Spaces:
Runtime error
Runtime error
Update app.py
Browse files
app.py
CHANGED
@@ -1,42 +1,5 @@
|
|
1 |
-
import os
|
2 |
-
from threading import Thread
|
3 |
-
from typing import Iterator
|
4 |
-
|
5 |
import gradio as gr
|
6 |
-
import
|
7 |
-
import torch
|
8 |
-
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
9 |
-
|
10 |
-
MAX_MAX_NEW_TOKENS = 1024
|
11 |
-
DEFAULT_MAX_NEW_TOKENS = 256
|
12 |
-
MAX_INPUT_TOKEN_LENGTH = 512
|
13 |
-
|
14 |
-
DESCRIPTION = """\
|
15 |
-
# Buzz-3B-Small
|
16 |
-
This Space demonstrates Buzz-3b-small-v0.6.3.
|
17 |
-
"""
|
18 |
-
|
19 |
-
LICENSE = """
|
20 |
-
<p/>
|
21 |
-
---
|
22 |
-
This demo uses Buzz-3b-small-v0.6.3. Please check the model card for details.
|
23 |
-
"""
|
24 |
-
|
25 |
-
if not torch.cuda.is_available():
|
26 |
-
DESCRIPTION += "\n<p>Running on CPU 🥶 This demo works better on GPU.</p>"
|
27 |
-
|
28 |
-
model_id = "H-D-T/Buzz-3b-small-v0.6.3"
|
29 |
-
|
30 |
-
if torch.cuda.is_available():
|
31 |
-
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto", trust_remote_code=True, low_cpu_mem_usage=True)
|
32 |
-
else:
|
33 |
-
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="cpu", trust_remote_code=True, low_cpu_mem_usage=True)
|
34 |
-
|
35 |
-
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
36 |
-
if tokenizer.pad_token == None:
|
37 |
-
tokenizer.pad_token = tokenizer.eos_token
|
38 |
-
tokenizer.pad_token_id = tokenizer.eos_token_id
|
39 |
-
model.config.pad_token_id = tokenizer.eos_token_id
|
40 |
|
41 |
# Define the special tokens
|
42 |
bos_token = "<|begin_of_text|>"
|
@@ -44,132 +7,43 @@ eos_token = "<|eot_id|>"
|
|
44 |
start_header_id = "<|start_header_id|>"
|
45 |
end_header_id = "<|end_header_id|>"
|
46 |
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
{{% for message in loop_messages %}}
|
55 |
-
{{% set content = '{start_header_id}' + message['role'] + '{end_header_id}\\n\\n' + message['content'].strip() + '{eos_token}' %}}
|
56 |
-
{{% if loop.index0 == 0 %}}{{% set content = bos_token + content %}}{{% endif %}}
|
57 |
-
{{ content }}
|
58 |
-
{{% endfor %}}
|
59 |
-
{{% if add_generation_prompt %}}{{ '{start_header_id}assistant{end_header_id}\\n\\n' }}{{% else %}}{{ eos_token }}{{% endif %}}
|
60 |
-
"""
|
61 |
-
chat_context = ""
|
62 |
for i, (user, assistant) in enumerate(chat_history):
|
63 |
-
user_msg = start_header_id
|
64 |
-
assistant_msg = start_header_id
|
65 |
if i == 0:
|
66 |
user_msg = bos_token + user_msg
|
67 |
-
|
68 |
-
|
69 |
-
if add_generation_prompt:
|
70 |
-
chat_context += start_header_id + "assistant" + end_header_id + "\n\n"
|
71 |
-
else:
|
72 |
-
chat_context += eos_token
|
73 |
-
|
74 |
-
return chat_context
|
75 |
|
76 |
-
|
77 |
-
def generate(
|
78 |
-
message: str,
|
79 |
-
chat_history: list[tuple[str, str]],
|
80 |
-
max_new_tokens: int = 1024,
|
81 |
-
temperature: float = 0.6,
|
82 |
-
top_p: float = 0.9,
|
83 |
-
top_k: int = 50,
|
84 |
-
repetition_penalty: float = 1.4,
|
85 |
-
) -> Iterator[str]:
|
86 |
-
|
87 |
chat_history.append(("user", message))
|
88 |
-
|
89 |
-
|
90 |
-
|
91 |
-
|
92 |
-
|
93 |
-
|
94 |
-
|
95 |
-
streamer = TextIteratorStreamer(tokenizer, timeout=10.0, skip_prompt=True, skip_special_tokens=True)
|
96 |
-
generate_kwargs = dict(
|
97 |
-
{"input_ids": input_ids},
|
98 |
-
streamer=streamer,
|
99 |
-
max_new_tokens=max_new_tokens,
|
100 |
-
do_sample=True,
|
101 |
-
top_p=top_p,
|
102 |
-
top_k=top_k,
|
103 |
-
temperature=temperature,
|
104 |
-
num_beams=1,
|
105 |
-
pad_token_id = tokenizer.eos_token_id,
|
106 |
-
repetition_penalty=repetition_penalty,
|
107 |
-
no_repeat_ngram_size=5,
|
108 |
-
early_stopping=False,
|
109 |
-
)
|
110 |
-
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
111 |
-
t.start()
|
112 |
-
|
113 |
-
outputs = []
|
114 |
-
for text in streamer:
|
115 |
-
outputs.append(text)
|
116 |
-
yield "".join(outputs)
|
117 |
-
|
118 |
-
|
119 |
-
chat_interface = gr.ChatInterface(
|
120 |
-
fn=generate,
|
121 |
-
additional_inputs=[
|
122 |
-
gr.Slider(
|
123 |
-
label="Max new tokens",
|
124 |
-
minimum=1,
|
125 |
-
maximum=MAX_MAX_NEW_TOKENS,
|
126 |
-
step=1,
|
127 |
-
value=DEFAULT_MAX_NEW_TOKENS,
|
128 |
-
),
|
129 |
-
gr.Slider(
|
130 |
-
label="Temperature",
|
131 |
-
minimum=0.1,
|
132 |
-
maximum=4.0,
|
133 |
-
step=0.1,
|
134 |
-
value=0.6,
|
135 |
-
),
|
136 |
-
gr.Slider(
|
137 |
-
label="Top-p (nucleus sampling)",
|
138 |
-
minimum=0.05,
|
139 |
-
maximum=1.0,
|
140 |
-
step=0.05,
|
141 |
-
value=0.9,
|
142 |
-
),
|
143 |
-
gr.Slider(
|
144 |
-
label="Top-k",
|
145 |
-
minimum=1,
|
146 |
-
maximum=1000,
|
147 |
-
step=1,
|
148 |
-
value=50,
|
149 |
-
),
|
150 |
-
gr.Slider(
|
151 |
-
label="Repetition penalty",
|
152 |
-
minimum=1.0,
|
153 |
-
maximum=2.0,
|
154 |
-
step=0.05,
|
155 |
-
value=1.4,
|
156 |
-
),
|
157 |
-
],
|
158 |
-
stop_btn=None,
|
159 |
-
examples=[
|
160 |
-
["A recipe for a chocolate cake:"],
|
161 |
-
["Can you explain briefly to me what is the Python programming language?"],
|
162 |
-
["Explain the plot of Cinderella in a sentence."],
|
163 |
-
["Question: What is the capital of France?\nAnswer:"],
|
164 |
-
["Question: I am very tired, what should I do?\nAnswer:"],
|
165 |
-
],
|
166 |
-
)
|
167 |
|
168 |
with gr.Blocks(css="style.css") as demo:
|
169 |
-
gr.Markdown(
|
170 |
-
gr.
|
171 |
-
|
172 |
-
gr.
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
173 |
|
174 |
if __name__ == "__main__":
|
175 |
demo.queue(max_size=20).launch()
|
|
|
|
|
|
|
|
|
|
|
1 |
import gradio as gr
|
2 |
+
from transformers import pipeline, Conversation, AutoTokenizer
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
3 |
|
4 |
# Define the special tokens
|
5 |
bos_token = "<|begin_of_text|>"
|
|
|
7 |
start_header_id = "<|start_header_id|>"
|
8 |
end_header_id = "<|end_header_id|>"
|
9 |
|
10 |
+
# Load the conversational pipeline and tokenizer
|
11 |
+
model_id = "H-D-T/Buzz-3b-small-v0.6.3"
|
12 |
+
chatbot = pipeline("conversational", model=model_id)
|
13 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id)
|
14 |
+
|
15 |
+
def format_conversation(chat_history):
|
16 |
+
formatted_history = ""
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
17 |
for i, (user, assistant) in enumerate(chat_history):
|
18 |
+
user_msg = f"{start_header_id}user{end_header_id}\n\n{user.strip()}{eos_token}"
|
19 |
+
assistant_msg = f"{start_header_id}assistant{end_header_id}\n\n{assistant.strip()}{eos_token}"
|
20 |
if i == 0:
|
21 |
user_msg = bos_token + user_msg
|
22 |
+
formatted_history += user_msg + assistant_msg
|
23 |
+
return formatted_history
|
|
|
|
|
|
|
|
|
|
|
|
|
24 |
|
25 |
+
def predict(message, chat_history):
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
26 |
chat_history.append(("user", message))
|
27 |
+
formatted_history = format_conversation(chat_history)
|
28 |
+
conversation = Conversation(formatted_history)
|
29 |
+
conversation = chatbot(conversation)
|
30 |
+
response = conversation.generated_responses[-1]
|
31 |
+
chat_history.append(("assistant", response))
|
32 |
+
return "", chat_history
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
|
34 |
with gr.Blocks(css="style.css") as demo:
|
35 |
+
gr.Markdown("# Buzz-3B-Small Conversational Demo")
|
36 |
+
with gr.Chatbot() as chatbot_ui:
|
37 |
+
chatbot_ui.append({"role": "assistant", "content": "Hi, how can I help you today?"})
|
38 |
+
with gr.Row():
|
39 |
+
with gr.Column():
|
40 |
+
textbox = gr.Textbox(label="Your message:")
|
41 |
+
with gr.Column():
|
42 |
+
submit_btn = gr.Button("Send")
|
43 |
+
|
44 |
+
chat_history = gr.State([])
|
45 |
+
|
46 |
+
submit_btn.click(predict, inputs=[textbox, chat_history], outputs=[textbox, chat_history])
|
47 |
|
48 |
if __name__ == "__main__":
|
49 |
demo.queue(max_size=20).launch()
|