upgrade to Qwen3-Coder-30B-A3B-Instruct (MoE, 3B active) - ZeroGPU xlarge

README.md

@@ -1,5 +1,5 @@
 ---
-title: CodeCraft - Qwen2.5-Coder-7B
+title: CodeCraft - Qwen3-Coder-30B
 emoji: 💻
 colorFrom: indigo
 colorTo: purple
@@ -12,29 +12,37 @@ python_version: "3.12"
 
 # CodeCraft - AI Coding Assistant
 
-Powered by **Qwen2.5-Coder-7B-Instruct** running on Hugging Face ZeroGPU.
+Powered by **Qwen3-Coder-30B-A3B-Instruct** running on Hugging Face ZeroGPU (xlarge).
+A Mixture-of-Experts model with 30B total params (3B active) — matches or exceeds much larger models on coding benchmarks.
 
-Chat with a state-of-the-art coding assistant. Supports code generation, debugging,
-refactoring, explanation, and general programming help across all major languages.
+Chat with a cutting-edge coding assistant. Supports code generation, debugging, refactoring, explanation, and general programming help across all major languages.
 
 ## Features
 
 - 💬 **Chat interface** with syntax-highlighted code blocks
-- ⚙️ **Adjustable parameters**: temperature, top-p, max tokens, system prompt
-- 📡 **Built-in API endpoint** at `/api` for programmatic use
+- ⚙️ **Adjustable parameters**: temperature, top-p, max tokens (up to 8192), system prompt
+- 📡 **Built-in API endpoint** at `/gradio_api/call/predict` for programmatic use
 - 🎨 **Syntax-highlighted output** via Gradio Markdown + code blocks
+- 🔄 **256K context** — handle entire codebases in conversation
+- 🛠️ **Agentic tool-use** support (function calling, ReAct, structured output)
 
 ## API Usage
 
-Every Gradio Space exposes a Rest API at `/api`. For this Space:
+Every Gradio Space exposes a REST API at `/gradio_api/call/predict`. For this Space:
 
 ```python
 import requests
 
 response = requests.post(
-    "https://<your-space>.hf.space/gradio_api/call/generate",
+    "https://serotoninboi-codecraft.hf.space/gradio_api/call/predict",
     json={
-        "data": ["write a fibonacci function in rust", "You are a helpful coding assistant.", 0.3, 0.9, 2048]
+        "data": [
+            "write a fibonacci function in rust",
+            "You are a helpful coding assistant.",
+            0.3,
+            0.9,
+            2048
+        ]
     }
 )
 print(response.json())
@@ -42,5 +50,4 @@ print(response.json())
 
 ## Model
 
-[Qwen/Qwen2.5-Coder-7B-Instruct](https://huggingface.co/Qwen/Qwen2.5-Coder-7B-Instruct) —
-7B parameter code-specific LLM with 128K context, instruction-tuned for chat and coding tasks.
+[Qwen/Qwen3-Coder-30B-A3B-Instruct](https://huggingface.co/Qwen/Qwen3-Coder-30B-A3B-Instruct) — 30B MoE (3B active) code-specific LLM with 256K context, built for agentic coding tasks.

app.py

@@ -8,12 +8,12 @@ from transformers import (
     TextIteratorStreamer,
 )
 from threading import Thread
-from typing import Optional, Generator
+from typing import Generator
 
 # ---------------------------------------------------------------------------
-# Module-scope model loading — ZeroGPU manages GPU offload transparently
+# Module-scope model loading - ZeroGPU manages GPU offload transparently
 # ---------------------------------------------------------------------------
-MODEL_ID = "Qwen/Qwen2.5-Coder-7B-Instruct"
+MODEL_ID = "Qwen/Qwen3-Coder-30B-A3B-Instruct"
 
 quant_config = BitsAndBytesConfig(
     load_in_4bit=True,
@@ -36,9 +36,9 @@ DEFAULT_SYSTEM = "You are an expert coding assistant. Write clean, efficient, we
 
 
 # ---------------------------------------------------------------------------
-# ZeroGPU-decorated generation
+# ZeroGPU-decorated generation - xlarge for 30B MoE model
 # ---------------------------------------------------------------------------
-@spaces.GPU(duration=120)
+@spaces.GPU(duration=180, size="xlarge")
 def generate(
     messages: list[dict],
     temperature: float,
@@ -46,9 +46,8 @@ def generate(
     max_new_tokens: int,
 ) -> str:
     """Run model inference inside a ZeroGPU worker process.
-
     Args are pickled across the process boundary.
-    Returns CPU text — safe for unpickling in the main process.
+    Returns CPU text - safe for unpickling in the main process.
     """
     inputs = tokenizer.apply_chat_template(
         messages,
@@ -72,9 +71,9 @@ def generate(
 
 
 # ---------------------------------------------------------------------------
-# Streaming variant — yields tokens as they're generated
+# Streaming variant - yields tokens as they're generated
 # ---------------------------------------------------------------------------
-@spaces.GPU(duration=120)
+@spaces.GPU(duration=180, size="xlarge")
 def generate_stream(
     messages: list[dict],
     temperature: float,
@@ -123,7 +122,7 @@ def predict(
     top_p: float,
     max_tokens: int,
 ):
-    """Chat function — called both from UI and the auto-generated Gradio API."""
+    """Chat function - called both from UI and the auto-generated Gradio API."""
     messages = [{"role": "system", "content": system_prompt}]
     for user_msg, asst_msg in history:
         messages.append({"role": "user", "content": user_msg})
@@ -161,7 +160,7 @@ def chat_fn(
 
 
 # ---------------------------------------------------------------------------
-# Lang / theme helper
+# Helpers
 # ---------------------------------------------------------------------------
 LANGUAGES = ["python", "javascript", "typescript", "rust", "go", "java", "cpp",
              "csharp", "ruby", "php", "sql", "bash", "html", "css", "json", "yaml"]
@@ -169,10 +168,11 @@ LANGUAGES = ["python", "javascript", "typescript", "rust", "go", "java", "cpp",
 
 def build_examples():
     return [
-        ["Write a Python function that checks if a string is a palindrome."],
+        ["Write a Python async function that downloads a URL and retries 3 times on failure."],
         ["Create a Rust function that reads a CSV file and returns the row count."],
-        ["Explain the difference between an interface and a type in TypeScript."],
-        ["Write a Go HTTP server that serves static files on port 8080."],
+        ["Explain the difference between an interface and a type in TypeScript with examples."],
+        ["Write a Go HTTP server that serves static files on port 8080 with CORS support."],
+        ["Refactor this Python class to use dependency injection: class Database: ..."],
     ]
 
 
@@ -189,8 +189,8 @@ def create_ui():
         fill_width=True,
     ) as demo:
         gr.Markdown(
-            "# 💻 CodeCraft — AI Coding Assistant\n"
-            "Powered by **Qwen2.5-Coder-7B-Instruct** · ZeroGPU"
+            "# CodeCraft - AI Coding Assistant\n"
+            "Powered by **Qwen3-Coder-30B-A3B-Instruct** (MoE, 3B active) - ZeroGPU xlarge"
         )
 
         chatbot = gr.Chatbot(
@@ -211,7 +211,7 @@ def create_ui():
             submit_btn = gr.Button("Send", variant="primary", scale=1, min_width=80)
             clear_btn = gr.Button("Clear", scale=1, min_width=80)
 
-        with gr.Accordion("⚙️ Settings", open=False):
+        with gr.Accordion("Settings", open=False):
             with gr.Row():
                 system_prompt = gr.Textbox(
                     label="System Prompt",
@@ -230,7 +230,7 @@ def create_ui():
                     value=0.9, step=0.05,
                 )
                 max_tokens = gr.Slider(
-                    label="Max Tokens", minimum=128, maximum=4096,
+                    label="Max Tokens", minimum=128, maximum=8192,
                     value=2048, step=128,
                 )
 
@@ -254,37 +254,33 @@ def create_ui():
                 yield "", history, []
             yield "", history, [message]
 
-        # Wire submit via message box
         msg.submit(
             respond,
             inputs=[msg, history_state, system_prompt, temperature, top_p, max_tokens],
             outputs=[msg, chatbot, history_state],
-            concurrency_limit=8,
+            concurrency_limit=4,
             api_name="predict",
         )
         submit_btn.click(
             respond,
             inputs=[msg, history_state, system_prompt, temperature, top_p, max_tokens],
             outputs=[msg, chatbot, history_state],
-            concurrency_limit=8,
+            concurrency_limit=4,
             api_name=False,
         )
 
-        # Clear conversation
         def clear_conversation():
             return [], "", []
 
         clear_btn.click(
             clear_conversation,
             outputs=[history_state, chatbot, msg],
-            concurrency_limit=8,
+            concurrency_limit=4,
         )
 
-        # -- API endpoint exposure (auto by Gradio, but re-binding as top-level fn) --
         gr.Markdown(
             """
-            ### 📡 API
-
+            ### API
             This Space exposes a REST API at `/gradio_api/call/predict`.
             See the [Gradio docs](https://www.gradio.app/guides/sharing-your-app#api) for usage.
             """
@@ -295,5 +291,5 @@ def create_ui():
 
 if __name__ == "__main__":
     demo = create_ui()
-    demo.queue(default_concurrency_limit=8)
+    demo.queue(default_concurrency_limit=4)
     demo.launch()