Create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,88 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import argparse
|
2 |
+
import gradio as gr
|
3 |
+
from openai import OpenAI
|
4 |
+
|
5 |
+
# Argument parser setup
|
6 |
+
parser = argparse.ArgumentParser(
|
7 |
+
description='Chatbot Interface with Customizable Parameters')
|
8 |
+
parser.add_argument('--model-url',
|
9 |
+
type=str,
|
10 |
+
default='http://134.28.190.100:8000/v1',
|
11 |
+
help='Model URL')
|
12 |
+
parser.add_argument('-m',
|
13 |
+
'--model',
|
14 |
+
type=str,
|
15 |
+
required=True,
|
16 |
+
default='TheBloke/Mistral-7B-Instruct-v0.2-AWQ',
|
17 |
+
help='Model name for the chatbot')
|
18 |
+
parser.add_argument('--temp',
|
19 |
+
type=float,
|
20 |
+
default=0.8,
|
21 |
+
help='Temperature for text generation')
|
22 |
+
parser.add_argument('--stop-token-ids',
|
23 |
+
type=str,
|
24 |
+
default='',
|
25 |
+
help='Comma-separated stop token IDs')
|
26 |
+
parser.add_argument("--host", type=str, default=None)
|
27 |
+
parser.add_argument("--port", type=int, default=8001)
|
28 |
+
|
29 |
+
# Parse the arguments
|
30 |
+
args = parser.parse_args()
|
31 |
+
|
32 |
+
# Set OpenAI's API key and API base to use vLLM's API server.
|
33 |
+
openai_api_key = "EMPTY"
|
34 |
+
openai_api_base = args.model_url
|
35 |
+
|
36 |
+
# Create an OpenAI client to interact with the API server
|
37 |
+
client = OpenAI(
|
38 |
+
api_key=openai_api_key,
|
39 |
+
base_url=openai_api_base,
|
40 |
+
)
|
41 |
+
|
42 |
+
# def add_document():
|
43 |
+
|
44 |
+
def predict(message, history):
|
45 |
+
# Convert chat history to OpenAI format
|
46 |
+
history_openai_format = []#[{
|
47 |
+
#"role": "system",
|
48 |
+
#"content": "You are a great ai assistant."
|
49 |
+
#}]
|
50 |
+
for human, assistant in history:
|
51 |
+
history_openai_format.append({"role": "user", "content": human})
|
52 |
+
history_openai_format.append({
|
53 |
+
"role": "assistant",
|
54 |
+
"content": assistant
|
55 |
+
})
|
56 |
+
history_openai_format.append({"role": "user", "content": message})
|
57 |
+
|
58 |
+
# Create a chat completion request and send it to the API server
|
59 |
+
stream = client.chat.completions.create(
|
60 |
+
model=args.model, # Model name to use
|
61 |
+
messages=history_openai_format, # Chat history
|
62 |
+
temperature=args.temp, # Temperature for text generation
|
63 |
+
stream=True, # Stream response
|
64 |
+
extra_body={
|
65 |
+
'repetition_penalty':
|
66 |
+
1,
|
67 |
+
'stop_token_ids': [
|
68 |
+
int(id.strip()) for id in args.stop_token_ids.split(',')
|
69 |
+
if id.strip()
|
70 |
+
] if args.stop_token_ids else []
|
71 |
+
})
|
72 |
+
|
73 |
+
# Read and return generated text from response stream
|
74 |
+
partial_message = ""
|
75 |
+
for chunk in stream:
|
76 |
+
partial_message += (chunk.choices[0].delta.content or "")
|
77 |
+
yield partial_message
|
78 |
+
|
79 |
+
with gr.Blocks(title="MethodAI 0.15", theme="Soft") as demo:
|
80 |
+
with gr.Row():
|
81 |
+
with gr.Column(scale=1):
|
82 |
+
gr.UploadButton("Click to upload PDFs",file_types=[".pdf"])
|
83 |
+
with gr.Column(scale=4):
|
84 |
+
# Create and launch a chat interface with Gradio
|
85 |
+
gr.ChatInterface(predict).queue()
|
86 |
+
# with demo:
|
87 |
+
# btn.upload(render_file, inputs=[btn], outputs=[show_img])
|
88 |
+
demo.launch(server_name=args.host, server_port=args.port, share=True)
|