Upload app.py

app.py

@@ -0,0 +1,93 @@
+"""
+Minimal Gradio smoke-test for Qwen3-30B-A3B on ZeroGPU.
+No FastAPI, no custom routes — just enough to confirm the Space runs.
+"""
+
+# ── Patch HfFolder before importing gradio ────────────────────────────────────
+import huggingface_hub as _hf_hub
+
+if not hasattr(_hf_hub, "HfFolder"):
+    class _HfFolder:
+        _token = None
+        @classmethod
+        def get_token(cls):
+            try:
+                from huggingface_hub.utils import get_token
+                return get_token()
+            except Exception:
+                return cls._token
+        @classmethod
+        def save_token(cls, token): cls._token = token
+        @classmethod
+        def delete_token(cls): cls._token = None
+
+    _hf_hub.HfFolder = _HfFolder  # type: ignore[attr-defined]
+
+# ── Imports ───────────────────────────────────────────────────────────────────
+import gradio as gr
+import spaces
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
+from threading import Thread
+
+MODEL_ID = "Qwen/Qwen3-30B-A3B"
+
+_tok   = None
+_model = None
+
+def get_model():
+    global _tok, _model
+    if _model is None:
+        print("Loading model…")
+        _tok   = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
+        _model = AutoModelForCausalLM.from_pretrained(
+            MODEL_ID, torch_dtype=torch.bfloat16,
+            device_map="auto", trust_remote_code=True,
+        )
+        _model.eval()
+        print("Model ready ✓")
+    return _tok, _model
+
+
+@spaces.GPU(duration=120)
+def respond(message, history):
+    tok, model = get_model()
+
+    msgs = [{"role": h["role"], "content": h["content"]} for h in history]
+    msgs.append({"role": "user", "content": message})
+
+    text = tok.apply_chat_template(msgs, tokenize=False, add_generation_prompt=True)
+    inputs = tok(text, return_tensors="pt").to(model.device)
+
+    streamer = TextIteratorStreamer(tok, skip_prompt=True, skip_special_tokens=True)
+    t = Thread(
+        target=model.generate,
+        kwargs=dict(**inputs, max_new_tokens=2048, do_sample=True,
+                    temperature=0.7, top_p=0.9,
+                    pad_token_id=tok.eos_token_id, streamer=streamer),
+        daemon=True,
+    )
+    t.start()
+
+    output = ""
+    for chunk in streamer:
+        output += chunk
+        yield output
+    t.join()
+
+
+with gr.Blocks() as demo:
+    # Sign-in is required so ZeroGPU can recognize *your* Pro account from
+    # inside the embedded iframe and apply your 40-min daily quota instead
+    # of the anonymous-visitor quota (~3 min).
+    gr.LoginButton()
+
+    gr.ChatInterface(
+        fn=respond,
+        type="messages",
+        title="Qwen3-30B-A3B",
+        description="ZeroGPU-backed Qwen3-30B-A3B chat",
+    )
+
+if __name__ == "__main__":
+    demo.launch()