fix a bug of fine-tuning on HPU (#239)

llm_on_ray/finetune/finetune.py

@@ -42,7 +42,16 @@
 from importlib import util
 
 
-def set_seed(config):
+def adapt_transformers_to_device(config: Dict):
+    device = config["Training"]["device"]
+    if device in ["hpu"]:
+        from optimum.habana.transformers.modeling_utils import adapt_transformers_to_gaudi
+
+        # adapt transformers to gaudi
+        adapt_transformers_to_gaudi()
+
+
+def set_seed(config: Dict):
     seed = config["Training"].get("seed", None)
     if seed is None:
         return
@@ -57,7 +66,7 @@ def set_seed(config):
         _set_seed(seed)
 
 
-def convert_to_training_args(cls, config):
+def convert_to_training_args(cls, config: Dict):
     device = config["Training"]["device"]
     accelerate_mode = config["Training"]["accelerate_mode"]
     save_strategy = config["General"]["save_strategy"]
@@ -312,11 +321,22 @@ def get_trainer(config: Dict, model, tokenizer, tokenized_dataset, data_collator
     elif device in ["hpu"]:
         from optimum.habana.transformers import GaudiTrainer
         from optimum.habana.transformers import GaudiTrainingArguments
+        from optimum.habana import GaudiConfig
+
+        # If gaudi_config_name is provided, load gaudi_config from huggingface model hub(https://huggingface.co/Habana), otherwise use default gaudi_config
+        gaudi_config_name = config["General"].get("gaudi_config_name", None)
+        if gaudi_config_name is not None:
+            gaudi_config = GaudiConfig.from_pretrained(gaudi_config_name)
+        else:
+            gaudi_config = GaudiConfig()
+            gaudi_config.use_fused_adam = True
+            gaudi_config.use_fused_clip_norm = True
 
         training_args = convert_to_training_args(GaudiTrainingArguments, config)
         trainer = GaudiTrainer(
             model=model,
             args=training_args,
+            gaudi_config=gaudi_config,
             train_dataset=tokenized_dataset["train"],
             eval_dataset=tokenized_dataset["validation"]
             if tokenized_dataset.get("validation") is not None
@@ -331,6 +351,8 @@ def get_trainer(config: Dict, model, tokenizer, tokenized_dataset, data_collator
 def train_func(config: Dict[str, Any]):
     os.chdir(config["cwd"])
 
+    adapt_transformers_to_device(config)
+
     set_seed(config)
 
     tokenizer = load_tokenizer(config)