Update app.py
Browse files
app.py
CHANGED
@@ -1,3 +1,103 @@
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
|
3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
2.77 kB
|
2 |
+
from huggingface_hub import InferenceClient
|
3 |
import gradio as gr
|
4 |
|
5 |
+
client = InferenceClient(
|
6 |
+
"mistralai/Mistral-7B-Instruct-v0.1"
|
7 |
+
)
|
8 |
+
|
9 |
+
|
10 |
+
def format_prompt(message, history):
|
11 |
+
prompt = "<s>"
|
12 |
+
for user_prompt, bot_response in history:
|
13 |
+
prompt += f"[INST] {user_prompt} [/INST]"
|
14 |
+
prompt += f" {bot_response}</s> "
|
15 |
+
prompt += f"[INST] {message} [/INST]"
|
16 |
+
return prompt
|
17 |
+
|
18 |
+
def generate(
|
19 |
+
prompt, history, temperature=0.9, max_new_tokens=256, top_p=0.95, repetition_penalty=1.0,
|
20 |
+
):
|
21 |
+
temperature = float(temperature)
|
22 |
+
if temperature < 1e-2:
|
23 |
+
temperature = 1e-2
|
24 |
+
top_p = float(top_p)
|
25 |
+
|
26 |
+
generate_kwargs = dict(
|
27 |
+
temperature=temperature,
|
28 |
+
max_new_tokens=max_new_tokens,
|
29 |
+
top_p=top_p,
|
30 |
+
repetition_penalty=repetition_penalty,
|
31 |
+
do_sample=True,
|
32 |
+
seed=42,
|
33 |
+
)
|
34 |
+
|
35 |
+
formatted_prompt = format_prompt(prompt, history)
|
36 |
+
|
37 |
+
stream = client.text_generation(formatted_prompt, **generate_kwargs, stream=True, details=True, return_full_text=False)
|
38 |
+
output = ""
|
39 |
+
|
40 |
+
for response in stream:
|
41 |
+
output += response.token.text
|
42 |
+
yield output
|
43 |
+
return output
|
44 |
+
|
45 |
+
|
46 |
+
additional_inputs=[
|
47 |
+
gr.Slider(
|
48 |
+
label="Temperature",
|
49 |
+
value=0.9,
|
50 |
+
minimum=0.0,
|
51 |
+
maximum=1.0,
|
52 |
+
step=0.05,
|
53 |
+
interactive=True,
|
54 |
+
info="Higher values produce more diverse outputs",
|
55 |
+
),
|
56 |
+
gr.Slider(
|
57 |
+
label="Max new tokens",
|
58 |
+
value=256,
|
59 |
+
minimum=0,
|
60 |
+
maximum=1048,
|
61 |
+
step=64,
|
62 |
+
interactive=True,
|
63 |
+
info="The maximum numbers of new tokens",
|
64 |
+
),
|
65 |
+
gr.Slider(
|
66 |
+
label="Top-p (nucleus sampling)",
|
67 |
+
value=0.90,
|
68 |
+
minimum=0.0,
|
69 |
+
maximum=1,
|
70 |
+
step=0.05,
|
71 |
+
interactive=True,
|
72 |
+
info="Higher values sample more low-probability tokens",
|
73 |
+
),
|
74 |
+
gr.Slider(
|
75 |
+
label="Repetition penalty",
|
76 |
+
value=1.2,
|
77 |
+
minimum=1.0,
|
78 |
+
maximum=2.0,
|
79 |
+
step=0.05,
|
80 |
+
interactive=True,
|
81 |
+
info="Penalize repeated tokens",
|
82 |
+
)
|
83 |
+
]
|
84 |
+
|
85 |
+
css = """
|
86 |
+
#mkd {
|
87 |
+
height: 500px;
|
88 |
+
overflow: auto;
|
89 |
+
border: 1px solid #ccc;
|
90 |
+
}
|
91 |
+
"""
|
92 |
+
|
93 |
+
with gr.Blocks(css=css) as demo:
|
94 |
+
gr.HTML("<h1><center>Mistral 7B Instruct<h1><center>")
|
95 |
+
gr.HTML("<h3><center>In this demo, you can chat with <a href='https://huggingface.co/mistralai/Mistral-7B-Instruct-v0.1'>Mistral-7B-Instruct</a> model. π¬<h3><center>")
|
96 |
+
gr.HTML("<h3><center>Learn more about the model <a href='https://huggingface.co/docs/transformers/main/model_doc/mistral'>here</a>. π<h3><center>")
|
97 |
+
gr.ChatInterface(
|
98 |
+
generate,
|
99 |
+
additional_inputs=additional_inputs,
|
100 |
+
examples=[["What is the secret to life?"], ["Write me a recipe for pancakes."]]
|
101 |
+
)
|
102 |
+
|
103 |
+
demo.queue(concurrency_count=75, max_size=100).launch(debug=True)
|