File size: 3,316 Bytes
88aba71 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 | {
"version": "0.2.21",
"common_args": {
"model_name_or_path": "./Qwen2.5-7B-Instruct",
"adapter_name_or_path": "./model_output", //同时做为train_sft_args的output_dir
"template": "qwen",
"default_system": "请你扮演一名人类，不要说自己是人工智能",
"finetuning_type": "lora",
"trust_remote_code": true
},
"cli_args": {
"full_log": false
},
"make_dataset_args": {
//数据处理配置
"include_type": [
"text",
// "image"
],
"blocked_words": [ // 禁用词
"例如 姓名",
"例如 密码",
"//....."
],
"single_combine_strategy": "time_window", // 单人组成单句策略
"qa_match_strategy": "time_window", // 组成qa策略
"single_combine_time_window": 2, // 单人组成单句时间窗口（分钟）,
"qa_match_time_window": 5, // 组成qa时间窗口（分钟）,
"combine_msg_max_length": 256, // 组合后消息最大长度 配合cutoff_len 使用
"prompt_with_history": false, // 是否在prompt中包含历史对话
"clean_dataset": {
"enable_clean": true,
"clean_strategy": "llm",
"llm": {
"accept_score": 2, //可以接受的llm打分阈值,1分最差，5分最好,低于此分数的数据不会用于训练
}
},
"online_llm_clear": false,
"base_url": "https://xxx/v1",
"llm_api_key": "xxxxx",
"model_name": "xxx", //建议使用参数较大的模型，例如DeepSeek-V3
"clean_batch_size": 10
},
"train_pt_args": {
//预训练微调配置
"stage": "pt",
"dataset": "wechat-pt",
"dataset_dir": "./dataset/res_csv/pt",
"lora_target": "q_proj,v_proj",
"lora_rank": 2,
"lora_dropout": 0.1,
"output_dir": "model_output",
"overwrite_cache": true,
"per_device_train_batch_size": 1,
"gradient_accumulation_steps": 1,
"lr_scheduler_type": "cosine",
"logging_steps": 10,
"save_steps": 1000,
"learning_rate": 0.001,
"num_train_epochs": 30,
"plot_loss": true,
"fp16": true
},
"train_sft_args": {
//微调配置
"stage": "sft",
"dataset": "wechat-sft",
"dataset_dir": "./dataset/res_csv/sft",
"use_fast_tokenizer": true,
"lora_target": "q_proj,v_proj",
"lora_rank": 4,
"lora_dropout": 0.3,
"weight_decay": 0.1,
"overwrite_cache": true,
"per_device_train_batch_size": 8,
"gradient_accumulation_steps": 4,
"lr_scheduler_type": "cosine",
"cutoff_len": 256,
"logging_steps": 10,
"save_steps": 100,
"learning_rate": 1e-4,
"warmup_ratio": 0.1,
"num_train_epochs": 2,
"plot_loss": true,
"fp16": true,
"flash_attn": "fa2",
// "deepspeed": "ds_config.json" //多卡训练
},
"infer_args": {
"repetition_penalty": 1.2,
"temperature": 0.5,
"max_length": 50,
"top_p": 0.65
}
} |