embed-bge-m3/FlagEmbedding/research/LLARA/finetune/load_model.py

65 lines
2.6 KiB
Python

import sys
import torch
from transformers import AutoConfig, AutoTokenizer, AutoModelForCausalLM, AutoModel
from peft import LoraConfig, TaskType, get_peft_model, PeftModel
def get_model(model_args):
# if model_args.use_flash_attn:
# from llama2_flash_attn_monkey_patch import replace_llama_attn_with_flash_attn
# replace_llama_attn_with_flash_attn()
if model_args.config_name:
config = AutoConfig.from_pretrained(model_args.config_name,
token=model_args.token,
cache_dir=model_args.cache_dir,
)
elif model_args.model_name_or_path:
config = AutoConfig.from_pretrained(model_args.model_name_or_path,
token=model_args.token,
cache_dir=model_args.cache_dir,
)
else:
raise ValueError(
"You are instantiating a new config instance from scratch. This is not supported by this script."
)
config.use_cache = False
if model_args.model_name_or_path:
model = AutoModelForCausalLM.from_pretrained(
model_args.model_name_or_path,
# load_in_8bit=True,
# torch_dtype=torch.bfloat16,
use_flash_attention_2=True if model_args.use_flash_attn else False,
token=model_args.token,
cache_dir=model_args.cache_dir,
from_tf=bool(".ckpt" in model_args.model_name_or_path),
config=config,
# low_cpu_mem_usage=model_args.low_cpu_mem_usage,
# device_map="auto",
)
else:
print("Training new model from scratch")
model = model_args.from_config(config)
if model_args.from_peft is not None:
model = PeftModel.from_pretrained(model, model_args.from_peft, is_trainable=True)
model.print_trainable_parameters()
else:
if model_args.use_lora:
peft_config = LoraConfig(
task_type=TaskType.FEATURE_EXTRACTION,
inference_mode=False,
r=model_args.lora_rank,
target_modules=model_args.target_modules,
lora_alpha=model_args.lora_alpha,
lora_dropout=model_args.lora_dropout
)
model = get_peft_model(model, peft_config)
# print(model.model.layers[0].self_attn.q_proj.weight.dtype)
# print(model.model.layers[0].self_attn.q_proj.lora_A.default.weight.dtype)
# sys.exit(0)
model.print_trainable_parameters()
return model