Spaces:
Running
Running
yentinglin
commited on
Commit
•
05eef7a
1
Parent(s):
ca877b2
Update app.py
Browse files
app.py
CHANGED
@@ -6,42 +6,12 @@ from text_generation import Client
|
|
6 |
from conversation import get_default_conv_template
|
7 |
|
8 |
|
9 |
-
endpoint_url = os.environ.get("ENDPOINT_URL")
|
10 |
client = Client(endpoint_url, timeout=120)
|
11 |
eos_token = "</s>"
|
12 |
|
13 |
|
14 |
|
15 |
-
def generate_response(history, max_new_token=512, top_p=0.9, temperature=0.8, do_sample=True):
|
16 |
-
conv = get_default_conv_template("vicuna").copy()
|
17 |
-
roles = {"human": conv.roles[0], "gpt": conv.roles[1]} # map human to USER and gpt to ASSISTANT
|
18 |
-
for user, bot in history:
|
19 |
-
conv.append_message(roles['human'], user)
|
20 |
-
conv.append_message(roles["gpt"], bot)
|
21 |
-
msg = conv.get_prompt()
|
22 |
-
|
23 |
-
for response in client.generate_stream(
|
24 |
-
msg,
|
25 |
-
max_new_tokens=max_new_token,
|
26 |
-
top_p=top_p,
|
27 |
-
temperature=temperature,
|
28 |
-
do_sample=do_sample,
|
29 |
-
):
|
30 |
-
if not response.token.special:
|
31 |
-
yield response.token.text
|
32 |
-
|
33 |
-
# res = client.generate(
|
34 |
-
# msg,
|
35 |
-
# stop_sequences=["<|assistant|>", eos_token, "<|system|>", "<|user|>"],
|
36 |
-
# max_new_tokens=max_new_token,
|
37 |
-
# top_p=top_p,
|
38 |
-
# top_k=top_k,
|
39 |
-
# do_sample=do_sample,
|
40 |
-
# temperature=temperature,
|
41 |
-
# repetition_penalty=repetition_penalty,
|
42 |
-
# )
|
43 |
-
# return [("assistant", res.generated_text)]
|
44 |
-
#
|
45 |
with gr.Blocks() as demo:
|
46 |
chatbot = gr.Chatbot()
|
47 |
msg = gr.Textbox()
|
@@ -51,22 +21,128 @@ with gr.Blocks() as demo:
|
|
51 |
return "", history + [[user_message, None]]
|
52 |
|
53 |
def bot(history):
|
54 |
-
|
|
|
|
|
|
|
|
|
|
|
55 |
|
56 |
-
import ipdb
|
57 |
-
ipdb.set_trace()
|
58 |
-
bot_message = random.choice(["How are you?", "I love you", "I'm very hungry"])
|
59 |
history[-1][1] = ""
|
60 |
-
for
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
61 |
history[-1][1] += character
|
|
|
62 |
time.sleep(0.05)
|
63 |
yield history
|
64 |
|
|
|
65 |
msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
|
66 |
-
|
67 |
)
|
68 |
clear.click(lambda: None, None, chatbot, queue=False)
|
69 |
|
70 |
demo.queue()
|
71 |
demo.launch()
|
72 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
6 |
from conversation import get_default_conv_template
|
7 |
|
8 |
|
9 |
+
endpoint_url = os.environ.get("ENDPOINT_URL", "http://127.0.0.1:8080")
|
10 |
client = Client(endpoint_url, timeout=120)
|
11 |
eos_token = "</s>"
|
12 |
|
13 |
|
14 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
15 |
with gr.Blocks() as demo:
|
16 |
chatbot = gr.Chatbot()
|
17 |
msg = gr.Textbox()
|
|
|
21 |
return "", history + [[user_message, None]]
|
22 |
|
23 |
def bot(history):
|
24 |
+
conv = get_default_conv_template("vicuna").copy()
|
25 |
+
roles = {"human": conv.roles[0], "gpt": conv.roles[1]} # map human to USER and gpt to ASSISTANT
|
26 |
+
for user, bot in history:
|
27 |
+
conv.append_message(roles['human'], user)
|
28 |
+
conv.append_message(roles["gpt"], bot)
|
29 |
+
msg = conv.get_prompt()
|
30 |
|
|
|
|
|
|
|
31 |
history[-1][1] = ""
|
32 |
+
for response in client.generate_stream(
|
33 |
+
msg,
|
34 |
+
max_new_tokens=512,
|
35 |
+
):
|
36 |
+
if not response.token.special:
|
37 |
+
character = response.token.text
|
38 |
+
history[-1][1] += character
|
39 |
+
yield history
|
40 |
+
|
41 |
+
|
42 |
+
def generate_response(history, max_new_token=512, top_p=0.9, temperature=0.8, do_sample=True):
|
43 |
+
conv = get_default_conv_template("vicuna").copy()
|
44 |
+
roles = {"human": conv.roles[0], "gpt": conv.roles[1]} # map human to USER and gpt to ASSISTANT
|
45 |
+
for user, bot in history:
|
46 |
+
conv.append_message(roles['human'], user)
|
47 |
+
conv.append_message(roles["gpt"], bot)
|
48 |
+
msg = conv.get_prompt()
|
49 |
+
|
50 |
+
for response in client.generate_stream(
|
51 |
+
msg,
|
52 |
+
max_new_tokens=max_new_token,
|
53 |
+
top_p=top_p,
|
54 |
+
temperature=temperature,
|
55 |
+
do_sample=do_sample,
|
56 |
+
):
|
57 |
+
history[-1][1] = ""
|
58 |
+
# if not response.token.special:
|
59 |
+
character = response.token.text
|
60 |
history[-1][1] += character
|
61 |
+
print(history[-1][1])
|
62 |
time.sleep(0.05)
|
63 |
yield history
|
64 |
|
65 |
+
|
66 |
msg.submit(user, [msg, chatbot], [msg, chatbot], queue=False).then(
|
67 |
+
bot, chatbot, chatbot
|
68 |
)
|
69 |
clear.click(lambda: None, None, chatbot, queue=False)
|
70 |
|
71 |
demo.queue()
|
72 |
demo.launch()
|
73 |
|
74 |
+
#
|
75 |
+
# with gr.Blocks() as demo:
|
76 |
+
# chatbot = gr.Chatbot()
|
77 |
+
# with gr.Row():
|
78 |
+
# with gr.Column(scale=4):
|
79 |
+
# with gr.Column(scale=12):
|
80 |
+
# user_input = gr.Textbox(
|
81 |
+
# show_label=False,
|
82 |
+
# placeholder="Shift + Enter傳送...",
|
83 |
+
# lines=10).style(
|
84 |
+
# container=False)
|
85 |
+
# with gr.Column(min_width=32, scale=1):
|
86 |
+
# submitBtn = gr.Button("Submit", variant="primary")
|
87 |
+
# with gr.Column(scale=1):
|
88 |
+
# emptyBtn = gr.Button("Clear History")
|
89 |
+
# max_new_token = gr.Slider(
|
90 |
+
# 1,
|
91 |
+
# 1024,
|
92 |
+
# value=128,
|
93 |
+
# step=1.0,
|
94 |
+
# label="Maximum New Token Length",
|
95 |
+
# interactive=True)
|
96 |
+
# top_p = gr.Slider(0, 1, value=0.9, step=0.01,
|
97 |
+
# label="Top P", interactive=True)
|
98 |
+
# temperature = gr.Slider(
|
99 |
+
# 0,
|
100 |
+
# 1,
|
101 |
+
# value=0.5,
|
102 |
+
# step=0.01,
|
103 |
+
# label="Temperature",
|
104 |
+
# interactive=True)
|
105 |
+
# top_k = gr.Slider(1, 40, value=40, step=1,
|
106 |
+
# label="Top K", interactive=True)
|
107 |
+
# do_sample = gr.Checkbox(
|
108 |
+
# value=True,
|
109 |
+
# label="Do Sample",
|
110 |
+
# info="use random sample strategy",
|
111 |
+
# interactive=True)
|
112 |
+
# repetition_penalty = gr.Slider(
|
113 |
+
# 1.0,
|
114 |
+
# 3.0,
|
115 |
+
# value=1.1,
|
116 |
+
# step=0.1,
|
117 |
+
# label="Repetition Penalty",
|
118 |
+
# interactive=True)
|
119 |
+
#
|
120 |
+
# params = [user_input, chatbot]
|
121 |
+
# predict_params = [
|
122 |
+
# chatbot,
|
123 |
+
# max_new_token,
|
124 |
+
# top_p,
|
125 |
+
# temperature,
|
126 |
+
# top_k,
|
127 |
+
# do_sample,
|
128 |
+
# repetition_penalty]
|
129 |
+
#
|
130 |
+
# submitBtn.click(
|
131 |
+
# generate_response,
|
132 |
+
# [user_input, max_new_token, top_p, top_k, temperature, do_sample, repetition_penalty],
|
133 |
+
# [chatbot],
|
134 |
+
# queue=False
|
135 |
+
# )
|
136 |
+
#
|
137 |
+
# user_input.submit(
|
138 |
+
# generate_response,
|
139 |
+
# [user_input, max_new_token, top_p, top_k, temperature, do_sample, repetition_penalty],
|
140 |
+
# [chatbot],
|
141 |
+
# queue=False
|
142 |
+
# )
|
143 |
+
#
|
144 |
+
# submitBtn.click(lambda: None, [], [user_input])
|
145 |
+
#
|
146 |
+
# emptyBtn.click(lambda: chatbot.reset(), outputs=[chatbot], show_progress=True)
|
147 |
+
#
|
148 |
+
# demo.launch()
|