SFT Training Test: Qwen3-4B¶

# Environment Setup
import os
from dotenv import load_dotenv
load_dotenv()

# Force text-based progress instead of HTML widgets
os.environ["TQDM_NOTEBOOK"] = "false"

# CRITICAL: Import unsloth FIRST for proper TRL patching
import unsloth
from unsloth import FastLanguageModel, is_bf16_supported

import torch

# Environment summary
gpu = torch.cuda.get_device_name(0) if torch.cuda.is_available() else "CPU"
print(f"Environment: unsloth {unsloth.__version__}, PyTorch {torch.__version__}, {gpu}")
print(f"HF_TOKEN loaded: {'Yes' if os.environ.get('HF_TOKEN') else 'No'}")

# Load Qwen3-4B with 4-bit quantization
MODEL_NAME = "unsloth/Qwen3-4B-unsloth-bnb-4bit"
print(f"\nLoading {MODEL_NAME.split('/')[-1]}...")

model, tokenizer = FastLanguageModel.from_pretrained(
    MODEL_NAME,
    max_seq_length=512,
    load_in_4bit=True,
    dtype=None,  # Auto-detect
)
print(f"Model loaded: {type(model).__name__}")

# Apply LoRA adapters
model = FastLanguageModel.get_peft_model(
    model,
    r=16,
    lora_alpha=16,
    lora_dropout=0,
    target_modules=["q_proj", "k_proj", "v_proj", "o_proj",
                    "gate_proj", "up_proj", "down_proj"],
    bias="none",
    use_gradient_checkpointing="unsloth",
    random_state=42,
)

trainable = sum(p.numel() for p in model.parameters() if p.requires_grad)
total = sum(p.numel() for p in model.parameters())
print(f"LoRA applied: {trainable:,} trainable / {total:,} total ({100*trainable/total:.2f}%)")

# Create minimal synthetic instruction dataset (5 samples)
from datasets import Dataset

# Synthetic instruction-response pairs for testing
synthetic_data = [
    {"instruction": "What is machine learning?",
     "response": "Machine learning is a subset of artificial intelligence where computers learn patterns from data."},
    {"instruction": "Explain Python in one sentence.",
     "response": "Python is a high-level programming language known for its readability and versatility."},
    {"instruction": "What is a neural network?",
     "response": "A neural network is a computational model inspired by biological neurons that processes information."},
    {"instruction": "Define supervised learning.",
     "response": "Supervised learning is training a model on labeled data to predict outcomes for new inputs."},
    {"instruction": "What is gradient descent?",
     "response": "Gradient descent is an optimization algorithm that minimizes loss by iteratively adjusting parameters."},
]

# Format as chat conversations
def format_conversation(sample):
    messages = [
        {"role": "user", "content": sample["instruction"]},
        {"role": "assistant", "content": sample["response"]}
    ]
    return {"text": tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=False)}

dataset = Dataset.from_list(synthetic_data)
dataset = dataset.map(format_conversation, remove_columns=["instruction", "response"])
print(f"Dataset created: {len(dataset)} samples")
print(f"Sample: {dataset[0]['text'][:100]}...")

# Post-training inference test
FastLanguageModel.for_inference(model)

messages = [{"role": "user", "content": "What is deep learning?"}]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

with torch.no_grad():
    outputs = model.generate(
        **inputs,
        max_new_tokens=64,
        temperature=0.7,
        top_p=0.9,
        do_sample=True,
        pad_token_id=tokenizer.pad_token_id,
    )

response = tokenizer.decode(outputs[0], skip_special_tokens=True)
print("=" * 60)
print("SFT Training Pipeline Test PASSED")
print("=" * 60)
print(f"Sample generation:\n{response[-200:]}")