train.py

import os
import torch
from datasets import load_dataset
from transformers import (
    AutoModelForCausalLM,
    AutoTokenizer,
    BitsAndBytesConfig,
    TrainingArguments,
    pipeline,
)
from peft import LoraConfig, PeftModel
from trl import SFTTrainer

# Model to fine-tune - you can change this to any of the models you want to train
# 'meta-llama/Meta-Llama-3-70B-Instruct'
# 'meta-llama/Llama-3.3-70B-Instruct'
# 'meta-llama/Meta-Llama-3-8B-Instruct'
base_model = "meta-llama/Meta-Llama-3-8B-Instruct"
new_model = "llama-3-8b-custom" # A name for your fine-tuned model

# Load the datasets
# Make sure your CSVs are in the same directory as this script
dataset = load_dataset('csv', data_files=['data_training.csv', 'data_training_1.csv'], split="train")

# 4-bit quantization configuration
compute_dtype = getattr(torch, "float16")
quant_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=compute_dtype,
    bnb_4bit_use_double_quant=False,
)

# Load the base model
model = AutoModelForCausalLM.from_pretrained(
    base_model,
    quantization_config=quant_config,
    device_map={"": 0},
    token=os.environ.get("HF_TOKEN") # Get token from secrets
)
model.config.use_cache = False
model.config.pretraining_tp = 1

# Load tokenizer
tokenizer = AutoTokenizer.from_pretrained(base_model, trust_remote_code=True, token=os.environ.get("HF_TOKEN"))
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = "right"

# PEFT configuration for LoRA
peft_params = LoraConfig(
    lora_alpha=16,
    lora_dropout=0.1,
    r=64,
    bias="none",
    task_type="CAUSAL_LM",
)

# Training parameters
training_params = TrainingArguments(
    output_dir="./results",
    num_train_epochs=1,
    per_device_train_batch_size=4,
    gradient_accumulation_steps=1,
    optim="paged_adamw_32bit",
    save_steps=25,
    logging_steps=25,
    learning_rate=2e-4,
    weight_decay=0.001,
    fp16=False,
    bf16=False,
    max_grad_norm=0.3,
    max_steps=-1,
    warmup_ratio=0.03,
    group_by_length=True,
    lr_scheduler_type="constant",
    report_to="tensorboard"
)

# Create the trainer
trainer = SFTTrainer(
    model=model,
    train_dataset=dataset,
    peft_config=peft_params,
    dataset_text_field="text", # IMPORTANT: Change "text" to the name of the column in your CSV that contains the training data
    max_seq_length=None,
    tokenizer=tokenizer,
    args=training_params,
    packing=False,
)

# Train the model
trainer.train()

# Save the fine-tuned model
trainer.model.save_pretrained(new_model)