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

fix #376

This commit is contained in:
hiyouga 2023-08-07 13:58:59 +08:00
parent 220175ab24
commit 081345baca
1 changed files with 6 additions and 6 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

12
src/llmtuner/tuner/core/loader.py
View File

@ -73,7 +73,12 @@ def load_model_and_tokenizer(
if tokenizer.pad_token_id is None: # add pad token
tokenizer.pad_token = tokenizer.eos_token
config = AutoConfig.from_pretrained(model_args.model_name_or_path, **config_kwargs)
if model_args.checkpoint_dir is not None and finetuning_args.finetuning_type == "full":
model_to_load = model_args.checkpoint_dir[0]
else:
model_to_load = model_args.model_name_or_path
config = AutoConfig.from_pretrained(model_to_load, **config_kwargs)
is_mergeable = True
# Quantization configurations (using bitsandbytes library).
@ -100,11 +105,6 @@ def load_model_and_tokenizer(
config_kwargs["device_map"] = {"": int(os.environ.get("LOCAL_RANK", "0"))}
logger.info("Quantizing model to {} bit.".format(model_args.quantization_bit))
if model_args.checkpoint_dir is not None and finetuning_args.finetuning_type == "full":
model_to_load = model_args.checkpoint_dir[0]
else:
model_to_load = model_args.model_name_or_path
# Load and prepare pretrained models (without valuehead).
model = AutoModelForCausalLM.from_pretrained(
model_to_load,