app_train.py

"""
PromptWizard Qwen Training — Configurable Dataset & Repo
Fine-tunes Qwen using a user-selected dataset and optionally uploads
the trained model to a Hugging Face Hub repo asynchronously with logs.
"""

import gradio as gr
import spaces
import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, Trainer, TrainingArguments
from datasets import load_dataset
from peft import LoraConfig, get_peft_model, TaskType
from huggingface_hub import upload_folder, HfFolder
import os, asyncio, threading
from datetime import datetime

# ==== Async upload wrapper ====
def start_async_upload(local_dir, hf_repo, output_log):
    """Starts async model upload in a background thread."""
    def runner():
        output_log.append(f"[INFO] 🚀 Async upload thread started for repo: {hf_repo}")
        asyncio.run(async_upload_model(local_dir, hf_repo, output_log))
        output_log.append(f"[INFO] 🛑 Async upload thread finished for repo: {hf_repo}")
    threading.Thread(target=runner, daemon=True).start()


async def async_upload_model(local_dir, hf_repo, output_log, max_retries=3):
    """Upload model folder to HF Hub via HTTP API."""
    try:
        token = HfFolder.get_token()
        output_log.append(f"[INFO] ☁️ Preparing to upload to repo: {hf_repo}")
        attempt = 0
        while attempt < max_retries:
            try:
                output_log.append(f"[INFO] 🔄 Attempt {attempt+1} to upload folder via HTTP API...")
                upload_folder(folder_path=local_dir, repo_id=hf_repo, repo_type="model", token=token, ignore_patterns=["*.lock","*.tmp"], create_pr=False)
                output_log.append("[SUCCESS] ✅ Model successfully uploaded to HF Hub!")
                break
            except Exception as e:
                attempt += 1
                output_log.append(f"[ERROR] Upload attempt {attempt} failed: {e}")
                if attempt < max_retries:
                    output_log.append("[INFO] Retrying in 5 seconds...")
                    await asyncio.sleep(5)
                else:
                    output_log.append("[ERROR] ❌ Max retries reached. Upload failed.")
    except Exception as e:
        output_log.append(f"[ERROR] ❌ Unexpected error during upload: {e}")


# ==== GPU check ====
def check_gpu_status():
    return "🚀 Zero GPU Ready - GPU will be allocated when training starts"

# ==== Logging helper ====
def log_message(output_log, msg):
    line = f"[{datetime.now().strftime('%H:%M:%S')}] {msg}"
    print(line)
    output_log.append(line)

# ==== Train model ====
@spaces.GPU(duration=300)
def train_model(base_model, dataset_name, num_epochs, batch_size, learning_rate):
    output_log = []
    test_split = 0.2

    try:
        log_message(output_log, "🔍 Initializing training sequence...")

        # ===== Device =====
        device = "cuda" if torch.cuda.is_available() else "cpu"
        log_message(output_log, f"🎮 Using device: {device}")
        if device == "cuda":
            log_message(output_log, f"✅ GPU: {torch.cuda.get_device_name(0)}")

        # ===== Load dataset =====
        log_message(output_log, f"\n📚 Loading dataset: {dataset_name} ...")
        dataset = load_dataset(dataset_name)
        dataset = dataset["train"].train_test_split(test_size=test_split)
        train_dataset = dataset["train"]
        test_dataset = dataset["test"]

        # ===== Format examples =====
        def format_example(item):
            text = item.get("text") or item.get("content") or " ".join(str(v) for v in item.values())
            prompt = f"""<|system|>
You are a wise teacher interpreting Bhagavad Gita with deep insights.
<|user|>
{text}
<|assistant|>
"""
            return {"text": prompt}

        train_dataset = train_dataset.map(format_example)
        test_dataset = test_dataset.map(format_example)
        log_message(output_log, f"✅ Formatted {len(train_dataset)} train + {len(test_dataset)} test examples")

        # ===== Load model & tokenizer =====
        tokenizer = AutoTokenizer.from_pretrained(base_model, trust_remote_code=True)
        if tokenizer.pad_token is None:
            tokenizer.pad_token = tokenizer.eos_token

        model = AutoModelForCausalLM.from_pretrained(
            base_model,
            trust_remote_code=True,
            torch_dtype=torch.float16 if device=="cuda" else torch.float32,
            low_cpu_mem_usage=True
        )
        if device == "cuda":
            model = model.to(device)

        # ===== LoRA configuration =====
        lora_config = LoraConfig(task_type=TaskType.CAUSAL_LM, r=8, lora_alpha=16, lora_dropout=0.1, target_modules=["q_proj","v_proj"], bias="none")
        model = get_peft_model(model, lora_config)

        # ===== Tokenization + labels =====
        def tokenize_fn(examples):
            tokenized = tokenizer(examples["text"], padding="max_length", truncation=True, max_length=256)
            tokenized["labels"] = tokenized["input_ids"].copy()
            return tokenized

        train_dataset = train_dataset.map(tokenize_fn, batched=True)
        test_dataset = test_dataset.map(tokenize_fn, batched=True)

        # ===== Training =====
        output_dir = "./qwen-gita-lora"
        training_args = TrainingArguments(
            output_dir=output_dir,
            num_train_epochs=num_epochs,
            per_device_train_batch_size=batch_size,
            gradient_accumulation_steps=2,
            warmup_steps=10,
            logging_steps=5,
            save_strategy="epoch",
            fp16=device=="cuda",
            optim="adamw_torch",
            learning_rate=learning_rate,
            max_steps=100
        )

        trainer = Trainer(model=model, args=training_args, train_dataset=train_dataset, eval_dataset=test_dataset, tokenizer=tokenizer)
        log_message(output_log, "\n🚀 Starting training...")
        trainer.train()
        trainer.save_model(output_dir)
        tokenizer.save_pretrained(output_dir)
        log_message(output_log, "\n✅ Training finished and model saved locally.")

        return "\n".join(output_log), model, tokenizer, output_dir

    except Exception as e:
        log_message(output_log, f"\n❌ Error during training: {e}")
        return "\n".join(output_log), None, None, None

# ==== Gradio Interface ====
def create_interface():
    with gr.Blocks(title="PromptWizard — Qwen Trainer") as demo:
        gr.Markdown("""
        # 🧘 PromptWizard Qwen Fine-tuning  
        Fine-tune Qwen and interact with it before optional upload.
        """)

        with gr.Row():
            with gr.Column():
                gr.Textbox(label="GPU Status", value=check_gpu_status(), interactive=False)
                base_model = gr.Textbox(label="Base Model", value="Qwen/Qwen2.5-0.5B")
                dataset_name = gr.Textbox(label="Dataset Name", value="rahul7star/Gita")
                hf_repo = gr.Textbox(label="HF Repo for Upload", value="rahul7star/Qwen0.5-3B-Gita")
                num_epochs = gr.Slider(1, 3, value=1, step=1, label="Epochs")
                batch_size = gr.Slider(1, 4, value=2, step=1, label="Batch Size")
                learning_rate = gr.Number(value=5e-5, label="Learning Rate")
                train_btn = gr.Button("🚀 Start Fine-tuning", variant="primary")
                upload_btn = gr.Button("☁️ Upload Model to HF Hub", variant="secondary", interactive=False)

            with gr.Column():
                output = gr.Textbox(label="Training Log", lines=25, max_lines=40,
                                    value="Click 'Start Fine-tuning' to train your model.")
                user_question = gr.Textbox(label="Ask your own question", placeholder="Type a question...")
                answer_box = gr.Textbox(label="Answer", lines=5, interactive=False)

        # ==== Train button ====
        def train_click(base_model, dataset_name, num_epochs, batch_size, learning_rate):
            log, model, tokenizer, output_dir = train_model(base_model, dataset_name, num_epochs, batch_size, learning_rate)
            return log, True, model, tokenizer, output_dir

        train_btn.click(
            fn=train_click,
            inputs=[base_model, dataset_name, num_epochs, batch_size, learning_rate],
            outputs=[output, upload_btn, gr.State(), gr.State(), gr.State()],
        )

        # ==== User question ====
        def ask_question(user_input, model, tokenizer):
            if not model or not tokenizer:
                return "Model not loaded yet."
            device = "cuda" if torch.cuda.is_available() else "cpu"
            inputs = tokenizer(f"<|system|>\nYou are a wise teacher interpreting Bhagavad Gita.\n<|user|>\n{user_input}\n<|assistant|>\n", return_tensors="pt").to(device)
            outputs = model.generate(**inputs, max_new_tokens=100)
            answer = tokenizer.decode(outputs[0], skip_special_tokens=True)
            return answer

        user_question.submit(ask_question, inputs=[user_question, gr.State(), gr.State()], outputs=[answer_box])

        # ==== Upload button ====
        def upload_click(hf_repo):
            output_log = []
            start_async_upload("./qwen-gita-lora", hf_repo, output_log)
            return "\n".join(output_log)

        upload_btn.click(upload_click, inputs=[hf_repo], outputs=[output])

    return demo

if __name__ == "__main__":
    demo = create_interface()
    demo.launch(server_name="0.0.0.0", server_port=7860)