Spaces:
Sleeping
Sleeping
Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,90 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr # Gradio for UI :contentReference[oaicite:3]{index=3}
|
2 |
+
from transformers import pipeline, set_seed # Transformers pipeline :contentReference[oaicite:4]{index=4}
|
3 |
+
|
4 |
+
# Set random seed reproducibility
|
5 |
+
def maybe_set_seed(seed: int):
|
6 |
+
if seed and seed > 0:
|
7 |
+
set_seed(seed)
|
8 |
+
|
9 |
+
# Load the model once per Space session
|
10 |
+
@gr.cache_resource # caches across sessions :contentReference[oaicite:5]{index=5}
|
11 |
+
def load_generator(model_name: str):
|
12 |
+
return pipeline(
|
13 |
+
"text-generation",
|
14 |
+
model=model_name,
|
15 |
+
trust_remote_code=True,
|
16 |
+
device_map="auto", # auto GPU/CPU placement :contentReference[oaicite:6]{index=6}
|
17 |
+
)
|
18 |
+
|
19 |
+
# The chat function called on each submit
|
20 |
+
def chat(
|
21 |
+
user_input: str,
|
22 |
+
history: list[tuple[str, str]],
|
23 |
+
model_name: str,
|
24 |
+
max_length: int,
|
25 |
+
temperature: float,
|
26 |
+
seed: int
|
27 |
+
):
|
28 |
+
maybe_set_seed(seed)
|
29 |
+
generator = load_generator(model_name)
|
30 |
+
# Prepare instruction-wrapped prompt
|
31 |
+
prompt = (
|
32 |
+
"[INST] <<SYS>>\nYou are a helpful assistant.\n<</SYS>>\n\n"
|
33 |
+
f"{user_input}\n[/INST]"
|
34 |
+
)
|
35 |
+
# Generate response
|
36 |
+
outputs = generator(
|
37 |
+
prompt,
|
38 |
+
max_length=max_length,
|
39 |
+
temperature=temperature,
|
40 |
+
do_sample=True,
|
41 |
+
num_return_sequences=1
|
42 |
+
)
|
43 |
+
raw = outputs[0]["generated_text"]
|
44 |
+
# Extract only assistant’s reply
|
45 |
+
response = raw.split("[/INST]")[-1].strip()
|
46 |
+
# Append to chat history
|
47 |
+
history.append((user_input, response))
|
48 |
+
return history, history
|
49 |
+
|
50 |
+
# Build the Gradio Blocks interface
|
51 |
+
with gr.Blocks() as demo:
|
52 |
+
gr.Markdown("## 🤖 Mistral-7B-Instruct Chatbot (Gradio)") # title :contentReference[oaicite:7]{index=7}
|
53 |
+
with gr.Row():
|
54 |
+
with gr.Column(scale=3):
|
55 |
+
chatbox = gr.Chatbot(label="Chat History")
|
56 |
+
with gr.Row():
|
57 |
+
inp = gr.Textbox(
|
58 |
+
placeholder="Type your message here...",
|
59 |
+
show_label=False,
|
60 |
+
lines=2
|
61 |
+
)
|
62 |
+
submit = gr.Button("Send")
|
63 |
+
with gr.Column(scale=1):
|
64 |
+
gr.Markdown("### Settings 🚀")
|
65 |
+
model_name = gr.Textbox(
|
66 |
+
label="Model name",
|
67 |
+
value="mistralai/Mistral-7B-Instruct-v0.3"
|
68 |
+
)
|
69 |
+
max_length = gr.Slider(
|
70 |
+
minimum=50, maximum=1024, step=50,
|
71 |
+
value=256, label="Max tokens"
|
72 |
+
)
|
73 |
+
temperature = gr.Slider(
|
74 |
+
minimum=0.0, maximum=1.0, step=0.05,
|
75 |
+
value=0.7, label="Temperature"
|
76 |
+
)
|
77 |
+
seed = gr.Number(
|
78 |
+
value=42, label="Random seed (0 disables)"
|
79 |
+
)
|
80 |
+
|
81 |
+
# Wire up the event
|
82 |
+
submit.click(
|
83 |
+
fn=chat,
|
84 |
+
inputs=[inp, chatbox, model_name, max_length, temperature, seed],
|
85 |
+
outputs=[chatbox, chatbox]
|
86 |
+
)
|
87 |
+
|
88 |
+
# Launch the app; in Spaces, no need to set share=True :contentReference[oaicite:8]{index=8}
|
89 |
+
if __name__ == "__main__":
|
90 |
+
demo.launch()
|