import pandas as pd
from datasets import Dataset
from openmind import AutoTokenizer, AutoModelForCausalLM, TrainingArguments, Trainer
import torch
from peft import LoraConfig, TaskType, get_peft_model
from transformers import DataCollatorForSeq2Seq
from torch.utils.data import DataLoader
import time
import wandb
df = pd.read_json('ruozhiba_qa.json')
ds = Dataset.from_pandas(df)
print(len(ds))

model_path = 'deepseek-coder-33b-base'

tokenizer = AutoTokenizer.from_pretrained(model_path, use_fast=False, trust_remote_code=True)

def process_func(example):
    MAX_LENGTH = 2048    
    input_ids, attention_mask, labels = [], [], []
    instruction = tokenizer(f"<|im_start|>system\n你是一个人工智能助手，请你根据要求回答问题。<|im_end|>\n<|im_start|>user\n{example['instruction'] + example['input']}<|im_end|>\n<|im_start|>assistant\n", add_special_tokens=False)  # 
    response = tokenizer(f"{example['output']}", add_special_tokens=False)
    input_ids = instruction["input_ids"] + response["input_ids"] + [tokenizer.pad_token_id]
    attention_mask = instruction["attention_mask"] + response["attention_mask"] + [1]  # 因为eos token咱们也是要关注的所以 补充为1
    labels = [-100] * len(instruction["input_ids"]) + response["input_ids"] + [tokenizer.pad_token_id]  
    if len(input_ids) > MAX_LENGTH:  # 做一个截断
        input_ids = input_ids[:MAX_LENGTH]
        attention_mask = attention_mask[:MAX_LENGTH]
        labels = labels[:MAX_LENGTH]
    return {
        "input_ids": input_ids,
        "attention_mask": attention_mask,
        "labels": labels
    }
tokenized_id = ds.map(process_func, remove_columns=ds.column_names)
## 加载训练模型
model = AutoModelForCausalLM.from_pretrained(model_path, device_map="auto",torch_dtype=torch.bfloat16)

config = LoraConfig(
    task_type=TaskType.CAUSAL_LM, 
    target_modules=["v_proj","k_proj","gate_proj","q_proj"],
    inference_mode=False, # 训练模式
    r=8, # Lora 秩
    lora_alpha=16, # Lora alaph，具体作用参见 Lora 原理
    lora_dropout=0.1# Dropout 比例
)

model = get_peft_model(model, config)

model.enable_input_require_grads()
lora_path = "./output/deepseek-ai/deepseek-coder-33b-base"

args = TrainingArguments(
    output_dir=lora_path,
    overwrite_output_dir=True,
    per_device_train_batch_size=1,
    gradient_accumulation_steps=1,
    logging_steps=1,
    logging_strategy='steps',
    logging_dir = lora_path,
    num_train_epochs=4,
    logging_nan_inf_filter=True,
    save_steps=500,
    learning_rate=1e-4,
    save_strategy='steps',
    dataloader_num_workers=2,
    gradient_checkpointing=True
)

data_loader = DataLoader(tokenized_id, batch_size=args.per_device_train_batch_size, num_workers=0)

trainer = Trainer(
    model=model,
    args=args,
    train_dataset=tokenized_id,
    data_collator=DataCollatorForSeq2Seq(tokenizer=tokenizer, padding=True)
)

start_time = time.time()
trainer.train()
end_time = time.time()
elapsed_time = end_time - start_time
wandb.log({"total_runtime_seconds": elapsed_time})

import argparse

import torch
from openmind import is_torch_npu_available
from openmind import AutoTokenizer, AutoModelForCausalLM


def parse_args():
    parser = argparse.ArgumentParser()
    parser.add_argument(
        "--model_name_or_path",
        type=str,
        help="Path to model",
        default="deepseek-coder-33b-base",
    )

    args = parser.parse_args()
    return args
def main():
    args = parse_args()
    if args.model_name_or_path:
        model_path = args.model_name_or_path
    else:
        model_path = "../"

    tokenizer = AutoTokenizer.from_pretrained(model_path, trust_remote_code=True)

    model = AutoModelForCausalLM.from_pretrained(
        model_path,
        trust_remote_code=True,
        torch_dtype=torch.float16, 
        device_map="auto"
    )

    input_text = """<｜fim▁begin｜>def quick_sort(arr):
        if len(arr) <= 1:
            return arr
        pivot = arr[0]
        left = []
        right = []
    <｜fim▁hole｜>
            if arr[i] < pivot:
                left.append(arr[i])
            else:
                right.append(arr[i])
        return quick_sort(left) + [pivot] + quick_sort(right)<｜fim▁end｜>"""
    inputs = tokenizer(input_text, return_tensors="pt").to(model.device)
    outputs = model.generate(**inputs, max_length=128)
    print(tokenizer.decode(outputs[0], skip_special_tokens=True)[len(input_text):])