Unsloth QAT LoRA

gistfile1.txt

from unsloth import FastLanguageModel
import torch
from torchao.quantization import Int4WeightOnlyConfig
from transformers import AutoModelForCausalLM, TextStreamer, TorchAoConfig


qat_scheme = "int4"
save_output_path = "/tmp/unsloth_model"
max_seq_length = 2048

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "unsloth/Meta-Llama-3.1-8B",
    max_seq_length = max_seq_length,
    dtype = torch.bfloat16,
    load_in_4bit = False,
    full_finetuning = False,
)

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 = "unsloth",
    random_state = 3407,
    use_rslora = False,
    loftq_config = None,
    qat_scheme = qat_scheme,
)


# ============
#  Data prep |
# ============

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("yahma/alpaca-cleaned", split = "train")
dataset = dataset.map(formatting_prompts_func, batched = True,)


# ========
#  Train |
# ========

from trl import SFTConfig, SFTTrainer
trainer = SFTTrainer(
    model = model,
    tokenizer = tokenizer,
    train_dataset = dataset,
    dataset_text_field = "text",
    max_seq_length = max_seq_length,
    packing = False, # Can make training 5x faster for short sequences.
    args = SFTConfig(
        per_device_train_batch_size = 16,
        gradient_accumulation_steps = 1,
        warmup_steps = 5,
        num_train_epochs = 1, # Set this for 1 full training run.
        max_steps = 10,
        learning_rate = 4e-5,
        logging_steps = 1,
        optim = "adamw_8bit",
        weight_decay = 0.01,
        lr_scheduler_type = "linear",
        seed = 3407,
        output_dir = "outputs",
        report_to = "none", # Use this for WandB etc
    ),
)

trainer_stats = trainer.train()


# =============
#  Model save |
# =============

model.save_pretrained(save_output_path)
tokenizer.save_pretrained(save_output_path)
quantization_config = TorchAoConfig(Int4WeightOnlyConfig())
quantized_model = AutoModelForCausalLM.from_pretrained(
    save_output_path,
    torch_dtype="auto",
    device_map="auto",
    quantization_config=quantization_config,
)
quantized_model.save_pretrained(save_output_path + "_quantized", safe_serialization=False)
tokenizer.save_pretrained(save_output_path + "_quantized")


# ============
#  Fibonacci |
# ============

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")

text_streamer = TextStreamer(tokenizer)
_ = quantized_model.generate(**inputs, streamer = text_streamer, max_new_tokens = 128)