Spaces:
Sleeping
Sleeping
aaaaa
Browse files
app.py
CHANGED
|
@@ -1,9 +1,12 @@
|
|
| 1 |
import gradio as gr
|
| 2 |
from huggingface_hub import InferenceClient
|
| 3 |
import time
|
|
|
|
|
|
|
| 4 |
|
| 5 |
# Inference client setup
|
| 6 |
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
|
|
|
|
| 7 |
|
| 8 |
# Global flag to handle cancellation
|
| 9 |
stop_inference = False
|
|
@@ -22,10 +25,31 @@ def respond(
|
|
| 22 |
|
| 23 |
if use_local_model:
|
| 24 |
# Simulate local inference
|
| 25 |
-
|
| 26 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 27 |
history.append((message, response))
|
| 28 |
-
yield history
|
|
|
|
| 29 |
else:
|
| 30 |
# API-based inference
|
| 31 |
messages = [{"role": "system", "content": system_message}]
|
|
@@ -49,12 +73,10 @@ def respond(
|
|
| 49 |
break
|
| 50 |
token = message_chunk.choices[0].delta.content
|
| 51 |
response += token
|
| 52 |
-
|
| 53 |
-
yield history # Yield the history list of tuples
|
| 54 |
|
| 55 |
-
|
| 56 |
-
history
|
| 57 |
-
yield history
|
| 58 |
|
| 59 |
def cancel_inference():
|
| 60 |
global stop_inference
|
|
@@ -127,8 +149,7 @@ with gr.Blocks(css=custom_css) as demo:
|
|
| 127 |
cancel_button = gr.Button("Cancel Inference", variant="danger")
|
| 128 |
|
| 129 |
def chat_fn(message, history):
|
| 130 |
-
|
| 131 |
-
return respond(
|
| 132 |
message,
|
| 133 |
history,
|
| 134 |
system_message.value,
|
|
@@ -137,6 +158,9 @@ with gr.Blocks(css=custom_css) as demo:
|
|
| 137 |
top_p.value,
|
| 138 |
use_local_model.value,
|
| 139 |
)
|
|
|
|
|
|
|
|
|
|
| 140 |
|
| 141 |
user_input.submit(chat_fn, [user_input, chat_history], chat_history)
|
| 142 |
cancel_button.click(cancel_inference)
|
|
|
|
| 1 |
import gradio as gr
|
| 2 |
from huggingface_hub import InferenceClient
|
| 3 |
import time
|
| 4 |
+
import torch
|
| 5 |
+
from transformers import pipeline
|
| 6 |
|
| 7 |
# Inference client setup
|
| 8 |
client = InferenceClient("HuggingFaceH4/zephyr-7b-beta")
|
| 9 |
+
pipe = pipeline("text-generation", "microsoft/Phi-3-mini-4k-instruct", torch_dtype=torch.bfloat16, device_map="auto")
|
| 10 |
|
| 11 |
# Global flag to handle cancellation
|
| 12 |
stop_inference = False
|
|
|
|
| 25 |
|
| 26 |
if use_local_model:
|
| 27 |
# Simulate local inference
|
| 28 |
+
messages = [{"role": "system", "content": system_message}]
|
| 29 |
+
|
| 30 |
+
for val in history:
|
| 31 |
+
if val[0]:
|
| 32 |
+
messages.append({"role": "user", "content": val[0]})
|
| 33 |
+
if val[1]:
|
| 34 |
+
messages.append({"role": "assistant", "content": val[1]})
|
| 35 |
+
|
| 36 |
+
messages.append({"role": "user", "content": message})
|
| 37 |
+
|
| 38 |
+
response = ""
|
| 39 |
+
for message in pipe(
|
| 40 |
+
messages,
|
| 41 |
+
max_new_tokens=max_tokens,
|
| 42 |
+
temperature=temperature,
|
| 43 |
+
do_sample=True,
|
| 44 |
+
top_p=top_p,
|
| 45 |
+
):
|
| 46 |
+
token = message['generated_text'][-1]['content']
|
| 47 |
+
response += token
|
| 48 |
+
yield response # Yielding response directly
|
| 49 |
+
|
| 50 |
history.append((message, response))
|
| 51 |
+
yield history # Yield the updated history
|
| 52 |
+
|
| 53 |
else:
|
| 54 |
# API-based inference
|
| 55 |
messages = [{"role": "system", "content": system_message}]
|
|
|
|
| 73 |
break
|
| 74 |
token = message_chunk.choices[0].delta.content
|
| 75 |
response += token
|
| 76 |
+
yield response # Yielding response directly
|
|
|
|
| 77 |
|
| 78 |
+
history.append((message, response))
|
| 79 |
+
yield history # Yield the updated history
|
|
|
|
| 80 |
|
| 81 |
def cancel_inference():
|
| 82 |
global stop_inference
|
|
|
|
| 149 |
cancel_button = gr.Button("Cancel Inference", variant="danger")
|
| 150 |
|
| 151 |
def chat_fn(message, history):
|
| 152 |
+
response_gen = respond(
|
|
|
|
| 153 |
message,
|
| 154 |
history,
|
| 155 |
system_message.value,
|
|
|
|
| 158 |
top_p.value,
|
| 159 |
use_local_model.value,
|
| 160 |
)
|
| 161 |
+
for response in response_gen:
|
| 162 |
+
history[-1] = (message, response)
|
| 163 |
+
yield history
|
| 164 |
|
| 165 |
user_input.submit(chat_fn, [user_input, chat_history], chat_history)
|
| 166 |
cancel_button.click(cancel_inference)
|