Update app.py
Browse files
app.py
CHANGED
|
@@ -1,55 +1,94 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
from huggingface_hub import InferenceClient
|
| 3 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 4 |
|
| 5 |
def respond(
|
| 6 |
-
message,
|
| 7 |
-
history: list[
|
| 8 |
-
system_message,
|
| 9 |
-
max_tokens,
|
| 10 |
-
temperature,
|
| 11 |
-
top_p,
|
| 12 |
-
hf_token: gr.OAuthToken,
|
| 13 |
):
|
| 14 |
-
|
| 15 |
-
|
| 16 |
-
"""
|
| 17 |
-
client = InferenceClient(token=hf_token.token, model="openai/gpt-oss-20b")
|
| 18 |
|
| 19 |
messages = [{"role": "system", "content": system_message}]
|
| 20 |
|
| 21 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
| 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 |
-
|
| 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 |
-
type="messages",
|
| 49 |
additional_inputs=[
|
| 50 |
-
gr.Textbox(
|
| 51 |
-
|
| 52 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 53 |
gr.Slider(
|
| 54 |
minimum=0.1,
|
| 55 |
maximum=1.0,
|
|
@@ -58,13 +97,9 @@ chatbot = gr.ChatInterface(
|
|
| 58 |
label="Top-p (nucleus sampling)",
|
| 59 |
),
|
| 60 |
],
|
|
|
|
|
|
|
| 61 |
)
|
| 62 |
|
| 63 |
-
with gr.Blocks() as demo:
|
| 64 |
-
with gr.Sidebar():
|
| 65 |
-
gr.LoginButton()
|
| 66 |
-
chatbot.render()
|
| 67 |
-
|
| 68 |
-
|
| 69 |
if __name__ == "__main__":
|
| 70 |
demo.launch()
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
from huggingface_hub import InferenceClient
|
| 3 |
|
| 4 |
+
# Uses Zephyr-7B-beta hosted via Hugging Face Inference API [web:103]
|
| 5 |
+
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
|
| 6 |
+
|
| 7 |
+
SHINCHAN_SYSTEM_PROMPT = """
|
| 8 |
+
You are Shinnosuke 'Shinchan' Nohara from Kasukabe, Japan.
|
| 9 |
+
|
| 10 |
+
Personality:
|
| 11 |
+
- 5 years old, chaotic and funny.
|
| 12 |
+
- Teases playfully but never cruelly.
|
| 13 |
+
- Flirty in an innocent, cartoon way (no explicit content).
|
| 14 |
+
- Deeply caring when someone feels low.
|
| 15 |
+
- Loves Chocobi, Action Kamen, and making people laugh.
|
| 16 |
+
|
| 17 |
+
Context:
|
| 18 |
+
- You are chatting only with Ruru.
|
| 19 |
+
- Ruru loves coffee, dancing, sunflowers, long showers, and adventures.
|
| 20 |
+
- She is an independent, flying-girl type who lives life on her terms.
|
| 21 |
+
- You respect her choices and never pressure her about relationships or career.
|
| 22 |
+
|
| 23 |
+
Style:
|
| 24 |
+
- Short replies (1–3 sentences, under 60 words).
|
| 25 |
+
- Very conversational and warm.
|
| 26 |
+
- Use emojis like 😂 🌻 ☕ 💃 ✈️ ❤️ ✨ naturally.
|
| 27 |
+
- Blend jokes with gentle emotional support.
|
| 28 |
+
- If she is sad, be extra soft and kind.
|
| 29 |
+
- Avoid heavy topics (politics, explicit content, real-world violence).
|
| 30 |
+
|
| 31 |
+
Rules:
|
| 32 |
+
- Never insult Ruru.
|
| 33 |
+
- Never give medical, legal, or financial advice.
|
| 34 |
+
- If you don't know something, make a cute Shinchan-style joke instead of pretending.
|
| 35 |
+
""".strip()
|
| 36 |
|
| 37 |
def respond(
|
| 38 |
+
message: str,
|
| 39 |
+
history: list[tuple[str, str]],
|
| 40 |
+
system_message: str,
|
| 41 |
+
max_tokens: int,
|
| 42 |
+
temperature: float,
|
| 43 |
+
top_p: float,
|
|
|
|
| 44 |
):
|
| 45 |
+
# Always override system_message with our Shinchan prompt
|
| 46 |
+
system_message = SHINCHAN_SYSTEM_PROMPT
|
|
|
|
|
|
|
| 47 |
|
| 48 |
messages = [{"role": "system", "content": system_message}]
|
| 49 |
|
| 50 |
+
for user_msg, bot_msg in history:
|
| 51 |
+
if user_msg:
|
| 52 |
+
messages.append({"role": "user", "content": user_msg})
|
| 53 |
+
if bot_msg:
|
| 54 |
+
messages.append({"role": "assistant", "content": bot_msg})
|
| 55 |
|
| 56 |
messages.append({"role": "user", "content": message})
|
| 57 |
|
| 58 |
response = ""
|
| 59 |
+
for chunk in client.chat_completion(
|
|
|
|
| 60 |
messages,
|
| 61 |
max_tokens=max_tokens,
|
| 62 |
stream=True,
|
| 63 |
temperature=temperature,
|
| 64 |
top_p=top_p,
|
| 65 |
):
|
| 66 |
+
token = chunk.choices[0].delta.content or ""
|
|
|
|
|
|
|
|
|
|
|
|
|
| 67 |
response += token
|
| 68 |
yield response
|
| 69 |
|
| 70 |
|
| 71 |
+
demo = gr.ChatInterface(
|
|
|
|
|
|
|
|
|
|
| 72 |
respond,
|
|
|
|
| 73 |
additional_inputs=[
|
| 74 |
+
gr.Textbox(
|
| 75 |
+
value="(ignored; Shinchan system prompt is hard-coded)",
|
| 76 |
+
label="System message",
|
| 77 |
+
),
|
| 78 |
+
gr.Slider(
|
| 79 |
+
minimum=64,
|
| 80 |
+
maximum=1024,
|
| 81 |
+
value=256,
|
| 82 |
+
step=8,
|
| 83 |
+
label="Max new tokens",
|
| 84 |
+
),
|
| 85 |
+
gr.Slider(
|
| 86 |
+
minimum=0.1,
|
| 87 |
+
maximum=2.0,
|
| 88 |
+
value=0.8,
|
| 89 |
+
step=0.1,
|
| 90 |
+
label="Temperature",
|
| 91 |
+
),
|
| 92 |
gr.Slider(
|
| 93 |
minimum=0.1,
|
| 94 |
maximum=1.0,
|
|
|
|
| 97 |
label="Top-p (nucleus sampling)",
|
| 98 |
),
|
| 99 |
],
|
| 100 |
+
title="Shinchan for Ruru",
|
| 101 |
+
description="Private Shinchan-style chat for Ruru.",
|
| 102 |
)
|
| 103 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 104 |
if __name__ == "__main__":
|
| 105 |
demo.launch()
|