train.py

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.

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "unsloth/Meta-Llama-3.1-8B-Instruct-bnb-4bit",
    max_seq_length = max_seq_length,
    dtype = dtype,
    load_in_4bit = load_in_4bit,
)
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
)

prompt_context = """Bạn là một tư vấn viên hữu ích về luật.
### Instruction and Input:
Dựa vào ngữ cảnh/tài liệu sau:
{}
Hãy trả lời câu hỏi: {}

### Câu trả lời:
{}
"""

prompt = """Bạn là một tư vấn viên hữu ích về luật. 
### Instruction and Input:
Hãy trả lời câu hỏi: {}
{}

### Câu trả lời:
{}"""

EOS_TOKEN = tokenizer.eos_token # Must add EOS_TOKEN
def formatting_prompts_func(examples):
    instructions = examples["context"]
    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!
        if instructions:
            text = prompt_context.format(instruction, input, output) + EOS_TOKEN
        else:
            text = prompt.format(input, output) + EOS_TOKEN
        texts.append(text)
    return { "text" : texts, }


from datasets import load_dataset
dataset = load_dataset("json", data_files="/root/unsloth/train_data.jsonl", split="train")
dataset = dataset.map(formatting_prompts_func, batched = True,)


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

from trl import SFTTrainer
from transformers import TrainingArguments
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=16,
        gradient_accumulation_steps=4,
        num_train_epochs=1,  # Set this for 1 full training run.
        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="outputs",
        report_to="none",  # Use this for WandB, etc.
        # Save every 1000 steps
        save_steps=500,
        save_total_limit=3,  # Keep only the last 3 checkpoints
    ),
)

trainer_stats = trainer.train()

model.save_pretrained("lora_model") # Local saving
tokenizer.save_pretrained("lora_model")

model.save_pretrained_merged("model", tokenizer, save_method = "merged_16bit",)
model.save_pretrained_merged("model", tokenizer, save_method = "lora",)