app.py

import gradio as gr
import numpy as np
import onnxruntime as ort
import re
import os

# ----------------------------
# CONFIGURATION (UPDATE THESE)
# ----------------------------
MODEL_PATH = "tinystories_lstm.onnx"      # Path to your ONNX model
VOCAB_PATH = "vocab.txt"                  # Path to vocab file

# Download files if not present (for Hugging Face Spaces or Colab)
if not os.path.exists(MODEL_PATH):
    from huggingface_hub import hf_hub_download
    MODEL_PATH = hf_hub_download(
        repo_id="phmd/TinyStories-LSTM-5.5M",  # 👈 REPLACE WITH YOUR HF MODEL ID
        filename="tinystories_lstm.onnx"
    )
if not os.path.exists(VOCAB_PATH):
    VOCAB_PATH = hf_hub_download(
        repo_id="phmd/TinyStories-LSTM-5.5M",
        filename="vocab.txt"
    )

# ----------------------------
# LOAD MODEL & VOCAB
# ----------------------------
print("Loading vocabulary...")
with open(VOCAB_PATH, "r") as f:
    vocab = [line.strip() for line in f]
word2idx = {word: idx for idx, word in enumerate(vocab)}
idx2word = {idx: word for word, idx in word2idx.items()}

SOS_IDX = word2idx["<SOS>"]
EOS_IDX = word2idx["<EOS>"]
PAD_IDX = word2idx["<PAD>"]
UNK_IDX = word2idx["<UNK>"]

print("Loading ONNX model...")
ort_session = ort.InferenceSession(MODEL_PATH)

# ----------------------------
# HELPER FUNCTIONS
# ----------------------------
def tokenize(text):
    text = re.sub(r'([.,!?])', r' \1 ', text.lower())
    return text.split()

def generate_story(prompt, max_new_tokens=64, temperature=0.8):
    if not prompt.strip():
        prompt = "once upon a time"
    
    tokens = tokenize(prompt)
    current_seq = [SOS_IDX] + [word2idx.get(t, UNK_IDX) for t in tokens]
    
    for _ in range(max_new_tokens):
        # Pad/truncate to 50
        padded = current_seq[-50:] if len(current_seq) > 50 else current_seq
        padded = padded + [PAD_IDX] * (50 - len(padded))
        
        input_tensor = np.array([padded], dtype=np.int64)
        logits = ort_session.run(None, {"input": input_tensor})[0]
        
        # Get logits at last real token position
        last_pos = min(len(current_seq) - 1, 49)
        next_token_logits = logits[0, last_pos, :] / temperature
        
        # Apply softmax sampling
        probs = np.exp(next_token_logits - np.max(next_token_logits))
        probs = probs / (np.sum(probs) + 1e-8)
        next_token = np.random.choice(len(probs), p=probs)
        
        if next_token == EOS_IDX:
            break
        current_seq.append(next_token)
    
    # Decode and clean
    words = [idx2word[idx] for idx in current_seq[1:] if idx != PAD_IDX]
    story = " ".join(words)
    story = re.sub(r'\s+([.,!?])', r'\1', story)  # Fix spacing
    story = story.replace(" †", '"').replace("â€", '"')  # Fix encoding artifacts
    return story

# ----------------------------
# GRADIO INTERFACE
# ----------------------------
with gr.Blocks(title="TinyStories LSTM") as demo:
    gr.Markdown("# 📖 TinyStories Word-Level LSTM")
    gr.Markdown("A **10.9 MB** LSTM that generates children's stories in seconds — **runs on CPU!**")
    
    with gr.Row():
        with gr.Column():
            prompt = gr.Textbox(
                label="Story Starter",
                value="once upon a time",
                placeholder="e.g., 'there was a brave little mouse...'"
            )
            with gr.Row():
                max_len = gr.Slider(10, 200, value=80, label="Max New Tokens")
                temp = gr.Slider(0.5, 1.5, value=0.8, label="Temperature")
            btn = gr.Button("Generate Story 🪄", variant="primary")
        
        with gr.Column():
            output = gr.Textbox(label="Generated Story", lines=15)
    
    btn.click(
        fn=generate_story,
        inputs=[prompt, max_len, temp],
        outputs=output
    )
    
    gr.Examples(
        examples=[
            ["once upon a time"],
            ["there was a robot who loved flowers"],
            ["in a faraway forest, a squirrel found a key"]
        ],
        inputs=prompt
    )
    
    gr.Markdown("""
    ---
    Model: [TinyStories LSTM (ONNX)](https://huggingface.co/your-username/tinystories-lstm-onnx) • 
    Size: 10.9 MB • 
    Runs on CPU in seconds • 
    Trained on 500k TinyStories
    """)

# Launch (for Colab or local)
if __name__ == "__main__":
    demo.launch()