Deploy Qwen2.5 Coder 3B Gradio app

README.md

@@ -1,13 +1,17 @@
----
-title: Devcoder Ai
-emoji: 📈
-colorFrom: pink
-colorTo: yellow
-sdk: gradio
-sdk_version: 6.30.0
-python_version: '3.12'
-app_file: app.py
-pinned: false
----
-
-Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
+---
+title: Qwen2.5 Coder 3B
+emoji: 💻
+colorFrom: blue
+colorTo: purple
+sdk: gradio
+python_version: 3.12
+app_file: app.py
+pinned: false
+---
+
+# Qwen2.5 Coder 3B
+
+A coding assistant using Qwen2.5-Coder-3B-Instruct with Hugging Face ZeroGPU.
+The model is downloaded from Hugging Face on first startup. Inference requests
+use the largest ZeroGPU allocation; xlarge requests consume twice the daily
+GPU quota of the default large size.

app.py

@@ -0,0 +1,114 @@
+import os
+
+import gradio as gr
+import spaces
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
+
+
+MODEL_ID = "Qwen/Qwen2.5-Coder-3B-Instruct"
+MAX_INPUT_TOKENS = 8192
+MAX_NEW_TOKENS = 512
+DEFAULT_SYSTEM_PROMPT = (
+    "You are a helpful coding assistant. "
+    "Give accurate, clear answers and use code examples when useful."
+)
+
+torch.set_num_threads(min(4, os.cpu_count() or 1))
+
+tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_ID,
+    torch_dtype=torch.float16,
+    low_cpu_mem_usage=True,
+    attn_implementation="sdpa",
+)
+model.to("cuda")
+model.eval()
+
+
+def normalize_history(history):
+    messages = []
+
+    for item in history or []:
+        if isinstance(item, dict):
+            role = item.get("role")
+            content = item.get("content")
+            if role in ("user", "assistant") and isinstance(content, str):
+                messages.append({"role": role, "content": content})
+        elif isinstance(item, (list, tuple)) and len(item) == 2:
+            user_message, assistant_message = item
+            if isinstance(user_message, str):
+                messages.append({"role": "user", "content": user_message})
+            if isinstance(assistant_message, str):
+                messages.append({"role": "assistant", "content": assistant_message})
+
+    return messages
+
+
+def prepare_inputs(system_prompt, messages):
+    chat_messages = [{"role": "system", "content": system_prompt}, *messages]
+    encoded = tokenizer.apply_chat_template(
+        chat_messages,
+        tokenize=True,
+        add_generation_prompt=True,
+        return_tensors="pt",
+        return_dict=True,
+    )
+    return encoded
+
+
+@spaces.GPU(size="xlarge", duration=60)
+def generate(message, history, system_prompt):
+    system_prompt = (
+        DEFAULT_SYSTEM_PROMPT if system_prompt is None else system_prompt
+    )
+    messages = normalize_history(history)
+    messages.append({"role": "user", "content": message})
+
+    inputs = prepare_inputs(system_prompt, messages)
+    while inputs["input_ids"].shape[-1] > MAX_INPUT_TOKENS:
+        if len(messages) <= 1:
+            raise gr.Error(
+                "This message or system prompt is too long. Shorten it and try again."
+            )
+        messages = messages[2:] if len(messages) > 2 else messages[1:]
+        inputs = prepare_inputs(system_prompt, messages)
+
+    inputs = {key: value.to("cuda") for key, value in inputs.items()}
+    with torch.inference_mode():
+        output = model.generate(
+            **inputs,
+            max_new_tokens=MAX_NEW_TOKENS,
+            do_sample=True,
+            temperature=0.2,
+            top_p=0.95,
+            repetition_penalty=1.05,
+            pad_token_id=tokenizer.eos_token_id,
+            eos_token_id=tokenizer.eos_token_id,
+        )
+
+    new_tokens = output[0, inputs["input_ids"].shape[-1]:]
+    return tokenizer.decode(new_tokens, skip_special_tokens=True).strip()
+
+
+system_prompt_input = gr.Textbox(
+    label="System prompt",
+    value=DEFAULT_SYSTEM_PROMPT,
+    lines=4,
+)
+
+demo = gr.ChatInterface(
+    fn=generate,
+    chatbot=gr.Chatbot(type="messages", height=500),
+    additional_inputs=[system_prompt_input],
+    title="Qwen2.5 Coder 3B",
+    description=(
+        "A coding assistant powered by Qwen2.5-Coder-3B on Hugging Face ZeroGPU."
+    ),
+)
+
+demo.queue()
+
+if __name__ == "__main__":
+    demo.launch()

deploy.py

@@ -0,0 +1,81 @@
+import argparse
+import json
+from pathlib import Path
+
+from huggingface_hub import HfApi
+
+
+ROOT = Path(__file__).resolve().parent
+CREDENTIALS_FILE = ROOT / "credentials.json"
+
+
+def read_token():
+    if not CREDENTIALS_FILE.is_file():
+        raise FileNotFoundError(
+            f"Missing {CREDENTIALS_FILE}. Add your Hugging Face access token "
+            'as {"token": "hf_..."} in that file.'
+        )
+
+    with CREDENTIALS_FILE.open(encoding="utf-8") as credentials_file:
+        credentials = json.load(credentials_file)
+
+    token = credentials.get("token")
+    if not isinstance(token, str) or not token.strip():
+        raise ValueError(
+            f"No token found in {CREDENTIALS_FILE}. Add it as "
+            '{"token": "hf_..."} and try again.'
+        )
+    return token.strip()
+
+
+def main():
+    parser = argparse.ArgumentParser(
+        description="Create a Hugging Face Gradio Space and upload this app."
+    )
+    parser.add_argument(
+        "--name",
+        default="devcoder-ai",
+        help="Space name (default: devcoder-ai)",
+    )
+    parser.add_argument(
+        "--private",
+        action="store_true",
+        help="Create the Space as private (public by default).",
+    )
+    args = parser.parse_args()
+
+    api = HfApi(token=read_token())
+    account = api.whoami()
+    username = account.get("name")
+    if not username:
+        raise RuntimeError("Could not determine the Hugging Face account name.")
+
+    repo_id = f"{username}/{args.name}"
+    api.create_repo(
+        repo_id=repo_id,
+        repo_type="space",
+        space_sdk="gradio",
+        space_hardware="zero-a10g",
+        private=args.private,
+        exist_ok=True,
+    )
+    print(f"Created Space: https://huggingface.co/spaces/{repo_id}")
+
+    result = api.upload_folder(
+        repo_id=repo_id,
+        repo_type="space",
+        folder_path=str(ROOT),
+        ignore_patterns=[
+            ".gitignore",
+            "credentials.json",
+            "__pycache__/**",
+            "*.py[cod]",
+        ],
+        commit_message="Deploy Qwen2.5 Coder 3B Gradio app",
+    )
+    print(f"Uploaded app. Deployment status: https://huggingface.co/spaces/{repo_id}")
+    print(f"Upload commit: {result.commit_url}")
+
+
+if __name__ == "__main__":
+    main()

requirements.txt

@@ -0,0 +1,6 @@
+torch==2.8.0
+transformers>=4.48.0,<5.0.0
+accelerate>=1.2.0,<2.0.0
+spaces>=0.37.0
+gradio>=5.0,<6.0
+huggingface_hub>=0.26.0,<1.0