Initial upload from AmkyawDev-LLM-V3

README.md

@@ -96,6 +96,42 @@ python scripts/push_to_hub.py
 
 MIT License
 
+## 🚀 Deploy to Hugging Face Spaces
+
+### Option 1: Create Space via UI
+1. Go to: https://huggingface.co/spaces
+2. Click "Create new Space"
+3. Select "Gradio" as SDK
+4. Upload the `deployment/web_ui/` folder contents
+5. Add required secrets (HF Token if needed)
+
+### Option 2: Push to Space programmatically
+
+```python
+from huggingface_hub import HfApi, login
+
+login(token="your_hf_token")
+
+api = HfApi()
+api.create_repo(
+    repo_id="amkyawdev/AmkyawDev-LLM-V3",
+    repo_type="space",
+    space_sdk="gradio"
+)
+
+api.upload_folder(
+    folder_path="deployment/web_ui",
+    repo_id="amkyawdev/AmkyawDev-LLM-V3",
+    repo_type="space",
+)
+```
+
+### Space Configuration
+The Space uses:
+- **SDK**: Gradio
+- **Python**: 3.10+
+- **Hardware**: CPU (or GPU if using Pro subscription)
+
 ## 🙏 Acknowledgments
 
 - [TRL](https://github.com/huggingface/trl) - SFTTrainer

deployment/web_ui/README.md

@@ -0,0 +1,16 @@
+# AmkyawDev-LLM-V3 Space Configuration
+
+title: AmkyawDev-LLM-V3
+emoji: 🇲🇲
+colorFrom: blue
+colorTo: green
+sdk: gradio
+sdk_version: 4.0.0
+app_file: app.py
+pinned: false
+license: mit
+tags:
+- burmese
+- language-model
+- llama
+- fine-tuned

deployment/web_ui/app.py

@@ -0,0 +1,225 @@
+#!/usr/bin/env python3
+"""
+AmkyawDev-LLM-V3 Gradio Web UI
+Burmese Language Model Chat Interface
+"""
+
+import os
+import torch
+from transformers import AutoTokenizer, AutoModelForCausalLM, AutoPeftModel
+from peft import PeftModel, PeftConfig
+import gradio as gr
+from threading import Thread
+
+
+# Model Configuration
+BASE_MODEL = "Qwen/Qwen2.5-1.5B-Instruct"  # Qwen2.5-1.5B model
+ADAPTER_PATH = "./model/adapter"  # Path to your LoRA weights
+
+# Load model and tokenizer
+def load_model():
+    """Load the fine-tuned model with LoRA adapters."""
+    
+    print("Loading tokenizer...")
+    tokenizer = AutoTokenizer.from_pretrained(
+        BASE_MODEL,
+        trust_remote_code=True
+    )
+    tokenizer.pad_token = tokenizer.eos_token
+    
+    print("Loading base model...")
+    base_model = AutoModelForCausalLM.from_pretrained(
+        BASE_MODEL,
+        trust_remote_code=True,
+        torch_dtype=torch.float16,
+        device_map="auto",
+    )
+    
+    # Check if adapter exists
+    if os.path.exists(ADAPTER_PATH) and os.listdir(ADAPTER_PATH):
+        print("Loading LoRA adapter...")
+        model = PeftModel.from_pretrained(
+            base_model,
+            ADAPTER_PATH,
+            torch_dtype=torch.float16,
+        )
+    else:
+        print("No adapter found, using base model.")
+        model = base_model
+    
+    model.eval()
+    
+    return model, tokenizer
+
+
+# Initialize model globally
+print("Initializing model... This may take a few minutes.")
+try:
+    model, tokenizer = load_model()
+    print("Model loaded successfully!")
+except Exception as e:
+    print(f"Error loading model: {e}")
+    print("Running in demo mode with mock responses.")
+    model = None
+    tokenizer = None
+
+
+def generate_response(prompt, system_prompt=None, temperature=0.7, max_tokens=512):
+    """Generate response from the model."""
+    
+    if model is None:
+        # Demo mode - return mock response
+        return "📝 ဤသည်မှာ demo mode ဖြစ်ပါတယ်။ မော်ဒယ်ပါခ်က်မပါတဲ့အတွက်စမ်းသပ်ဖြေဆိုပါတယ်။"
+    
+    # Build conversation
+    if system_prompt:
+        full_prompt = f"System: {system_prompt}\n\nUser: {prompt}\nAssistant:"
+    else:
+        full_prompt = f"User: {prompt}\nAssistant:"
+    
+    # Tokenize
+    inputs = tokenizer(full_prompt, return_tensors="pt").to(model.device)
+    
+    # Generate
+    with torch.no_grad():
+        outputs = model.generate(
+            **inputs,
+            temperature=temperature,
+            max_new_tokens=max_tokens,
+            do_sample=True,
+            top_p=0.9,
+            repetition_penalty=1.1,
+        )
+    
+    # Decode response
+    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
+    
+    # Extract assistant response
+    if "Assistant:" in response:
+        response = response.split("Assistant:")[-1].strip()
+    
+    return response
+
+
+def chat(message, history, system_prompt, temperature, max_tokens):
+    """Chat function for Gradio."""
+    
+    response = generate_response(
+        message, 
+        system_prompt=system_prompt,
+        temperature=temperature,
+        max_tokens=max_tokens
+    )
+    
+    return response
+
+
+# Build Gradio Interface
+def create_ui():
+    """Create the Gradio web UI."""
+    
+    with gr.Blocks(
+        title="AmkyawDev-LLM-V3",
+        theme=gr.themes.Soft(),
+        css="""
+        .gradio-container {max-width: 1200px !important;}
+        .main {background: linear-gradient(135deg, #667eea 0%, #764ba2 100%);}
+        """
+    ) as demo:
+        
+        gr.Markdown("""
+        # 🇲🇲 AmkyawDev-LLM-V3
+        ### Burmese Language Model Chat Interface
+        
+        ဤသည်မှာ မြန်မာဘာသာစကား Large Language Model ဖြစ်ပါတယ်။
+        """)
+        
+        with gr.Row():
+            with gr.Column(scale=3):
+                chatbot = gr.Chatbot(
+                    height=500,
+                    show_copy_button=True,
+                    bubble_full_width=False,
+                )
+                
+                with gr.Row():
+                    msg = gr.Textbox(
+                        label="Message",
+                        placeholder="မေးခွန်းရိုက်ပါတယ်...",
+                        lines=3,
+                        container=True,
+                    )
+                
+                with gr.Row():
+                    submit_btn = gr.Button("📤 ပို့မယ်", variant="primary")
+                    clear_btn = gr.Button("🗑️ ရှင်းမယ်", variant="secondary")
+            
+            with gr.Column(scale=1):
+                gr.Markdown("### ⚙️ Settings")
+                
+                system_prompt = gr.Textbox(
+                    label="System Prompt",
+                    value="You are a helpful Burmese language assistant.",
+                    lines=3,
+                )
+                
+                temperature = gr.Slider(
+                    label="Temperature",
+                    minimum=0.1,
+                    maximum=1.5,
+                    value=0.7,
+                    step=0.1,
+                )
+                
+                max_tokens = gr.Slider(
+                    label="Max Tokens",
+                    minimum=64,
+                    maximum=2048,
+                    value=512,
+                    step=64,
+                )
+        
+        # Chat functionality
+        def respond(message, history, system_prompt, temperature, max_tokens):
+            response = generate_response(
+                message,
+                system_prompt=system_prompt,
+                temperature=temperature,
+                max_tokens=max_tokens
+            )
+            history.append((message, response))
+            return "", history
+        
+        submit_btn.click(
+            respond,
+            inputs=[msg, chatbot, system_prompt, temperature, max_tokens],
+            outputs=[msg, chatbot],
+        )
+        
+        msg.submit(
+            respond,
+            inputs=[msg, chatbot, system_prompt, temperature, max_tokens],
+            outputs=[msg, chatbot],
+        )
+        
+        clear_btn.click(lambda: (None, [])), outputs=[msg, chatbot])
+        
+        gr.Markdown("""
+        ---
+        ### 📝 Notes
+        - မော်ဒယ်မပါပါက demo mode ဖြစ်ပပါတယ်။
+        - LoRA weights ပါချင်းဆိုင်းပါတယ်။
+        """)
+    
+    return demo
+
+
+# Main
+if __name__ == "__main__":
+    print("Starting AmkyawDev-LLM-V3 Web UI...")
+    demo = create_ui()
+    demo.launch(
+        server_name="0.0.0.0",
+        server_port=7860,
+        share=False,
+    )

deployment/web_ui/requirements.txt

@@ -0,0 +1,15 @@
+# AmkyawDev-LLM-V3 Web UI Requirements
+
+# Core ML
+torch>=2.0.0
+transformers>=4.36.0
+
+# PEFT for LoRA
+peft>=0.8.0
+
+# Web UI
+gradio>=4.0.0
+
+# Additional
+accelerate>=0.25.0
+numpy>=1.24.0

scripts/push_space.py

@@ -0,0 +1,53 @@
+#!/usr/bin/env python3
+"""
+Push AmkyawDev-LLM-V3 to Hugging Face Spaces
+Creates a Gradio Space for the Burmese Language Model
+"""
+
+import os
+from huggingface_hub import HfApi, login
+
+
+def create_space():
+    """Create and push the Gradio Space to Hugging Face."""
+    
+    # Get token from environment
+    token = os.environ.get("HF_TOKEN")
+    
+    if not token:
+        print("Error: No Hugging Face token found.")
+        print("Please set HF_TOKEN environment variable")
+        return
+    
+    # Login
+    print("Logging in to Hugging Face...")
+    login(token=token)
+    
+    # Initialize API
+    api = HfApi()
+    
+    repo_id = "amkyawdev/AmkyawDev-LLM-V3"
+    
+    # Create Space repository
+    print(f"Creating Space: {repo_id}")
+    api.create_repo(
+        repo_id=repo_id,
+        repo_type="space",
+        space_sdk="gradio",
+    )
+    
+    # Upload files to Space
+    print("Uploading web_ui files to Space...")
+    api.upload_folder(
+        folder_path="deployment/web_ui",
+        repo_id=repo_id,
+        repo_type="space",
+        commit_message="Initial Space upload"
+    )
+    
+    print(f"✅ Successfully created Space!")
+    print(f"   URL: https://huggingface.co/spaces/{repo_id}")
+
+
+if __name__ == "__main__":
+    create_space()

training/config.yaml

@@ -2,7 +2,7 @@
 
 # Model Configuration
 model:
-  name: "meta-llama/Meta-Llama-3-8B"  # or your base model
+  name: "Qwen/Qwen2.5-1.5B-Instruct"  # Qwen2.5-1.5B model
   trust_remote_code: true
 
 # LoRA Configuration