update model

app.py

@@ -1,7 +1,19 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
 import spaces
+import torch
+from transformers import AutoModelForCausalLM, AutoTokenizer
 
+# 只加载一次模型和分词器
+MODEL_NAME = "inclusionAI/Ring-mini-2.0"
+device = "cuda" if torch.cuda.is_available() else "cpu"
+tokenizer = AutoTokenizer.from_pretrained(MODEL_NAME, trust_remote_code=True)
+model = AutoModelForCausalLM.from_pretrained(
+    MODEL_NAME,
+    torch_dtype=torch.float16 if device == "cuda" else torch.float32,
+    trust_remote_code=True
+).to(device)
+
+@spaces.GPU
 def respond(
     message,
     history: list[dict[str, str]],
@@ -9,35 +21,39 @@ def respond(
     max_tokens,
     temperature,
     top_p,
-    hf_token: gr.OAuthToken,
+    hf_token: gr.OAuthToken = None,  # 保持参数兼容
 ):
     """
-    For more information on `huggingface_hub` Inference API support, please check the docs: https://huggingface.co/docs/huggingface_hub/v0.22.2/en/guides/inference
+    使用 transformers 在 GPU 上本地推理 inclusionAI/Ring-mini-2.0
     """
-    client = InferenceClient(token=hf_token.token, model="openai/gpt-oss-20b")
-
-    messages = [{"role": "system", "content": system_message}]
+    # 拼接历史和 system prompt，兼容 gradio ChatInterface 的消息格式
+    prompt = system_message + "\n"
+    # gradio history: [{"role": "user"/"assistant", "content": "..."}, ...]
+    last_role = None
+    for turn in history:
+        if turn.get("role") == "user":
+            prompt += f"User: {turn['content']}\n"
+            last_role = "user"
+        elif turn.get("role") == "assistant":
+            prompt += f"Assistant: {turn['content']}\n"
+            last_role = "assistant"
+    prompt += f"User: {message}\nAssistant:"
 
-    messages.extend(history)
-
-    messages.append({"role": "user", "content": message})
-
-    response = ""
-
-    for message in client.chat_completion(
-        messages,
-        max_tokens=max_tokens,
-        stream=True,
+    input_ids = tokenizer(prompt, return_tensors="pt").input_ids.to(device)
+    output_ids = model.generate(
+        input_ids,
+        max_new_tokens=max_tokens,
         temperature=temperature,
         top_p=top_p,
-    ):
-        choices = message.choices
-        token = ""
-        if len(choices) and choices[0].delta.content:
-            token = choices[0].delta.content
-
-        response += token
-        yield response
+        do_sample=True,
+        pad_token_id=tokenizer.eos_token_id,
+    )
+    output = tokenizer.decode(output_ids[0][input_ids.shape[1]:], skip_special_tokens=True)
+    # 流式输出
+    response = ""
+    for token in output.split():
+        response += token + " "
+        yield response.strip()
 
 
 """
@@ -61,6 +77,7 @@ chatbot = gr.ChatInterface(
 )
 
 with gr.Blocks() as demo:
+    gr.Markdown("# HuggingFace Running")
     with gr.Sidebar():
         gr.LoginButton()
     chatbot.render()
@@ -68,8 +85,3 @@ with gr.Blocks() as demo:
 
 if __name__ == "__main__":
     demo.launch()
-
-# 在这里写个 @spaces.GPU 函数，蒙混过静态规则
-@spaces.GPU
-def foo():
-    pass

requirements.txt

@@ -1,2 +1,4 @@
 gradio
 huggingface-hub
+transformers
+accelerate