Files
pipelines/prompt_bench/finetune.py
2026-04-13 15:43:01 -04:00

215 lines
6.6 KiB
Python

"""Fine-tune Qwen 3.5 4B on bill summarization data using Unsloth.
Loads a ChatML-style JSONL dataset (system/user/assistant messages),
applies QLoRA with 4-bit quantization, and saves the merged model
in HuggingFace format. Designed for a single RTX 3090 (24GB).
Usage:
python -m python.prompt_bench.finetune \
--dataset output/finetune_dataset.jsonl \
--output-dir output/qwen-bill-summarizer
"""
from __future__ import annotations
import json
import logging
from dataclasses import dataclass
from pathlib import Path
from typing import Annotated
import tomllib
import typer
from unsloth import FastLanguageModel
from datasets import Dataset
from transformers import TrainingArguments
from trl import SFTTrainer
logger = logging.getLogger(__name__)
@dataclass
class LoraConfig:
"""LoRA adapter hyperparameters."""
rank: int
alpha: int
dropout: float
targets: list[str]
@dataclass
class TrainingConfig:
"""Training loop hyperparameters."""
learning_rate: float
epochs: int
batch_size: int
gradient_accumulation: int
max_seq_length: int
warmup_ratio: float
weight_decay: float
logging_steps: int
save_steps: int
@dataclass
class FinetuneConfig:
"""Top-level finetune configuration."""
base_model: str
lora: LoraConfig
training: TrainingConfig
@classmethod
def from_toml(cls, config_path: Path) -> FinetuneConfig:
"""Load finetune config from a TOML file."""
raw = tomllib.loads(config_path.read_text())["finetune"]
return cls(
base_model=raw["base_model"],
lora=LoraConfig(**raw["lora"]),
training=TrainingConfig(**raw["training"]),
)
def _messages_to_chatml(messages: list[dict]) -> str:
r"""Convert a message list to Qwen ChatML format.
Produces:
<|im_start|>system\n...\n<|im_end|>
<|im_start|>user\n...\n<|im_end|>
<|im_start|>assistant\n...\n<|im_end|>
"""
parts = []
for message in messages:
role = message["role"]
content = message["content"]
parts.append(f"<|im_start|>{role}\n{content}<|im_end|>")
return "\n".join(parts)
def load_dataset_from_jsonl(path: Path) -> Dataset:
"""Load a ChatML JSONL file into a HuggingFace Dataset.
Each line must have {"messages": [{"role": ..., "content": ...}, ...]}.
Pre-formats into a `text` column with the Qwen ChatML template applied,
which SFTTrainer consumes directly.
"""
records = []
with path.open(encoding="utf-8") as handle:
for raw_line in handle:
stripped = raw_line.strip()
if stripped:
entry = json.loads(stripped)
records.append({"text": _messages_to_chatml(entry["messages"])})
logger.info("Loaded %d examples from %s", len(records), path)
return Dataset.from_list(records)
def main(
dataset_path: Annotated[Path, typer.Option("--dataset", help="Fine-tuning JSONL")] = Path(
"output/finetune_dataset.jsonl",
),
validation_split: Annotated[float, typer.Option("--val-split", help="Fraction held out for validation")] = 0.1,
output_dir: Annotated[Path, typer.Option("--output-dir", help="Where to save the merged model")] = Path(
"output/qwen-bill-summarizer",
),
config_path: Annotated[
Path,
typer.Option("--config", help="TOML config file"),
] = Path(__file__).parent / "config.toml",
save_gguf: Annotated[bool, typer.Option("--save-gguf/--no-save-gguf", help="Also save GGUF")] = False,
) -> None:
"""Fine-tune Qwen 3.5 4B on bill summarization with Unsloth + QLoRA."""
logging.basicConfig(level="INFO", format="%(asctime)s %(levelname)s %(name)s: %(message)s")
if not dataset_path.is_file():
message = f"Dataset not found: {dataset_path}"
raise typer.BadParameter(message)
config = FinetuneConfig.from_toml(config_path)
logger.info("Loading base model: %s", config.base_model)
model, tokenizer = FastLanguageModel.from_pretrained(
model_name=config.base_model,
max_seq_length=config.training.max_seq_length,
load_in_4bit=True,
dtype=None,
)
logger.info("Applying LoRA (rank=%d, alpha=%d)", config.lora.rank, config.lora.alpha)
model = FastLanguageModel.get_peft_model(
model,
r=config.lora.rank,
lora_alpha=config.lora.alpha,
lora_dropout=config.lora.dropout,
target_modules=config.lora.targets,
bias="none",
use_gradient_checkpointing="unsloth",
random_state=42,
)
full_dataset = load_dataset_from_jsonl(dataset_path)
split = full_dataset.train_test_split(test_size=validation_split, seed=42)
train_dataset = split["train"]
validation_dataset = split["test"]
logger.info("Split: %d train, %d validation", len(train_dataset), len(validation_dataset))
training_args = TrainingArguments(
output_dir=str(output_dir / "checkpoints"),
num_train_epochs=config.training.epochs,
per_device_train_batch_size=config.training.batch_size,
gradient_accumulation_steps=config.training.gradient_accumulation,
learning_rate=config.training.learning_rate,
warmup_ratio=config.training.warmup_ratio,
weight_decay=config.training.weight_decay,
lr_scheduler_type="cosine",
logging_steps=config.training.logging_steps,
save_steps=config.training.save_steps,
save_total_limit=3,
eval_strategy="steps",
eval_steps=config.training.save_steps,
load_best_model_at_end=True,
bf16=True,
optim="adamw_8bit",
seed=42,
report_to="none",
)
trainer = SFTTrainer(
model=model,
tokenizer=tokenizer,
train_dataset=train_dataset,
eval_dataset=validation_dataset,
args=training_args,
max_seq_length=config.training.max_seq_length,
packing=True,
)
logger.info(
"Starting training: %d train, %d val, %d epochs",
len(train_dataset),
len(validation_dataset),
config.training.epochs,
)
trainer.train()
merged_path = str(output_dir / "merged")
logger.info("Saving merged model to %s", merged_path)
model.save_pretrained_merged(merged_path, tokenizer, save_method="merged_16bit")
if save_gguf:
gguf_path = str(output_dir / "gguf")
logger.info("Saving GGUF to %s", gguf_path)
model.save_pretrained_gguf(gguf_path, tokenizer, quantization_method="q4_k_m")
logger.info("Done! Model saved to %s", output_dir)
def cli() -> None:
"""Typer entry point."""
typer.run(main)
if __name__ == "__main__":
cli()