传入Trainer中训练的为何是model而不是delta_model呢？ #12
Description
lora_config = json.load(open("config/lora_config.json"))
lora_config["lora_r"] = sparse_args.lora_r
lora_config = LoraConfig.from_dict(lora_config)
**delta_model** = LoraModel.from_config(lora_config, backbone_model=model)
**delta_model**.freeze_module(set_state_dict = True)
**delta_mode**l.log(delta_ratio=True, trainable_ratio=True, visualization=False)
if training_args.train_sparse:
print("building sparse optimizer and scheduler")
from src.trainer import GATE_PARAM_NAME
valid_param_name = []
for n, p in **delta_model**.named_parameters():
print(f"Parameter name: {n}, requires_grad: {p.requires_grad}")
if GATE_PARAM_NAME in n:
valid_param_name.append(n)
print("valid param name:", valid_param_name)
sparse_optimizer = SparseAdamW(sparse_lambda=sparse_args.sparse_lambda_2, lambda_schedule=sparse_args.lambda_schedule, max_lambda=sparse_args.max_lambda, lambda_num=sparse_args.lambda_num, params=[p for n, p in model.named_parameters() if GATE_PARAM_NAME in n and p.requires_grad], lr=sparse_args.sparse_lr)
sparse_scheduler = get_linear_schedule_with_warmup(sparse_optimizer,
num_warmup_steps=int(training_args.num_train_epochs*(len(train_dataset) / training_args.train_batch_size)*training_args.warmup_ratio),
num_training_steps=int(training_args.num_train_epochs*(len(train_dataset) / training_args.train_batch_size)))
trainer = SparseTrainer(
model=**model**,
args=training_args,
train_dataset=train_dataset if training_args.do_train else None,
eval_dataset=eval_dataset if training_args.do_eval else None,
compute_metrics=compute_metrics,
tokenizer=tokenizer,
data_collator=data_collator,
optimizers = (optimizer, lr_scheduler),
sparse_lambda = sparse_args.sparse_lambda,
sparse_optimizer = (sparse_optimizer, sparse_scheduler)
)
我比较困惑为什么delta_model在后续代码中没有出现过了，按照我浅显的理解，在封装好了delta_model之后，在Trainer中进行微调的应该是delta_model呀。目前delta_model定义好之后好像没有起到什么实际作用。
我对opendelta的了解还不够深入，希望大家不吝赐教。