unsloth_test_alpaca(一个unsloth炼丹的小例子)

aifeifei798 aifeifei

准备数据

先把需要的数据转到alpaca的格式,并且重新整理

from datasets import load_dataset, Dataset
# Load the dataset and select only the required columns
dataset = load_dataset("./Chinese-DeepSeek-R1-Distill-data-110k", split="train")
dataset = dataset.select_columns(["input", "content", "reasoning_content"])

new_data = {
    "input": dataset["input"],
    "instruction": ["深度思索"] * len(dataset),
    "output": ['<think>' + rc + '</think>' + c for rc, c in zip(dataset["reasoning_content"], dataset["content"])]
}
new_dataset = Dataset.from_dict(new_data)
new_dataset.save_to_disk("Chinese-DeepSeek-R1-Distill-data-110k-alpaca")

开始炼丹

from unsloth import FastLanguageModel
import torch
max_seq_length = 2048 # Choose any! We auto support RoPE Scaling internally!
dtype = None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+
load_in_4bit = True # Use 4bit quantization to reduce memory usage. Can be False.

# 4bit pre quantized models we support for 4x faster downloading + no OOMs.
fourbit_models = [
    "unsloth/Meta-Llama-3.1-8B-bnb-4bit",      # Llama-3.1 2x faster
    "unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit",
    "unsloth/Meta-Llama-3.1-70B-bnb-4bit",
    "unsloth/Meta-Llama-3.1-405B-bnb-4bit",    # 4bit for 405b!
    "unsloth/Mistral-Small-Instruct-2409",     # Mistral 22b 2x faster!
    "unsloth/mistral-7b-instruct-v0.3-bnb-4bit",
    "unsloth/Phi-3.5-mini-instruct",           # Phi-3.5 2x faster!
    "unsloth/Phi-3-medium-4k-instruct",
    "unsloth/gemma-2-9b-bnb-4bit",
    "unsloth/gemma-2-27b-bnb-4bit",            # Gemma 2x faster!

    "unsloth/Llama-3.2-1B-bnb-4bit",           # NEW! Llama 3.2 models
    "unsloth/Llama-3.2-1B-Instruct-bnb-4bit",
    "unsloth/Llama-3.2-3B-bnb-4bit",
    "unsloth/Llama-3.2-3B-Instruct-bnb-4bit",

    "unsloth/Llama-3.3-70B-Instruct-bnb-4bit" # NEW! Llama 3.3 70B!
] # More models at https://huggingface.co/unsloth

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "./Llama-3.2-1B-Instruct-unsloth-bnb-4bit", # or choose "unsloth/Llama-3.2-1B-Instruct"
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
    # token = "hf_...", # use one if using gated models like meta-llama/Llama-2-7b-hf
)

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
)

from unsloth.chat_templates import get_chat_template

tokenizer = get_chat_template(
    tokenizer,
    chat_template = "llama-3.2",
)

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 # Must add 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):
        # Must add EOS_TOKEN, otherwise your generation will go on forever!
        text = alpaca_prompt.format(instruction, input, output) + EOS_TOKEN
        texts.append(text)
    return { "text" : texts, }
pass

from datasets import load_dataset
dataset = load_dataset("aifeifei798/Chinese-DeepSeek-R1-Distill-data-110k-alpaca", split = "train")
dataset = dataset.map(formatting_prompts_func, batched = True,)
print(dataset[5])

from trl import SFTTrainer
from transformers import TrainingArguments, DataCollatorForSeq2Seq
from unsloth import is_bfloat16_supported

trainer = SFTTrainer(
    model = model,
    tokenizer = tokenizer,
    train_dataset = dataset,
    dataset_text_field = "text",
    max_seq_length = max_seq_length,
    dataset_num_proc = 16,
    packing = False, # Can make training 5x faster for short sequences.
    args = TrainingArguments(
        per_device_train_batch_size = 1,
        gradient_accumulation_steps = 4,
        warmup_steps = 5,
        # num_train_epochs = 1, # Set this for 1 full training run.
        max_steps = 100,
        learning_rate = 2e-4,
        fp16 = not is_bfloat16_supported(),
        bf16 = is_bfloat16_supported(),
        logging_steps = 1,
        optim = "adamw_8bit",
        weight_decay = 0.01,
        lr_scheduler_type = "linear",
        seed = 3407,
        output_dir = "Llama-3.2-1B-Instruct-unsloth-bnb-4bit-outputs",
        report_to = "none", # Use this for WandB etc
        # 新增：每5步保存一次检查点
        save_steps=5, # 中途断了可以继续,救命的地方
        # 可选：限制总检查点数量，避免硬盘爆满
        save_total_limit=10,  # 保留最近10个检查点
        # resume_from_checkpoint="Llama-3.2-1B-Instruct-unsloth-bnb-4bit-outputs/checkpoint-300",  # 中途断了继续,救命的地方
    ),
)

trainer_stats = trainer.train()

# alpaca_prompt = Copied from above
FastLanguageModel.for_inference(model) # Enable native 2x faster inference
inputs = tokenizer(
[
    alpaca_prompt.format(
        "Continue the fibonnaci sequence.", # instruction
        "1, 1, 2, 3, 5, 8", # input
        "", # output - leave this blank for generation!
    )
], return_tensors = "pt").to("cuda")

# alpaca_prompt = Copied from above
FastLanguageModel.for_inference(model) # Enable native 2x faster inference
inputs = tokenizer(
[
    alpaca_prompt.format(
        "Continue the fibonnaci sequence.", # instruction
        "1, 1, 2, 3, 5, 8", # input
        "", # output - leave this blank for generation!
    )
], return_tensors = "pt").to("cuda")

from transformers import TextStreamer
text_streamer = TextStreamer(tokenizer)
_ = model.generate(**inputs, streamer = text_streamer, max_new_tokens = 128)

outputs = model.generate(**inputs, max_new_tokens = 64, use_cache = True)
tokenizer.batch_decode(outputs)
                   
model.save_pretrained("Llama-3.2-1B-Instruct-unsloth-bnb-4bit-lora") # Local saving
tokenizer.save_pretrained("Llama-3.2-1B-Instruct-unsloth-bnb-4bit-lora")
model.save_pretrained_merged("Llama-3.2-1B-Instruct-unsloth-bnb-4bit-end", tokenizer)

炼完看看

from unsloth import FastLanguageModel
from unsloth import PatchFastRL
PatchFastRL("GRPO", FastLanguageModel)
import torch
max_seq_length = 2048 # Choose any! We auto support RoPE Scaling internally!
dtype = None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+
load_in_4bit = True # Use 4bit quantization to reduce memory usage. Can be False.

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:
{}"""

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "./Llama-3.2-1B-Instruct-unsloth-bnb-4bit-end", # or choose "unsloth/Llama-3.2-1B-Instruct"
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
    # token = "hf_...", # use one if using gated models like meta-llama/Llama-2-7b-hf
)

if False:
    model, tokenizer = FastLanguageModel.from_pretrained(
        model_name = "./DeepSeek-R1-Distill-Llama-8B-bnb-4bit-lora", # YOUR MODEL YOU USED FOR TRAINING
        max_seq_length = max_seq_length,
        dtype = dtype,
        load_in_4bit = load_in_4bit,
    )
    FastLanguageModel.for_inference(model) # Enable native 2x faster inference

# alpaca_prompt = You MUST copy from above!

inputs = tokenizer(
[
    alpaca_prompt.format(
        "深度思索", # instruction
        """ 能给我讲一个寓意深刻的故事吗？ """, # input
        "", # output - leave this blank for generation!
    )
], return_tensors = "pt").to("cuda")

from transformers import TextStreamer
text_streamer = TextStreamer(tokenizer)
_ = model.generate(**inputs, streamer = text_streamer, max_new_tokens = 1024)

总结

• 'loss': 2.2363 只看这个结果,好像0.75左右好点吧,我训练了400步才这样,算了,电脑嗷嗷叫
• 这就是炼丹,看结果,重复炼丹,看结果.训练模型没那么难.就是拼硬件,拼电费,拼准备的数据