Spaces:
Running
Running
Lohia, Aditya
commited on
Commit
•
a9409d4
1
Parent(s):
12e4d9f
Updated Spaces
Browse files- app.py +147 -0
- dialog.py +45 -0
- gateway.py +90 -0
- requirements.txt +5 -0
app.py
ADDED
@@ -0,0 +1,147 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import os
|
2 |
+
import gradio as gr
|
3 |
+
from typing import Iterator
|
4 |
+
|
5 |
+
from dialog import get_dialog_box
|
6 |
+
from gateway import check_server_health, request_generation
|
7 |
+
|
8 |
+
# CONSTANTS
|
9 |
+
MAX_NEW_TOKENS: int = 2048
|
10 |
+
|
11 |
+
# GET ENVIRONMENT VARIABLES
|
12 |
+
CLOUD_GATEWAY_API = os.getenv("API_ENDPOINT")
|
13 |
+
|
14 |
+
|
15 |
+
def toggle_ui():
|
16 |
+
"""
|
17 |
+
Function to toggle the visibility of the UI based on the server health
|
18 |
+
Returns:
|
19 |
+
hide/show main ui/dialog
|
20 |
+
"""
|
21 |
+
health = check_server_health(cloud_gateway_api=CLOUD_GATEWAY_API)
|
22 |
+
if health:
|
23 |
+
return gr.update(visible=True), gr.update(visible=False) # Show main UI, hide dialog
|
24 |
+
else:
|
25 |
+
return gr.update(visible=False), gr.update(visible=True) # Hide main UI, show dialog
|
26 |
+
|
27 |
+
|
28 |
+
def generate(
|
29 |
+
message: str,
|
30 |
+
chat_history: list,
|
31 |
+
system_prompt: str,
|
32 |
+
max_new_tokens: int = 1024,
|
33 |
+
temperature: float = 0.6,
|
34 |
+
top_p: float = 0.9,
|
35 |
+
top_k: int = 50,
|
36 |
+
repetition_penalty: float = 1.2,
|
37 |
+
) -> Iterator[str]:
|
38 |
+
"""Send a request to backend, fetch the streaming responses and emit to the UI.
|
39 |
+
|
40 |
+
Args:
|
41 |
+
message (str): input message from the user
|
42 |
+
chat_history (list[tuple[str, str]]): entire chat history of the session
|
43 |
+
system_prompt (str): system prompt
|
44 |
+
max_new_tokens (int, optional): maximum number of tokens to generate, ignoring the number of tokens in the
|
45 |
+
prompt. Defaults to 1024.
|
46 |
+
temperature (float, optional): the value used to module the next token probabilities. Defaults to 0.6.
|
47 |
+
top_p (float, optional): if set to float<1, only the smallest set of most probable tokens with probabilities
|
48 |
+
that add up to top_p or higher are kept for generation. Defaults to 0.9.
|
49 |
+
top_k (int, optional): the number of highest probability vocabulary tokens to keep for top-k-filtering.
|
50 |
+
Defaults to 50.
|
51 |
+
repetition_penalty (float, optional): the parameter for repetition penalty. 1.0 means no penalty.
|
52 |
+
Defaults to 1.2.
|
53 |
+
|
54 |
+
Yields:
|
55 |
+
Iterator[str]: Streaming responses to the UI
|
56 |
+
"""
|
57 |
+
# sample method to yield responses from the llm model
|
58 |
+
outputs = []
|
59 |
+
for text in request_generation(message=message,
|
60 |
+
system_prompt=system_prompt,
|
61 |
+
max_new_tokens=max_new_tokens,
|
62 |
+
temperature=temperature,
|
63 |
+
top_p=top_p,
|
64 |
+
top_k=top_k,
|
65 |
+
repetition_penalty=repetition_penalty,
|
66 |
+
cloud_gateway_api=CLOUD_GATEWAY_API):
|
67 |
+
outputs.append(text)
|
68 |
+
yield "".join(outputs)
|
69 |
+
|
70 |
+
|
71 |
+
chat_interface = gr.ChatInterface(
|
72 |
+
fn=generate,
|
73 |
+
additional_inputs=[
|
74 |
+
gr.Textbox(label="System prompt", lines=6),
|
75 |
+
gr.Slider(
|
76 |
+
label="Max New Tokens",
|
77 |
+
minimum=1,
|
78 |
+
maximum=MAX_NEW_TOKENS,
|
79 |
+
step=1,
|
80 |
+
value=1024,
|
81 |
+
),
|
82 |
+
gr.Slider(
|
83 |
+
label="Temperature",
|
84 |
+
minimum=0.1,
|
85 |
+
maximum=4.0,
|
86 |
+
step=0.1,
|
87 |
+
value=0.1,
|
88 |
+
),
|
89 |
+
gr.Slider(
|
90 |
+
label="Top-p (nucleus sampling)",
|
91 |
+
minimum=0.05,
|
92 |
+
maximum=1.0,
|
93 |
+
step=0.05,
|
94 |
+
value=0.95,
|
95 |
+
),
|
96 |
+
gr.Slider(
|
97 |
+
label="Top-k",
|
98 |
+
minimum=1,
|
99 |
+
maximum=1000,
|
100 |
+
step=1,
|
101 |
+
value=50,
|
102 |
+
),
|
103 |
+
gr.Slider(
|
104 |
+
label="Repetition penalty",
|
105 |
+
minimum=1.0,
|
106 |
+
maximum=2.0,
|
107 |
+
step=0.05,
|
108 |
+
value=1.2,
|
109 |
+
),
|
110 |
+
],
|
111 |
+
stop_btn=None,
|
112 |
+
examples=[
|
113 |
+
["Hello there! How are you doing?"],
|
114 |
+
["Can you explain briefly to me what is the Python programming language?"],
|
115 |
+
["Explain the plot of Cinderella in a sentence."],
|
116 |
+
["How many hours does it take a man to eat a Helicopter?"],
|
117 |
+
["Write a 100-word article on 'Benefits of Open-Source in AI research'."],
|
118 |
+
],
|
119 |
+
cache_examples=False,
|
120 |
+
chatbot=gr.Chatbot(
|
121 |
+
height=600)
|
122 |
+
)
|
123 |
+
|
124 |
+
with gr.Blocks(css="style.css", theme=gr.themes.Default()) as demo:
|
125 |
+
# Get the server status before displaying UI
|
126 |
+
visibility = check_server_health(CLOUD_GATEWAY_API)
|
127 |
+
|
128 |
+
# Container for the main interface
|
129 |
+
with gr.Column(visible=visibility, elem_id="main_ui") as main_ui:
|
130 |
+
gr.Markdown(f"""
|
131 |
+
# Llama-3 8B Chat
|
132 |
+
This Space is an Alpha release that demonstrates model [Llama-3-8b-chat](https://huggingface.co/meta-llama/Meta-Llama-3-8B) by Meta, a Llama 3 model with 8B parameters fine-tuned for chat instructions, running on AMD MI210 infrastructure. Feel free to play with it!
|
133 |
+
""")
|
134 |
+
chat_interface.render()
|
135 |
+
|
136 |
+
# Dialog box using Markdown for the error message
|
137 |
+
with gr.Row(visible=(not visibility), elem_id="dialog_box") as dialog_box:
|
138 |
+
# Add spinner and message
|
139 |
+
get_dialog_box()
|
140 |
+
|
141 |
+
# Timer to check server health every 5 seconds and update UI
|
142 |
+
timer = gr.Timer(value=10)
|
143 |
+
timer.tick(fn=toggle_ui, outputs=[main_ui, dialog_box])
|
144 |
+
|
145 |
+
|
146 |
+
if __name__ == "__main__":
|
147 |
+
demo.queue(max_size=int(os.getenv("QUEUE"))).launch()
|
dialog.py
ADDED
@@ -0,0 +1,45 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import gradio as gr
|
2 |
+
|
3 |
+
|
4 |
+
def get_dialog_box():
|
5 |
+
return gr.HTML("""
|
6 |
+
<div style="display: flex; align-items: center; justify-content: center; min-height: 80vh;">
|
7 |
+
<div style="display: flex; flex-direction: column; align-items: center;">
|
8 |
+
<!-- Spinner -->
|
9 |
+
<div class="loader" style="margin-top: 20px;"></div>
|
10 |
+
<!-- Message -->
|
11 |
+
<h2 style="color: orange; font-family: trebuchet ms, sans-serif; align-items: center;">The service is not working, please refresh or try again later!</h2>
|
12 |
+
</div>
|
13 |
+
</div>
|
14 |
+
|
15 |
+
<!-- Spinner CSS -->
|
16 |
+
<style>
|
17 |
+
/* HTML: <div class="loader"></div> */
|
18 |
+
.loader {
|
19 |
+
width: 120px;
|
20 |
+
height: 22px;
|
21 |
+
border-radius: 40px;
|
22 |
+
color: orange !important;
|
23 |
+
border: 2px solid;
|
24 |
+
position: relative;
|
25 |
+
overflow: hidden;
|
26 |
+
}
|
27 |
+
.loader::before {
|
28 |
+
content: "";
|
29 |
+
position: absolute;
|
30 |
+
margin: 2px;
|
31 |
+
width: 14px;
|
32 |
+
top: 0;
|
33 |
+
bottom: 0;
|
34 |
+
left: -20px;
|
35 |
+
border-radius: inherit;
|
36 |
+
background: currentColor;
|
37 |
+
box-shadow: -10px 0 12px 3px currentColor;
|
38 |
+
clip-path: polygon(0 5%, 100% 0,100% 100%,0 95%,-30px 50%);
|
39 |
+
animation: l14 1s infinite linear;
|
40 |
+
}
|
41 |
+
@keyframes l14 {
|
42 |
+
100% {left: calc(100% + 20px)}
|
43 |
+
}
|
44 |
+
</style>
|
45 |
+
""")
|
gateway.py
ADDED
@@ -0,0 +1,90 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import json
|
2 |
+
import requests
|
3 |
+
|
4 |
+
|
5 |
+
def check_server_health(cloud_gateway_api: str):
|
6 |
+
"""
|
7 |
+
Use the appropriate API endpoint to check the server health.
|
8 |
+
Args:
|
9 |
+
cloud_gateway_api: API endpoint to probe.
|
10 |
+
|
11 |
+
Returns:
|
12 |
+
True if server is active, false otherwise.
|
13 |
+
"""
|
14 |
+
try:
|
15 |
+
response = requests.get(cloud_gateway_api + "/health")
|
16 |
+
if response.status_code == 200:
|
17 |
+
return True
|
18 |
+
except requests.ConnectionError:
|
19 |
+
print("Failed to establish connection to the server.")
|
20 |
+
|
21 |
+
return False
|
22 |
+
|
23 |
+
|
24 |
+
def request_generation(message: str,
|
25 |
+
system_prompt: str,
|
26 |
+
cloud_gateway_api: str,
|
27 |
+
max_new_tokens: int = 1024,
|
28 |
+
temperature: float = 0.6,
|
29 |
+
top_p: float = 0.9,
|
30 |
+
top_k: int = 50,
|
31 |
+
repetition_penalty: float = 1.2, ):
|
32 |
+
"""
|
33 |
+
Request streaming generation from the cloud gateway API. Uses the simple requests module with stream=True to utilize
|
34 |
+
token-by-token generation from LLM.
|
35 |
+
|
36 |
+
Args:
|
37 |
+
message: prompt from the user.
|
38 |
+
system_prompt: system prompt to append.
|
39 |
+
cloud_gateway_api (str): API endpoint to send the request.
|
40 |
+
max_new_tokens: maximum number of tokens to generate, ignoring the number of tokens in the prompt.
|
41 |
+
temperature: the value used to module the next token probabilities.
|
42 |
+
top_p: if set to float<1, only the smallest set of most probable tokens with probabilities that add up to top_p
|
43 |
+
or higher are kept for generation.
|
44 |
+
top_k: the number of highest probability vocabulary tokens to keep for top-k-filtering.
|
45 |
+
repetition_penalty: the parameter for repetition penalty. 1.0 means no penalty.
|
46 |
+
|
47 |
+
Returns:
|
48 |
+
|
49 |
+
"""
|
50 |
+
|
51 |
+
payload = {
|
52 |
+
"model": "meta-llama/Meta-Llama-3-8B-Instruct",
|
53 |
+
"messages": [
|
54 |
+
{"role": "system", "content": system_prompt},
|
55 |
+
{"role": "user", "content": message}
|
56 |
+
],
|
57 |
+
"max_tokens": max_new_tokens,
|
58 |
+
"temperature": temperature,
|
59 |
+
"top_p": top_p,
|
60 |
+
"repetition_penalty": repetition_penalty,
|
61 |
+
"top_k": top_k,
|
62 |
+
"stream": True # Enable streaming
|
63 |
+
}
|
64 |
+
|
65 |
+
with requests.post(cloud_gateway_api + "/v1/chat/completions", json=payload, stream=True) as response:
|
66 |
+
for chunk in response.iter_lines():
|
67 |
+
if chunk:
|
68 |
+
# Convert the chunk from bytes to a string and then parse it as json
|
69 |
+
chunk_str = chunk.decode('utf-8')
|
70 |
+
|
71 |
+
# Remove the `data: ` prefix from the chunk if it exists
|
72 |
+
if chunk_str.startswith("data: "):
|
73 |
+
chunk_str = chunk_str[len("data: "):]
|
74 |
+
|
75 |
+
# Skip empty chunks
|
76 |
+
if chunk_str.strip() == "[DONE]":
|
77 |
+
break
|
78 |
+
|
79 |
+
# Parse the chunk into a JSON object
|
80 |
+
try:
|
81 |
+
chunk_json = json.loads(chunk_str)
|
82 |
+
# Extract the "content" field from the choices
|
83 |
+
content = chunk_json["choices"][0]["delta"].get("content", "")
|
84 |
+
|
85 |
+
# Print the generated content as it's streamed
|
86 |
+
if content:
|
87 |
+
yield content
|
88 |
+
except json.JSONDecodeError:
|
89 |
+
# Handle any potential errors in decoding
|
90 |
+
continue
|
requirements.txt
ADDED
@@ -0,0 +1,5 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
numpy==1.26.4
|
2 |
+
pillow==10.4.0
|
3 |
+
gradio==4.43.0
|
4 |
+
fastapi==0.111.1
|
5 |
+
websockets==11.0.3
|