from unsloth import FastLanguageModel, get_chat_template
import torch

# 加载模型并合并 LoRA
model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "mistralai/Mistral-Small-3.1-24B-Base-2503",
    dtype = None,
    max_seq_length = 2048,
    load_in_4bit = True,
)
model = FastLanguageModel.get_peft_model(
    model,
    r = 16, # Choose any number > 0 ! Suggested 8, 16, 32, 64, 128
    target_modules = ["q_proj", "k_proj", "v_proj", "o_proj",
                      "gate_proj", "up_proj", "down_proj",],
    lora_alpha = 16,
    lora_dropout = 0, # Supports any, but = 0 is optimized
    bias = "none",    # Supports any, but = "none" is optimized
    # [NEW] "unsloth" uses 30% less VRAM, fits 2x larger batch sizes!
    use_gradient_checkpointing = "unsloth", # True or "unsloth" for very long context
    random_state = 3407,
    use_rslora = False,  # We support rank stabilized LoRA
    loftq_config = None, # And LoftQ
)
model.load_adapter("/home/outputs/checkpoint-4000","default")
tokenizer = get_chat_template(tokenizer, chat_template="chatml", map_eos_token=True)

print(type(model)) 


# 正确保存合并模型（使用 Unsloth 官方函数）
model.save_pretrained_merged(
    save_directory = "/home/merged_model1",
    tokenizer = tokenizer,
    save_method = "merged_16bit",
)