create app.py
Browse files
app.py
ADDED
@@ -0,0 +1,81 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import time
|
2 |
+
import gradio as gr
|
3 |
+
|
4 |
+
API_URL = "https://joi-20b.ngrok.io/generate_stream"
|
5 |
+
|
6 |
+
def predict(inputs, history=[], top_p, temperature, top_k, repetition_penalty):
|
7 |
+
if not inputs.startswith("User: "):
|
8 |
+
inputs = "User: " + inputs + "\n"
|
9 |
+
payload = {
|
10 |
+
"inputs": inputs, #"My name is Jane and I",
|
11 |
+
"parameters": {
|
12 |
+
"details": True,
|
13 |
+
"do_sample": True,
|
14 |
+
"max_new_tokens": 20,
|
15 |
+
"repetition_penalty": 1.03,
|
16 |
+
"seed": 0,
|
17 |
+
"stop": ["photographer"],
|
18 |
+
"temperature": 0.5,
|
19 |
+
"top_k": 10,
|
20 |
+
"top_p": 0.95
|
21 |
+
}
|
22 |
+
}
|
23 |
+
|
24 |
+
headers = {
|
25 |
+
'accept': 'text/event-stream',
|
26 |
+
'Content-Type': 'application/json'
|
27 |
+
}
|
28 |
+
|
29 |
+
history.append(inputs)
|
30 |
+
response = requests.post(API_URL, headers=headers, json=payload)
|
31 |
+
responses = response.text.split("\n\n")
|
32 |
+
|
33 |
+
partial_words = ""
|
34 |
+
for idx, resp in enumerate(responses):
|
35 |
+
if resp[:4] == 'data':
|
36 |
+
partial_words = partial_words + json.loads(resp[5:])['token']['text']
|
37 |
+
#print(partial_words)
|
38 |
+
time.sleep(0.05)
|
39 |
+
if idx == 0:
|
40 |
+
history.append(" " + partial_words)
|
41 |
+
else:
|
42 |
+
history[-1] = partial_words
|
43 |
+
|
44 |
+
chat = [(history[i], history[i + 1]) for i in range(0, len(history) - 1, 2) ] # convert to tuples of list
|
45 |
+
|
46 |
+
yield chat, history #resembles {chatbot: chat, state: history}
|
47 |
+
|
48 |
+
title = """<h1 align="center">Gradio Streaming</h1>"""
|
49 |
+
description = """Language models can be conditioned to act like dialogue agents through a conversational prompt that typically takes the form:
|
50 |
+
```
|
51 |
+
User: <utterance>
|
52 |
+
Assistant: <utterance>
|
53 |
+
User: <utterance>
|
54 |
+
Assistant: <utterance>
|
55 |
+
...
|
56 |
+
```
|
57 |
+
In this app, you can explore the outputs of the Joi alpha language models.
|
58 |
+
"""
|
59 |
+
|
60 |
+
with gr.Blocks(css = "#chatbot {height: 400px, overflow: auto;}") as demo:
|
61 |
+
gr.HTML(title)
|
62 |
+
inputs = gr.Textbox(placeholder= "Hi my name is Joe.", label= "Type an input and press Enter") #t
|
63 |
+
chatbot = gr.Chatbot(elem_id='chatbot') #c
|
64 |
+
state = gr.State([]) #s
|
65 |
+
b1 = gr.Button()
|
66 |
+
|
67 |
+
#inputs, top_p, temperature, top_k, repetition_penalty
|
68 |
+
with gr.Accordion("Parameters", open=False):
|
69 |
+
top_p = gr.Slider( minimum=-0, maximum=1.0, value=0.95, step=0.05, interactive=True, label="Top-p (nucleus sampling)",)
|
70 |
+
temperature = gr.Slider( minimum=-0, maximum=5.0, value=0.5, step=0.1, interactive=True, label="Temperature",)
|
71 |
+
top_k = gr.Slider( minimum=1, maximum=50, value=4, step=1, interactive=True, label="Top-k",)
|
72 |
+
repetition_penalty = gr.Slider( minimum=0.1, maximum=3.0, value=1.03, step=0.01, interactive=True, label="Repetition Penalty", )
|
73 |
+
|
74 |
+
#b1.click(predict, [t,s], [c,s])
|
75 |
+
#inputs.submit(predict, [t,s], [c,s])
|
76 |
+
inputs.submit( inference_chat, [inputs, state, top_p, temperature, top_k, repetition_penalty,], [chatbot, state],)
|
77 |
+
b1.click( inference_chat, [inputs, state, top_p, temperature, top_k, repetition_penalty,], [chatbot, state],)
|
78 |
+
|
79 |
+
gr.HTML(description)
|
80 |
+
demo.queue().launch(debug=True)
|
81 |
+
|