admin/Project02
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

调整训练参数

Browse Source
This commit is contained in:
997146918 2025-08-11 10:29:12 +08:00
parent 5c27e14038
commit 66dade2f3f
1 changed files with 9 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
AITrain/train_npc_dialogue_lora.py
View File

@ -39,7 +39,7 @@ def process_func(example, tokenizer):
instruction = tokenizer( instruction = tokenizer(
f"<s><|im_start|>system\n{system_prompt}<|im_end|>\n" f"<s><|im_start|>system\n{system_prompt}<|im_end|>\n"
f"<|im_start|>user\n{instruction + user_input}<|im_end|>\n" f"<|im_start|>user\n{instruction + user_input}<|im_end|>\n"
f"<|im_start|>assistant\n", f"<|im_start|>assistant\n<think>\n\n</think>\n\n",
add_special_tokens=False add_special_tokens=False
) )
@ -142,12 +142,12 @@ def train_lora_model(model_path, data_path, output_dir):
model.config.use_cache = False # 关闭缓存以节省显存 model.config.use_cache = False # 关闭缓存以节省显存
# 5. 准备数据集 # 5. 准备数据集
train_dataset = prepare_dataset(data_path, tokenizer) train_preparedataset = prepare_dataset(data_path, tokenizer)
# 6. 配置训练参数 - 针对3080显卡优化 # 6. 配置训练参数 - 针对3080显卡优化
training_args = TrainingArguments( training_args = TrainingArguments(
output_dir=output_dir, output_dir=output_dir,
per_device_train_batch_size=1, # 减小batch size per_device_train_batch_size=2, # 减小batch size
gradient_accumulation_steps=4, # 增加梯度累积 gradient_accumulation_steps=4, # 增加梯度累积
logging_steps=10, logging_steps=10,
num_train_epochs=3, # 增加训练轮数以充分学习角色特征 num_train_epochs=3, # 增加训练轮数以充分学习角色特征
@ -161,6 +161,7 @@ def train_lora_model(model_path, data_path, output_dir):
dataloader_pin_memory=False, # 减少内存使用 dataloader_pin_memory=False, # 减少内存使用
remove_unused_columns=False, remove_unused_columns=False,
report_to="none", report_to="none",
#bf16=True,
#fp16=True, # 使用混合精度训练 #fp16=True, # 使用混合精度训练
save_total_limit=3, # 只保留最新的3个检查点 save_total_limit=3, # 只保留最新的3个检查点
) )
@ -175,7 +176,7 @@ def train_lora_model(model_path, data_path, output_dir):
trainer = Trainer( trainer = Trainer(
model=model, model=model,
args=training_args, args=training_args,
train_dataset=train_dataset, train_dataset=train_preparedataset,
data_collator=DataCollatorForSeq2Seq(tokenizer=tokenizer, padding=True), data_collator=DataCollatorForSeq2Seq(tokenizer=tokenizer, padding=True),
callbacks=[swanlab_callback] # 传入之前的swanlab_callback callbacks=[swanlab_callback] # 传入之前的swanlab_callback
) )
@ -270,13 +271,13 @@ def test_trained_model(model_path, lora_path):
def main(): def main():
# 配置路径 # 配置路径
model_path = '/mnt/g/Project02/AITrain/Qwen/Qwen3-8B-AWQ' # 基础模型路径 model_path = '/mnt/e/AI/Project02/AITrain/Qwen/Qwen3-8B-AWQ' # 基础模型路径
data_path = './npc_dialogue_dataset.json' # 训练数据路径 data_path = './npc_dialogue_dataset.json' # 训练数据路径
output_dir = './output/NPC_Dialogue_LoRA' # 输出目录 output_dir = './output/NPC_Dialogue_LoRA' # 输出目录
#####test # #####test
final_model_path = os.path.join(output_dir, "final_model") # final_model_path = os.path.join(output_dir, "final_model")
test_trained_model(model_path, final_model_path) # test_trained_model(model_path, final_model_path)
# 确保数据文件存在 # 确保数据文件存在
if not os.path.exists(data_path): if not os.path.exists(data_path):
print(f"数据文件不存在: {data_path}") print(f"数据文件不存在: {data_path}")