader = dataloader
+                    self.batch_size = dataloader.total_batch_size
+            return INCDataLoader()
+        eval_dataloader = _build_inc_dataloader(eval_dataloader)
+    batch_size = eval_dataloader.batch_size
+
     # Tune
     def eval_func_for_nc(model_tuned):
         trainer.model = model_tuned
@@ -564,7 +580,7 @@ def eval_func_for_nc(model_tuned):
         conf = PostTrainingQuantConfig(accuracy_criterion=accuracy_criterion)
         q_model = quantization.fit(model,
                                    conf,
-                                   calib_dataloader=trainer.get_eval_dataloader(),
+                                   calib_dataloader=eval_dataloader,
                                    eval_func=eval_func_for_nc)
         q_model.save(training_args.output_dir)
         exit(0)
@@ -582,7 +598,7 @@ def eval_func_for_nc(model_tuned):
             from neural_compressor.config import BenchmarkConfig
             from neural_compressor import benchmark
             b_conf = BenchmarkConfig(warmup=5, iteration=100, cores_per_instance=4, num_of_instance=1)
-            benchmark.fit(new_model, b_conf, b_dataloader=trainer.get_eval_dataloader())
+            benchmark.fit(new_model, b_conf, b_dataloader=eval_dataloader)
         else:
             eval_func_for_nc(new_model)
 

examples/pytorch/nlp/huggingface_models/language-modeling/quantization/ptq_weight_only/run_clm.py

@@ -552,6 +552,22 @@ def compute_metrics(eval_preds):
         else None,
     )
 
+    eval_dataloader = trainer.get_eval_dataloader()
+    # transformer issue #1
+    # for transformers 4.31.0: accelerate dataloader
+    # *** ValueError: batch_size attribute should not be set 
+    # after DataLoaderShard is initialized
+    if eval_dataloader.batch_size is None:
+        def _build_inc_dataloader(dataloader):
+            class INCDataLoader:
+                __iter__ = dataloader.__iter__
+                def __init__(self) -> None:
+                    self.dataloader = dataloader
+                    self.batch_size = dataloader.total_batch_size
+            return INCDataLoader()
+        eval_dataloader = _build_inc_dataloader(eval_dataloader)
+    batch_size = eval_dataloader.batch_size
+
     # Tune
     def eval_func_for_nc(model_tuned):
         trainer.model = model_tuned
@@ -592,7 +608,7 @@ def eval_func_for_nc(model_tuned):
                                     op_type_dict=op_type_dict)
         q_model = quantization.fit(model,
                                    conf,
-                                   calib_dataloader=trainer.get_eval_dataloader(),
+                                   calib_dataloader=eval_dataloader,
                                    eval_func=eval_func_for_nc)
         q_model.save(training_args.output_dir)
         exit(0)
@@ -610,7 +626,7 @@ def eval_func_for_nc(model_tuned):
             from neural_compressor.config import BenchmarkConfig
             from neural_compressor import benchmark
             b_conf = BenchmarkConfig(warmup=5, iteration=100, cores_per_instance=4, num_of_instance=1)
-            benchmark.fit(new_model, b_conf, b_dataloader=trainer.get_eval_dataloader())
+            benchmark.fit(new_model, b_conf, b_dataloader=eval_dataloader)
         else:
             eval_func_for_nc(new_model)
 

examples/pytorch/nlp/huggingface_models/question-answering/quantization/ptq_static/fx/run_qa.py

@@ -605,6 +605,19 @@ def compute_metrics(p: EvalPrediction):
     )
 
     eval_dataloader = trainer.get_eval_dataloader()
+    # transformer issue #1
+    # for transformers 4.31.0: accelerate dataloader
+    # *** ValueError: batch_size attribute should not be set 
+    # after DataLoaderShard is initialized
+    if eval_dataloader.batch_size is None:
+        def _build_inc_dataloader(dataloader):
+            class INCDataLoader:
+                __iter__ = dataloader.__iter__
+                def __init__(self) -> None:
+                    self.dataloader = dataloader
+                    self.batch_size = dataloader.total_batch_size
+            return INCDataLoader()
+        eval_dataloader = _build_inc_dataloader(eval_dataloader)
     batch_size = eval_dataloader.batch_size
     metric_name = "eval_f1"
 

examples/pytorch/nlp/huggingface_models/question-answering/quantization/ptq_static/ipex/run_qa.py

@@ -612,6 +612,19 @@ def compute_metrics(p: EvalPrediction):
     )
 
     eval_dataloader = trainer.get_eval_dataloader()
+    # transformer issue #1
+    # for transformers 4.31.0: accelerate dataloader
+    # *** ValueError: batch_size attribute should not be set 
+    # after DataLoaderShard is initialized
+    if eval_dataloader.batch_size is None:
+        def _build_inc_dataloader(dataloader):
+            class INCDataLoader:
+                __iter__ = dataloader.__iter__
+                def __init__(self) -> None:
+                    self.dataloader = dataloader
+                    self.batch_size = dataloader.total_batch_size
+            return INCDataLoader()
+        eval_dataloader = _build_inc_dataloader(eval_dataloader)
     batch_size = eval_dataloader.batch_size
     metric_name = "eval_f1"
 

examples/pytorch/nlp/huggingface_models/summarization/quantization/ptq_dynamic/fx/run_summarization.py

@@ -588,6 +588,22 @@ def compute_metrics(eval_preds):
     )
     num_beams = data_args.num_beams if data_args.num_beams is not None else training_args.generation_num_beams
 
+    eval_dataloader = trainer.get_eval_dataloader()
+    # transformer issue #1
+    # for transformers 4.31.0: accelerate dataloader
+    # *** ValueError: batch_size attribute should not be set 
+    # after DataLoaderShard is initialized
+    if eval_dataloader.batch_size is None:
+        def _build_inc_dataloader(dataloader):
+            class INCDataLoader:
+                __iter__ = dataloader.__iter__
+                def __init__(self) -> None:
+                    self.dataloader = dataloader
+                    self.batch_size = dataloader.total_batch_size
+            return INCDataLoader()
+        eval_dataloader = _build_inc_dataloader(eval_dataloader)
+    batch_size = eval_dataloader.batch_size
+
     def eval_func_for_nc(model):
         trainer.model = model
         results = trainer.evaluate(
@@ -616,7 +632,7 @@ def eval_func_for_nc(model):
         conf = PostTrainingQuantConfig(approach="dynamic")
         q_model = quantization.fit(model,
                                    conf,
-                                   calib_dataloader=trainer.get_eval_dataloader(),
+                                   calib_dataloader=eval_dataloader,
                                    eval_func=eval_func_for_nc)
         q_model.save(training_args.output_dir)
         exit(0)
@@ -634,7 +650,7 @@ def eval_func_for_nc(model):
             from neural_compressor.config import BenchmarkConfig
             from neural_compressor import benchmark
             b_conf = BenchmarkConfig(warmup=5, iteration=100, cores_per_instance=4, num_of_instance=1)
-            benchmark.fit(new_model, b_conf, b_dataloader=trainer.get_eval_dataloader())
+            benchmark.fit(new_model, b_conf, b_dataloader=eval_dataloader)
         else:
             eval_func_for_nc(new_model)
 

examples/pytorch/nlp/huggingface_models/text-classification/export/fx/run_glue.py

@@ -494,6 +494,19 @@ def compute_metrics(p: EvalPrediction):
     )
 
     eval_dataloader = trainer.get_eval_dataloader()
+    # transformer issue #1
+    # for transformers 4.31.0: accelerate dataloader
+    # *** ValueError: batch_size attribute should not be set 
+    # after DataLoaderShard is initialized
+    if eval_dataloader.batch_size is None:
+        def _build_inc_dataloader(dataloader):
+            class INCDataLoader:
+                __iter__ = dataloader.__iter__
+                def __init__(self) -> None:
+                    self.dataloader = dataloader
+                    self.batch_size = dataloader.total_batch_size
+            return INCDataLoader()
+        eval_dataloader = _build_inc_dataloader(eval_dataloader)
     batch_size = eval_dataloader.batch_size
 
     def take_eval_steps(model, trainer, save_metrics=False):

examples/pytorch/nlp/huggingface_models/text-classification/quantization/ptq_dynamic/fx/run_glue.py

@@ -479,6 +479,19 @@ def compute_metrics(p: EvalPrediction):
     )
 
     eval_dataloader = trainer.get_eval_dataloader()
+    # transformer issue #1
+    # for transformers 4.31.0: accelerate dataloader
+    # *** ValueError: batch_size attribute should not be set 
+    # after DataLoaderShard is initialized
+    if eval_dataloader.batch_size is None:
+        def _build_inc_dataloader(dataloader):
+            class INCDataLoader:
+                __iter__ = dataloader.__iter__
+                def __init__(self) -> None:
+                    self.dataloader = dataloader
+                    self.batch_size = dataloader.total_batch_size
+            return INCDataLoader()
+        eval_dataloader = _build_inc_dataloader(eval_dataloader)
     batch_size = eval_dataloader.batch_size
 
     def take_eval_steps(model, trainer, save_metrics=False):

examples/pytorch/nlp/huggingface_models/text-classification/quantization/ptq_static/fx/run_glue.py

@@ -479,6 +479,19 @@ def compute_metrics(p: EvalPrediction):
     )
 
     eval_dataloader = trainer.get_eval_dataloader()
+    # transformer issue #1
+    # for transformers 4.31.0: accelerate dataloader
+    # *** ValueError: batch_size attribute should not be set 
+    # after DataLoaderShard is initialized
+    if eval_dataloader.batch_size is None:
+        def _build_inc_dataloader(dataloader):
+            class INCDataLoader:
+                __iter__ = dataloader.__iter__
+                def __init__(self) -> None:
+                    self.dataloader = dataloader
+                    self.batch_size = dataloader.total_batch_size
+            return INCDataLoader()
+        eval_dataloader = _build_inc_dataloader(eval_dataloader)
     batch_size = eval_dataloader.batch_size
 
     def take_eval_steps(model, trainer, save_metrics=False):

examples/pytorch/nlp/huggingface_models/text-classification/quantization/qat/fx/run_glue.py

@@ -506,6 +506,19 @@ def compute_metrics(p: EvalPrediction):
                                                             early_stopping_threshold))
 
     eval_dataloader = trainer.get_eval_dataloader()
+    # transformer issue #1
+    # for transformers 4.31.0: accelerate dataloader
+    # *** ValueError: batch_size attribute should not be set 
+    # after DataLoaderShard is initialized
+    if eval_dataloader.batch_size is None:
+        def _build_inc_dataloader(dataloader):
+            class INCDataLoader:
+                __iter__ = dataloader.__iter__
+                def __init__(self) -> None:
+                    self.dataloader = dataloader
+                    self.batch_size = dataloader.total_batch_size
+            return INCDataLoader()
+        eval_dataloader = _build_inc_dataloader(eval_dataloader)
     batch_size = eval_dataloader.batch_size
 
     def eval_func(model):

examples/pytorch/nlp/huggingface_models/translation/quantization/ptq_dynamic/fx/run_translation.py

@@ -543,6 +543,22 @@ def compute_metrics(eval_preds):
         compute_metrics=compute_metrics if training_args.predict_with_generate else None
     )
 
+    eval_dataloader = trainer.get_eval_dataloader()
+    # transformer issue #1
+    # for transformers 4.31.0: accelerate dataloader
+    # *** ValueError: batch_size attribute should not be set 
+    # after DataLoaderShard is initialized
+    if eval_dataloader.batch_size is None:
+        def _build_inc_dataloader(dataloader):
+            class INCDataLoader:
+                __iter__ = dataloader.__iter__
+                def __init__(self) -> None:
+                    self.dataloader = dataloader
+                    self.batch_size = dataloader.total_batch_size
+            return INCDataLoader()
+        eval_dataloader = _build_inc_dataloader(eval_dataloader)
+    batch_size = eval_dataloader.batch_size
+
     results = {}
     max_length = (
         training_args.generation_max_length
@@ -576,7 +592,7 @@ def eval_func_for_nc(model):
         conf = PostTrainingQuantConfig(approach="dynamic")
         q_model = quantization.fit(model,
                                    conf,
-                                   calib_dataloader=trainer.get_eval_dataloader(),
+                                   calib_dataloader=eval_dataloader,
                                    eval_func=eval_func_for_nc)
         q_model.save(training_args.output_dir)
         exit(0)
@@ -595,7 +611,7 @@ def eval_func_for_nc(model):
             from neural_compressor.config import BenchmarkConfig
             from neural_compressor import benchmark
             b_conf = BenchmarkConfig(warmup=5, iteration=100, cores_per_instance=4, num_of_instance=1)
-            benchmark.fit(new_model, b_conf, b_dataloader=trainer.get_eval_dataloader())
+            benchmark.fit(new_model, b_conf, b_dataloader=eval_dataloader)
         else:
             eval_func_for_nc(new_model)
         exit(0)