Fix for text-generation, AttributeError: 'GenerationConfig' object ha…

…s no attribute 'use_fused_rope' (#1823)

Co-authored-by: regisss <15324346+regisss@users.noreply.github.com>

examples/text-generation/run_lm_eval.py

@@ -27,6 +27,7 @@
 import psutil
 import torch
 import torch.nn.functional as F
+import transformers
 from lm_eval import evaluator, utils
 from lm_eval.models.huggingface import HFLM, TemplateLM
 
@@ -36,6 +37,7 @@
 from transformers.generation import GenerationConfig
 from utils import finalize_quantization, initialize_model, save_model
 
+from optimum.habana.transformers.generation import GaudiGenerationConfig
 from optimum.habana.utils import get_hpu_memory_stats
 
 
@@ -248,6 +250,7 @@ def get_model_dtype(model) -> str:
 def main() -> None:
     # Modified based on cli_evaluate function in https://github.com/EleutherAI/lm-evaluation-harness/blob/v0.4.7/lm_eval/__main__.py/#L268
     args = setup_lm_eval_parser()
+    transformers.GenerationConfig = GaudiGenerationConfig
     model, _, tokenizer, generation_config = initialize_model(args, logger)
     if args.trust_remote_code:
         # trust_remote_code fix was introduced in lm_eval 0.4.3

optimum/habana/transformers/modeling_utils.py

@@ -376,7 +376,6 @@ def adapt_transformers_to_gaudi():
         GaudiGenerationMixin._prepare_cache_for_generation
     )
     transformers.generation.GenerationConfig = GaudiGenerationConfig
-    transformers.GenerationConfig = GaudiGenerationConfig
     transformers.generation.configuration_utils.GenerationConfig = GaudiGenerationConfig
     transformers.modeling_utils.GenerationConfig = GaudiGenerationConfig
     transformers.generation.MaxLengthCriteria.__call__ = gaudi_MaxLengthCriteria_call