SnifferCaptain commited on
Commit
ea70086
·
verified ·
1 Parent(s): 419a41d

Update app.py

Browse files
Files changed (1) hide show
  1. app.py +169 -56
app.py CHANGED
@@ -2,68 +2,181 @@ import gradio as gr
2
  from huggingface_hub import InferenceClient
3
 
4
 
 
 
 
5
  def respond(
6
- message,
7
- history: list[dict[str, str]],
8
- system_message,
9
- max_tokens,
10
- temperature,
11
- top_p,
12
- hf_token: gr.OAuthToken,
 
13
  ):
14
- """
15
- 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
16
- """
17
- client = InferenceClient(token=hf_token.token, model="openai/gpt-oss-20b")
18
-
19
- messages = [{"role": "system", "content": system_message}]
20
-
21
- messages.extend(history)
22
-
23
- messages.append({"role": "user", "content": message})
24
-
25
- response = ""
26
-
27
- for message in client.chat_completion(
28
- messages,
29
- max_tokens=max_tokens,
30
- stream=True,
31
- temperature=temperature,
32
- top_p=top_p,
33
- ):
34
- choices = message.choices
35
- token = ""
36
- if len(choices) and choices[0].delta.content:
37
- token = choices[0].delta.content
38
-
39
- response += token
40
- yield response
41
-
42
-
43
- """
44
- For information on how to customize the ChatInterface, peruse the gradio docs: https://www.gradio.app/docs/chatinterface
45
- """
46
- chatbot = gr.ChatInterface(
47
- respond,
48
- additional_inputs=[
49
- gr.Textbox(value="You are a friendly Chatbot.", label="System message"),
50
- gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"),
51
- gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"),
52
- gr.Slider(
53
- minimum=0.1,
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
54
  maximum=1.0,
55
  value=0.95,
56
  step=0.05,
57
- label="Top-p (nucleus sampling)",
58
- ),
59
- ],
60
- )
61
 
62
- with gr.Blocks() as demo:
63
- with gr.Sidebar():
64
- gr.LoginButton()
65
- chatbot.render()
 
 
 
 
 
 
 
 
 
 
 
 
66
 
67
 
68
  if __name__ == "__main__":
69
- demo.launch()
 
 
 
 
2
  from huggingface_hub import InferenceClient
3
 
4
 
5
+ MODEL_ID = "openai/gpt-oss-20b"
6
+
7
+
8
  def respond(
9
+ message: str,
10
+ history: list[dict],
11
+ system_message: str,
12
+ reasoning_effort: str,
13
+ max_tokens: int,
14
+ temperature: float,
15
+ top_p: float,
16
+ hf_token: gr.OAuthToken | None,
17
  ):
18
+ if hf_token is None or not getattr(hf_token, "token", None):
19
+ yield "请先点击左侧的“Sign in with Hugging Face”登录,然后再发送消息。"
20
+ return
21
+
22
+ client = InferenceClient(
23
+ model=MODEL_ID,
24
+ provider="auto",
25
+ token=hf_token.token,
26
+ )
27
+
28
+ messages = []
29
+
30
+ if system_message.strip():
31
+ messages.append(
32
+ {
33
+ "role": "system",
34
+ "content": system_message.strip(),
35
+ }
36
+ )
37
+
38
+ # Gradio messages 模式下,history 通常是:
39
+ # [{"role": "user", "content": "..."}, ...]
40
+ for item in history:
41
+ role = item.get("role")
42
+ content = item.get("content")
43
+
44
+ if role in {"user", "assistant", "system"} and isinstance(content, str):
45
+ messages.append(
46
+ {
47
+ "role": role,
48
+ "content": content,
49
+ }
50
+ )
51
+
52
+ messages.append(
53
+ {
54
+ "role": "user",
55
+ "content": message,
56
+ }
57
+ )
58
+
59
+ response_text = ""
60
+
61
+ try:
62
+ stream = client.chat_completion(
63
+ messages=messages,
64
+ max_tokens=int(max_tokens),
65
+ temperature=float(temperature),
66
+ top_p=float(top_p),
67
+ stream=True,
68
+
69
+ # 转交给兼容 OpenAI 风格的推理后端。
70
+ # 若当前 Provider 不接受该字段,删除 extra_body 即可。
71
+ extra_body={
72
+ "reasoning_effort": reasoning_effort,
73
+ },
74
+ )
75
+
76
+ for chunk in stream:
77
+ if not chunk.choices:
78
+ continue
79
+
80
+ delta = chunk.choices[0].delta
81
+ token = getattr(delta, "content", None)
82
+
83
+ if token:
84
+ response_text += token
85
+ yield response_text
86
+
87
+ if not response_text:
88
+ yield "模型没有返回可显示的文本。请稍后重试或更换推理参数。"
89
+
90
+ except Exception as error:
91
+ error_text = str(error)
92
+
93
+ if "401" in error_text or "unauthorized" in error_text.lower():
94
+ yield (
95
+ "身份验证失败。请退出后重新登录 Hugging Face,"
96
+ "并确认账户允许调用 Inference Providers。"
97
+ )
98
+ elif "402" in error_text or "payment" in error_text.lower():
99
+ yield (
100
+ "当前 Hugging Face 账户的推理额度不足,"
101
+ "请检查 Inference Providers 余额或计费设置。"
102
+ )
103
+ elif "429" in error_text or "rate limit" in error_text.lower():
104
+ yield "请求过于频繁或免费额度已达到限制,请稍后再试。"
105
+ else:
106
+ yield f"调用模型时发生错误:{error_text}"
107
+
108
+
109
+ with gr.Blocks(title="GPT-OSS 20B Chat") as demo:
110
+ gr.Markdown(
111
+ """
112
+ # GPT-OSS 20B Chat
113
+
114
+ 使用 Hugging Face Inference Providers 调用 `openai/gpt-oss-20b`。
115
+
116
+ 请先登录 Hugging Face。推理请求将使用登录用户自己的 HF 账户和额度。
117
+ """
118
+ )
119
+
120
+ with gr.Sidebar():
121
+ gr.Markdown("### Hugging Face 账户")
122
+ gr.LoginButton()
123
+
124
+ system_message = gr.Textbox(
125
+ value="You are a helpful and friendly assistant.",
126
+ label="System message",
127
+ lines=4,
128
+ )
129
+
130
+ reasoning_effort = gr.Radio(
131
+ choices=["low", "medium", "high"],
132
+ value="medium",
133
+ label="Reasoning effort",
134
+ )
135
+
136
+ max_tokens = gr.Slider(
137
+ minimum=64,
138
+ maximum=8192,
139
+ value=1024,
140
+ step=64,
141
+ label="Max output tokens",
142
+ )
143
+
144
+ temperature = gr.Slider(
145
+ minimum=0.0,
146
+ maximum=2.0,
147
+ value=0.7,
148
+ step=0.05,
149
+ label="Temperature",
150
+ )
151
+
152
+ top_p = gr.Slider(
153
+ minimum=0.05,
154
  maximum=1.0,
155
  value=0.95,
156
  step=0.05,
157
+ label="Top-p",
158
+ )
 
 
159
 
160
+ gr.ChatInterface(
161
+ fn=respond,
162
+ type="messages",
163
+ additional_inputs=[
164
+ system_message,
165
+ reasoning_effort,
166
+ max_tokens,
167
+ temperature,
168
+ top_p,
169
+ ],
170
+ examples=[
171
+ ["请解释 MoE 模型中的总参数和激活参数有什么区别。"],
172
+ ["用 Python 写一个并发批量请求 API 的示例。"],
173
+ ],
174
+ cache_examples=False,
175
+ )
176
 
177
 
178
  if __name__ == "__main__":
179
+ demo.queue(
180
+ default_concurrency_limit=8,
181
+ max_size=32,
182
+ ).launch()