From 871f7de3d06264060193093bc9cb06909cf43eab Mon Sep 17 00:00:00 2001 From: hiyouga Date: Mon, 7 Aug 2023 13:58:59 +0800 Subject: [PATCH] fix #376 Former-commit-id: 081345baca263b5f0a6e936e71605e7cb127b3cd --- src/llmtuner/tuner/core/loader.py | 12 ++++++------ 1 file changed, 6 insertions(+), 6 deletions(-) diff --git a/src/llmtuner/tuner/core/loader.py b/src/llmtuner/tuner/core/loader.py index 5610deef..ec19e962 100644 --- a/src/llmtuner/tuner/core/loader.py +++ b/src/llmtuner/tuner/core/loader.py @@ -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,