p04798526
/
LLaMA-Factory-310P3
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

fix gptq training

This commit is contained in:
hiyouga 2023-12-02 00:27:15 +08:00
parent a973ce6e89
commit f57445c7a0
3 changed files with 12 additions and 10 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

4
src/llmtuner/extras/misc.py
View File

@ -72,7 +72,9 @@ def get_current_device() -> str:
import accelerate import accelerate
if accelerate.utils.is_xpu_available(): if accelerate.utils.is_xpu_available():
return "xpu:{}".format(os.environ.get("LOCAL_RANK", "0")) return "xpu:{}".format(os.environ.get("LOCAL_RANK", "0"))
elif accelerate.utils.is_npu_available() or torch.cuda.is_available(): elif accelerate.utils.is_npu_available():
return "npu:{}".format(os.environ.get("LOCAL_RANK", "0"))
elif torch.cuda.is_available():
return "cuda:{}".format(os.environ.get("LOCAL_RANK", "0")) return "cuda:{}".format(os.environ.get("LOCAL_RANK", "0"))
else: else:
return "cpu" return "cpu"

2
src/llmtuner/model/adapter.py
View File

@ -87,7 +87,7 @@ def init_adapter(
if is_trainable and checkpoint_to_resume is None: # create new lora weights while training if is_trainable and checkpoint_to_resume is None: # create new lora weights while training
if len(finetuning_args.lora_target) == 1 and finetuning_args.lora_target[0] == "all": if len(finetuning_args.lora_target) == 1 and finetuning_args.lora_target[0] == "all":
target_modules = find_all_linear_modules(model, model_args.quantization_bit) target_modules = find_all_linear_modules(model)
else: else:
target_modules = finetuning_args.lora_target target_modules = finetuning_args.lora_target

16
src/llmtuner/model/utils.py
View File

@ -42,18 +42,18 @@ def dispatch_model(model: "PreTrainedModel") -> "PreTrainedModel":
return model.cuda() return model.cuda()
def find_all_linear_modules( def find_all_linear_modules(model: "PreTrainedModel") -> List[str]:
model: "PreTrainedModel",
quantization_bit: Optional[int] = None
) -> List[str]:
r""" r"""
Finds all available modules to apply lora. Finds all available modules to apply lora.
""" """
if quantization_bit is not None: quantization_method = getattr(model, "quantization_method", None)
import bitsandbytes as bnb if quantization_method is None:
linear_cls = bnb.nn.Linear4bit if quantization_bit == 4 else bnb.nn.Linear8bitLt
else:
linear_cls = torch.nn.Linear linear_cls = torch.nn.Linear
elif quantization_method == "bitsandbytes":
import bitsandbytes as bnb
linear_cls = bnb.nn.Linear4bit if getattr(model, "is_loaded_in_4bit", False) else bnb.nn.Linear8bitLt
else:
raise ValueError("Finding linear modules for {} models is not supported.".format(quantization_method))
output_layer_names = ["lm_head"] output_layer_names = ["lm_head"]
if model.config.model_type == "chatglm": if model.config.model_type == "chatglm":