import os
from unsloth import FastLanguageModel
import torch
from trl import SFTTrainer
from transformers import TrainingArguments
from datasets import load_dataset

# 加载模型
max_seq_length = 2048
dtype = None
load_in_4bit = True
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "/home/zk/ai/base_model/llama-3-8b-bnb-4bit",
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
)

# 准备训练数据
alpaca_prompt = """Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.
### Instruction：
{}
### Input:
{}
### Response:
{}"""

EOS_TOKEN = tokenizer.eos_token  # 必须添加 EOS_TOKEN
def formatting_prompts_func(examples):
    instructions = examples["instruction"]
    inputs = examples["input"]
    outputs = examples["output"]
    texts = []
    for instruction, input, output in zip(instructions, inputs, outputs):
        # 必须添加EOS_TOKEN，否则无限生成
        text = alpaca_prompt.format(instruction, input, output) + EOS_TOKEN
        texts.append(text)
    return {"text": texts}
pass

# 本地数据集
dataset = load_dataset("json", data_files = "/home/zk/ai/dataset/caishui_2011_100hao.json", split="train")
dataset = dataset.map(formatting_prompts_func, batched = True)

# 设置训练参数
model = FastLanguageModel.get_peft_model(
    model,
    r = 16,
    target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
                        "gate_proj", "up_proj", "down_proj", ],
    lora_alpha = 16,
    lora_dropout = 0,
    bias = "none",
    use_gradient_checkpointing = True,
    random_state = 3407,
    max_seq_length = max_seq_length,
    use_rslora = False,
    loftq_config = None,
)

trainer = SFTTrainer(
    model = model,
    train_dataset = dataset,
    dataset_text_field = "text",
    max_seq_length = max_seq_length,
    tokenizer = tokenizer,
    args = TrainingArguments(
        per_device_train_batch_size = 1,
        gradient_accumulation_steps = 4,
        warmup_steps = 2,
        max_steps = 20,
        fp16 = not torch.cuda.is_bf16_supported(),
        bf16 = torch.cuda.is_bf16_supported(),
        logging_steps = 1,
        output_dir = "outputs",
        optim = "adamw_8bit",
        weight_decay = 0.01,
        lr_scheduler_type = "linear",
        seed = 3407,
        learning_rate = 2e-5,
    ),
)

# 开始训练
trainer.train()

# 保存微调模型
model.save_pretrained("lora_model")

# 选项：保存为16位hf模型
save_16bit = input("是否保存为16位hf模型？(y/n): ")
if save_16bit.lower() == "y":
    model.save_pretrained_merged("outputs", tokenizer, save_method="merged_16bit")

# 选项：保存为gguf模型
save_gguf = input("是否保存为gguf模型？(y/n): ")
if save_gguf.lower() == "y":
    os.system("python /home/zk/ai/llama.cpp/convert_hf_to_gguf.py --outfile /home/zk/ai/gguf_model/lm38b_tax_jzjt.gguf /home/zk/ai/outputs")

# 选项：量化为4位gguf模型
quantize_4bit = input("是否量化为4位gguf模型？(y/n): ")
if quantize_4bit.lower() == "y":
    os.system("/home/zk/ai/llama.cpp/build/bin/llama-quantize /home/zk/ai/gguf_model/QWQ_tax_jzjt.gguf /home/zk/ai/gguf_model/lm38b_tax_jzjt-Q4_K_M.gguf Q4_K_M")