Upload folder using huggingface_hub

README.md

@@ -1,12 +1,36 @@
 ---
-title: Gemma 3n Luau Demo
-emoji: 🐨
-colorFrom: purple
-colorTo: gray
+title: Gemma 3n E4B Luau Generator
+emoji: 🎮
+colorFrom: blue
+colorTo: purple
 sdk: gradio
-sdk_version: 5.44.1
+sdk_version: 4.44.0
 app_file: app.py
 pinned: false
+license: apache-2.0
 ---
 
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+# 🎮 Gemma-3n-E4B Luau Code Generator
+
+This Space hosts the Gemma-3n-E4B model (8B parameters, 4B runtime efficiency) fine-tuned on the Roblox Luau corpus.
+
+## Model Details
+- **Base Model**: google/gemma-3n-E4B
+- **Architecture**: Novel architecture with 8B parameters but 4B model runtime footprint
+- **Fine-tuned Dataset**: Roblox/luau_corpus
+- **Task**: Luau code generation for Roblox development
+- **Model Repository**: [Laserhun/gemma-3n-E4B-luau-finetuned](https://huggingface.co/Laserhun/gemma-3n-E4B-luau-finetuned)
+
+## About Gemma-3n-E4B
+Gemma-3n-E4B uses an innovative architecture that provides 8 billion parameter model quality while maintaining the runtime efficiency of a 4 billion parameter model.
+
+## Features
+- Advanced Luau code generation
+- Roblox-specific patterns and best practices
+- Efficient memory usage despite large parameter count
+- High-quality code output
+
+## Training Details
+- Fine-tuned on Roblox Luau corpus
+- Uses LoRA for efficient adaptation
+- Optimized for code generation tasks

app.py

@@ -0,0 +1,103 @@
+import gradio as gr
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+from peft import PeftModel, PeftConfig
+
+# Model configuration - Gemma-3n-E4B fine-tuned
+MODEL_ID = "Laserhun/gemma-3n-E4B-luau-finetuned"
+BASE_MODEL_ID = "google/gemma-3n-E4B"
+
+print("Loading Gemma-3n-E4B fine-tuned model...")
+try:
+    # Try loading as PEFT model
+    peft_config = PeftConfig.from_pretrained(MODEL_ID)
+    
+    # Load base model
+    base_model = AutoModelForCausalLM.from_pretrained(
+        BASE_MODEL_ID,
+        torch_dtype=torch.float16,
+        device_map="auto",
+        trust_remote_code=True,
+        ignore_mismatched_sizes=True
+    )
+    
+    # Load PEFT adapters
+    model = PeftModel.from_pretrained(base_model, MODEL_ID)
+    print("Loaded Gemma-3n-E4B as PEFT model")
+except:
+    # Load as regular model
+    model = AutoModelForCausalLM.from_pretrained(
+        MODEL_ID,
+        torch_dtype=torch.float16,
+        device_map="auto",
+        trust_remote_code=True
+    )
+    print("Loaded as regular model")
+
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+if not tokenizer.pad_token:
+    tokenizer.pad_token = tokenizer.eos_token
+
+def generate_luau_code(prompt, max_length=512, temperature=0.7, top_p=0.95):
+    """Generate Luau code using Gemma-3n-E4B model"""
+    
+    # Format for Gemma-3n
+    formatted_prompt = f"<start_of_turn>user\n{prompt}<end_of_turn>\n<start_of_turn>model\n"
+    
+    # Tokenize
+    inputs = tokenizer(formatted_prompt, return_tensors="pt", truncation=True, max_length=512)
+    
+    # Move to device
+    inputs = {k: v.to(model.device) for k, v in inputs.items()}
+    
+    # Generate
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            max_new_tokens=max_length,
+            temperature=temperature,
+            top_p=top_p,
+            do_sample=True,
+            pad_token_id=tokenizer.pad_token_id,
+            eos_token_id=tokenizer.eos_token_id
+        )
+    
+    # Decode
+    generated_text = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    
+    # Extract response
+    if "<start_of_turn>model" in generated_text:
+        response = generated_text.split("<start_of_turn>model")[-1].strip()
+    else:
+        response = generated_text[len(formatted_prompt):].strip()
+    
+    return response
+
+# Create Gradio interface
+iface = gr.Interface(
+    fn=generate_luau_code,
+    inputs=[
+        gr.Textbox(
+            lines=4,
+            placeholder="Describe the Luau code you want to generate...",
+            label="Enter your Luau code request"
+        ),
+        gr.Slider(minimum=100, maximum=1000, value=512, step=50, label="Max Length"),
+        gr.Slider(minimum=0.1, maximum=1.0, value=0.7, step=0.1, label="Temperature"),
+        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top P")
+    ],
+    outputs=gr.Code(language="lua", label="Generated Luau Code"),
+    title="🎮 Gemma-3n-E4B Luau Code Generator",
+    description="Generate Roblox Luau code using Gemma-3n-E4B model (8B parameters, 4B runtime) fine-tuned on Luau corpus.",
+    examples=[
+        ["Create a smooth part movement function with easing", 512, 0.7, 0.95],
+        ["Write a door script with click interaction and smooth animation", 512, 0.7, 0.95],
+        ["Generate a complete inventory system with add, remove, and display functions", 700, 0.7, 0.95],
+        ["Create a spawning system for objects at random positions", 400, 0.7, 0.95],
+        ["Write a leaderboard system that saves player scores", 600, 0.7, 0.95]
+    ],
+    theme=gr.themes.Soft()
+)
+
+if __name__ == "__main__":
+    iface.launch()

requirements.txt

@@ -0,0 +1,9 @@
+transformers>=4.35.0
+torch>=2.0.0
+gradio>=4.0.0
+accelerate
+peft
+sentencepiece
+protobuf
+bitsandbytes
+timm