train_infrastructure_model.py

#!/usr/bin/env python3
# /// script
# dependencies = [
#   "trl>=0.12.0",
#   "peft>=0.7.0",
#   "transformers>=4.38.0",
#   "datasets>=2.16.0",
#   "torch>=2.1.0",
#   "accelerate>=0.26.0",
#   "bitsandbytes>=0.42.0",
#   "trackio>=0.3.0"
# ]
# ///

"""
Infrastructure Security Training - SFT Fine-tuning
Trains Qwen 2.5 7B on infrastructure management tasks
"""

from datasets import load_dataset
from transformers import AutoTokenizer, AutoModelForCausalLM, BitsAndBytesConfig
from peft import LoraConfig
from trl import SFTTrainer, SFTConfig
import torch
import trackio

# Model and dataset configuration
BASE_MODEL = "Qwen/Qwen2.5-7B-Instruct"
DATASET_NAME = "lokegud/infrastructure-security-training"
OUTPUT_MODEL = "lokegud/infrastructure-assistant-7b"

print("=" * 60)
print("Infrastructure Assistant Training")
print("=" * 60)
print(f"Base Model: {BASE_MODEL}")
print(f"Dataset: {DATASET_NAME}")
print(f"Output: {OUTPUT_MODEL}")
print("=" * 60)

# Load dataset
print("\nLoading dataset...")
dataset = load_dataset(DATASET_NAME)
train_dataset = dataset["train"]
eval_dataset = dataset["validation"]

print(f"Train examples: {len(train_dataset):,}")
print(f"Eval examples: {len(eval_dataset):,}")

# Format dataset for instruction tuning
def format_instruction(example):
    """Format examples as instruction-following prompts"""
    instruction = example["instruction"]
    input_text = example.get("input", "")
    output = example["output"]

    if input_text:
        prompt = f"### Instruction:\n{instruction}\n\n### Input:\n{input_text}\n\n### Response:\n{output}"
    else:
        prompt = f"### Instruction:\n{instruction}\n\n### Response:\n{output}"

    return {"text": prompt}

print("\nFormatting dataset...")
train_dataset = train_dataset.map(format_instruction, remove_columns=train_dataset.column_names)
eval_dataset = eval_dataset.map(format_instruction, remove_columns=eval_dataset.column_names)

# Load tokenizer
print("\nLoading tokenizer...")
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
tokenizer.pad_token = tokenizer.eos_token
tokenizer.padding_side = "right"

# QLoRA configuration for efficient training
print("Configuring QLoRA...")
bnb_config = BitsAndBytesConfig(
    load_in_4bit=True,
    bnb_4bit_quant_type="nf4",
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_use_double_quant=True,
)

# Load model
print("Loading model...")
model = AutoModelForCausalLM.from_pretrained(
    BASE_MODEL,
    quantization_config=bnb_config,
    device_map="auto",
    trust_remote_code=True,
)
model.config.use_cache = False
model.config.pretraining_tp = 1

# LoRA configuration
print("Configuring LoRA adapters...")
peft_config = LoraConfig(
    r=64,
    lora_alpha=16,
    lora_dropout=0.1,
    bias="none",
    task_type="CAUSAL_LM",
    target_modules=["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj"],
)

# Training configuration
print("Configuring training...")
training_args = SFTConfig(
    output_dir=OUTPUT_MODEL,

    # Training parameters
    num_train_epochs=3,
    per_device_train_batch_size=4,
    per_device_eval_batch_size=4,
    gradient_accumulation_steps=4,
    gradient_checkpointing=True,
    max_seq_length=2048,

    # Optimization
    learning_rate=2e-4,
    lr_scheduler_type="cosine",
    warmup_ratio=0.1,
    weight_decay=0.01,
    optim="paged_adamw_8bit",

    # Evaluation and logging
    eval_strategy="steps",
    eval_steps=50,
    logging_steps=10,
    save_strategy="steps",
    save_steps=200,
    save_total_limit=3,

    # Hub integration
    push_to_hub=True,
    hub_model_id=OUTPUT_MODEL,
    hub_strategy="every_save",
    hub_private_repo=False,

    # Tracking
    report_to="trackio",
    run_name="infrastructure-assistant-qwen-7b",

    # Performance
    bf16=True,
    max_grad_norm=0.3,
    group_by_length=True,

    # Misc
    seed=42,
    dataset_text_field="text",
    packing=False,
)

# Initialize trainer
print("Initializing trainer...")
trainer = SFTTrainer(
    model=model,
    train_dataset=train_dataset,
    eval_dataset=eval_dataset,
    peft_config=peft_config,
    args=training_args,
)

# Train
print("\n" + "=" * 60)
print("Starting training...")
print("=" * 60)

trainer.train()

# Save final model
print("\nSaving final model...")
trainer.save_model()

# Push to Hub
print("Pushing to Hub...")
trainer.push_to_hub()

print("\n" + "=" * 60)
print("Training complete!")
print("=" * 60)
print(f"Model saved to: https://huggingface.co/{OUTPUT_MODEL}")
print("\nNext steps:")
print("  1. Test the model on HuggingFace Hub")
print("  2. Convert to GGUF for Ollama deployment")
print("  3. Deploy to your infrastructure")