Update model_utils.py

model_utils.py

@@ -14,47 +14,12 @@
 # See the License for the specific language governing permissions and
 # limitations under the License.
 
-from typing import Dict
-
-import torch
-from transformers import AutoTokenizer, BitsAndBytesConfig, PreTrainedTokenizer
-
-from accelerate import Accelerator
-from huggingface_hub import list_repo_files
-from peft import LoraConfig, PeftConfig
+from transformers import AutoTokenizer, PreTrainedTokenizer
 
 from .configs import DataArguments, ModelArguments
 from .data import DEFAULT_CHAT_TEMPLATE
 
 
-def get_current_device() -> int:
-    """Get the current device. For GPU we return the local process index to enable multiple GPU training."""
-    return Accelerator().local_process_index if torch.cuda.is_available() else "cpu"
-
-
-def get_kbit_device_map() -> Dict[str, int] | None:
-    """Useful for running inference with quantized models by setting `device_map=get_peft_device_map()`"""
-    return {"": get_current_device()} if torch.cuda.is_available() else None
-
-
-def get_quantization_config(model_args) -> BitsAndBytesConfig | None:
-    if model_args.load_in_4bit:
-        quantization_config = BitsAndBytesConfig(
-            load_in_4bit=True,
-            bnb_4bit_compute_dtype=torch.float16,  # For consistency with model weights, we use the same value as `torch_dtype` which is float16 for PEFT models
-            bnb_4bit_quant_type=model_args.bnb_4bit_quant_type,
-            bnb_4bit_use_double_quant=model_args.use_bnb_nested_quant,
-        )
-    elif model_args.load_in_8bit:
-        quantization_config = BitsAndBytesConfig(
-            load_in_8bit=True,
-        )
-    else:
-        quantization_config = None
-
-    return quantization_config
-
-
 def get_tokenizer(model_args: ModelArguments, data_args: DataArguments) -> PreTrainedTokenizer:
     """Get the tokenizer for the model."""
     tokenizer = AutoTokenizer.from_pretrained(
@@ -76,26 +41,4 @@ def get_tokenizer(model_args: ModelArguments, data_args: DataArguments) -> PreTr
     elif tokenizer.chat_template is None:
         tokenizer.chat_template = DEFAULT_CHAT_TEMPLATE
 
-    return tokenizer
-
-
-def get_peft_config(model_args: ModelArguments) -> PeftConfig | None:
-    if model_args.use_peft is False:
-        return None
-
-    peft_config = LoraConfig(
-        r=model_args.lora_r,
-        lora_alpha=model_args.lora_alpha,
-        lora_dropout=model_args.lora_dropout,
-        bias="none",
-        task_type="CAUSAL_LM",
-        target_modules=model_args.lora_target_modules,
-        modules_to_save=model_args.lora_modules_to_save,
-    )
-
-    return peft_config
-
-
-def is_adapter_model(model_name_or_path: str, revision: str = "main") -> bool:
-    repo_files = list_repo_files(model_name_or_path, revision=revision)
-    return "adapter_model.safetensors" in repo_files or "adapter_model.bin" in repo_files
+    return tokenizer