gabrielclark3330
commited on
Commit
•
2a15330
1
Parent(s):
bcab068
Cleanups
Browse files
app.py
CHANGED
@@ -1,286 +1,3 @@
|
|
1 |
-
'''
|
2 |
-
import os
|
3 |
-
import gradio as gr
|
4 |
-
from transformers import AutoTokenizer, AutoModelForCausalLM
|
5 |
-
import torch
|
6 |
-
|
7 |
-
model_name_2_7B_instruct = "Zyphra/Zamba2-2.7B-instruct"
|
8 |
-
model_name_7B_instruct = "Zyphra/Zamba2-7B-instruct"
|
9 |
-
max_context_length = 4096
|
10 |
-
|
11 |
-
tokenizer_2_7B_instruct = AutoTokenizer.from_pretrained(model_name_2_7B_instruct)
|
12 |
-
model_2_7B_instruct = AutoModelForCausalLM.from_pretrained(
|
13 |
-
model_name_2_7B_instruct, device_map="cuda", torch_dtype=torch.bfloat16
|
14 |
-
)
|
15 |
-
|
16 |
-
tokenizer_7B_instruct = AutoTokenizer.from_pretrained(model_name_7B_instruct)
|
17 |
-
model_7B_instruct = AutoModelForCausalLM.from_pretrained(
|
18 |
-
model_name_7B_instruct, device_map="cuda", torch_dtype=torch.bfloat16
|
19 |
-
)
|
20 |
-
|
21 |
-
def extract_assistant_response(generated_text):
|
22 |
-
assistant_token = '<|im_start|> assistant'
|
23 |
-
end_token = '<|im_end|>'
|
24 |
-
start_idx = generated_text.rfind(assistant_token)
|
25 |
-
if start_idx == -1:
|
26 |
-
# Assistant token not found
|
27 |
-
return generated_text.strip()
|
28 |
-
start_idx += len(assistant_token)
|
29 |
-
end_idx = generated_text.find(end_token, start_idx)
|
30 |
-
if end_idx == -1:
|
31 |
-
# End token not found, return from start_idx to end
|
32 |
-
return generated_text[start_idx:].strip()
|
33 |
-
else:
|
34 |
-
return generated_text[start_idx:end_idx].strip()
|
35 |
-
|
36 |
-
def generate_response(chat_history, max_new_tokens, model, tokenizer):
|
37 |
-
sample = []
|
38 |
-
for turn in chat_history:
|
39 |
-
if turn[0]:
|
40 |
-
sample.append({'role': 'user', 'content': turn[0]})
|
41 |
-
if turn[1]:
|
42 |
-
sample.append({'role': 'assistant', 'content': turn[1]})
|
43 |
-
chat_sample = tokenizer.apply_chat_template(sample, tokenize=False)
|
44 |
-
input_ids = tokenizer(chat_sample, return_tensors='pt', add_special_tokens=False).to(model.device)
|
45 |
-
|
46 |
-
max_new_tokens = int(max_new_tokens)
|
47 |
-
max_input_length = max_context_length - max_new_tokens
|
48 |
-
if input_ids['input_ids'].size(1) > max_input_length:
|
49 |
-
input_ids['input_ids'] = input_ids['input_ids'][:, -max_input_length:]
|
50 |
-
if 'attention_mask' in input_ids:
|
51 |
-
input_ids['attention_mask'] = input_ids['attention_mask'][:, -max_input_length:]
|
52 |
-
|
53 |
-
with torch.no_grad():
|
54 |
-
outputs = model.generate(**input_ids, max_new_tokens=int(max_new_tokens), return_dict_in_generate=False, output_scores=False, use_cache=True, num_beams=1, do_sample=False)
|
55 |
-
"""
|
56 |
-
outputs = model.generate(
|
57 |
-
input_ids=input_ids,
|
58 |
-
max_new_tokens=int(max_new_tokens),
|
59 |
-
do_sample=True,
|
60 |
-
use_cache=True,
|
61 |
-
temperature=temperature,
|
62 |
-
top_k=int(top_k),
|
63 |
-
top_p=top_p,
|
64 |
-
repetition_penalty=repetition_penalty,
|
65 |
-
num_beams=int(num_beams),
|
66 |
-
length_penalty=length_penalty,
|
67 |
-
num_return_sequences=1
|
68 |
-
)
|
69 |
-
"""
|
70 |
-
generated_text = tokenizer.decode(outputs[0])
|
71 |
-
assistant_response = extract_assistant_response(generated_text)
|
72 |
-
|
73 |
-
del input_ids
|
74 |
-
del outputs
|
75 |
-
torch.cuda.empty_cache()
|
76 |
-
|
77 |
-
return assistant_response
|
78 |
-
|
79 |
-
with gr.Blocks() as demo:
|
80 |
-
gr.Markdown("# Zamba2 Model Selector")
|
81 |
-
with gr.Tabs():
|
82 |
-
with gr.TabItem("7B Instruct Model"):
|
83 |
-
gr.Markdown("### Zamba2-7B Instruct Model")
|
84 |
-
with gr.Column():
|
85 |
-
chat_history_7B_instruct = gr.State([])
|
86 |
-
chatbot_7B_instruct = gr.Chatbot()
|
87 |
-
message_7B_instruct = gr.Textbox(lines=2, placeholder="Enter your message...", label="Your Message")
|
88 |
-
with gr.Accordion("Generation Parameters", open=False):
|
89 |
-
max_new_tokens_7B_instruct = gr.Slider(50, 1000, step=50, value=500, label="Max New Tokens")
|
90 |
-
# temperature_7B_instruct = gr.Slider(0.1, 1.5, step=0.1, value=0.2, label="Temperature")
|
91 |
-
# top_k_7B_instruct = gr.Slider(1, 100, step=1, value=50, label="Top K")
|
92 |
-
# top_p_7B_instruct = gr.Slider(0.1, 1.0, step=0.1, value=1.0, label="Top P")
|
93 |
-
# repetition_penalty_7B_instruct = gr.Slider(1.0, 2.0, step=0.1, value=1.2, label="Repetition Penalty")
|
94 |
-
# num_beams_7B_instruct = gr.Slider(1, 10, step=1, value=1, label="Number of Beams")
|
95 |
-
# length_penalty_7B_instruct = gr.Slider(0.0, 2.0, step=0.1, value=1.0, label="Length Penalty")
|
96 |
-
|
97 |
-
def user_message_7B_instruct(message, chat_history):
|
98 |
-
chat_history = chat_history + [[message, None]]
|
99 |
-
return gr.update(value=""), chat_history, chat_history
|
100 |
-
|
101 |
-
def bot_response_7B_instruct(chat_history, max_new_tokens):
|
102 |
-
response = generate_response(chat_history, max_new_tokens, model_7B_instruct, tokenizer_7B_instruct)
|
103 |
-
chat_history[-1][1] = response
|
104 |
-
return chat_history, chat_history
|
105 |
-
|
106 |
-
send_button_7B_instruct = gr.Button("Send")
|
107 |
-
send_button_7B_instruct.click(
|
108 |
-
fn=user_message_7B_instruct,
|
109 |
-
inputs=[message_7B_instruct, chat_history_7B_instruct],
|
110 |
-
outputs=[message_7B_instruct, chat_history_7B_instruct, chatbot_7B_instruct]
|
111 |
-
).then(
|
112 |
-
fn=bot_response_7B_instruct,
|
113 |
-
inputs=[
|
114 |
-
chat_history_7B_instruct,
|
115 |
-
max_new_tokens_7B_instruct
|
116 |
-
],
|
117 |
-
outputs=[chat_history_7B_instruct, chatbot_7B_instruct]
|
118 |
-
)
|
119 |
-
with gr.TabItem("2.7B Instruct Model"):
|
120 |
-
gr.Markdown("### Zamba2-2.7B Instruct Model")
|
121 |
-
with gr.Column():
|
122 |
-
chat_history_2_7B_instruct = gr.State([])
|
123 |
-
chatbot_2_7B_instruct = gr.Chatbot()
|
124 |
-
message_2_7B_instruct = gr.Textbox(lines=2, placeholder="Enter your message...", label="Your Message")
|
125 |
-
with gr.Accordion("Generation Parameters", open=False):
|
126 |
-
max_new_tokens_2_7B_instruct = gr.Slider(50, 1000, step=50, value=500, label="Max New Tokens")
|
127 |
-
# temperature_2_7B_instruct = gr.Slider(0.1, 1.5, step=0.1, value=0.2, label="Temperature")
|
128 |
-
# top_k_2_7B_instruct = gr.Slider(1, 100, step=1, value=50, label="Top K")
|
129 |
-
# top_p_2_7B_instruct = gr.Slider(0.1, 1.0, step=0.1, value=1.0, label="Top P")
|
130 |
-
# repetition_penalty_2_7B_instruct = gr.Slider(1.0, 2.0, step=0.1, value=1.2, label="Repetition Penalty")
|
131 |
-
# num_beams_2_7B_instruct = gr.Slider(1, 10, step=1, value=1, label="Number of Beams")
|
132 |
-
# length_penalty_2_7B_instruct = gr.Slider(0.0, 2.0, step=0.1, value=1.0, label="Length Penalty")
|
133 |
-
|
134 |
-
def user_message_2_7B_instruct(message, chat_history):
|
135 |
-
chat_history = chat_history + [[message, None]]
|
136 |
-
return gr.update(value=""), chat_history, chat_history
|
137 |
-
|
138 |
-
def bot_response_2_7B_instruct(chat_history, max_new_tokens):
|
139 |
-
response = generate_response(chat_history, max_new_tokens, model_2_7B_instruct, tokenizer_2_7B_instruct)
|
140 |
-
chat_history[-1][1] = response
|
141 |
-
return chat_history, chat_history
|
142 |
-
|
143 |
-
send_button_2_7B_instruct = gr.Button("Send")
|
144 |
-
send_button_2_7B_instruct.click(
|
145 |
-
fn=user_message_2_7B_instruct,
|
146 |
-
inputs=[message_2_7B_instruct, chat_history_2_7B_instruct],
|
147 |
-
outputs=[message_2_7B_instruct, chat_history_2_7B_instruct, chatbot_2_7B_instruct]
|
148 |
-
).then(
|
149 |
-
fn=bot_response_2_7B_instruct,
|
150 |
-
inputs=[
|
151 |
-
chat_history_2_7B_instruct,
|
152 |
-
max_new_tokens_2_7B_instruct
|
153 |
-
],
|
154 |
-
outputs=[chat_history_2_7B_instruct, chatbot_2_7B_instruct]
|
155 |
-
)
|
156 |
-
|
157 |
-
if __name__ == "__main__":
|
158 |
-
demo.queue().launch(max_threads=1)
|
159 |
-
'''
|
160 |
-
|
161 |
-
'''
|
162 |
-
import os
|
163 |
-
import gradio as gr
|
164 |
-
from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
|
165 |
-
import torch
|
166 |
-
import threading
|
167 |
-
import re
|
168 |
-
|
169 |
-
model_name_2_7B_instruct = "Zyphra/Zamba2-2.7B-instruct"
|
170 |
-
model_name_7B_instruct = "Zyphra/Zamba2-7B-instruct"
|
171 |
-
max_context_length = 4096
|
172 |
-
|
173 |
-
tokenizer_2_7B_instruct = AutoTokenizer.from_pretrained(model_name_2_7B_instruct)
|
174 |
-
model_2_7B_instruct = AutoModelForCausalLM.from_pretrained(
|
175 |
-
model_name_2_7B_instruct, device_map="cuda", torch_dtype=torch.bfloat16
|
176 |
-
)
|
177 |
-
|
178 |
-
tokenizer_7B_instruct = AutoTokenizer.from_pretrained(model_name_7B_instruct)
|
179 |
-
model_7B_instruct = AutoModelForCausalLM.from_pretrained(
|
180 |
-
model_name_7B_instruct, device_map="cuda", torch_dtype=torch.bfloat16
|
181 |
-
)
|
182 |
-
|
183 |
-
def generate_response(chat_history, max_new_tokens, model, tokenizer):
|
184 |
-
sample = []
|
185 |
-
for turn in chat_history:
|
186 |
-
if turn[0]:
|
187 |
-
sample.append({'role': 'user', 'content': turn[0]})
|
188 |
-
if turn[1]:
|
189 |
-
sample.append({'role': 'assistant', 'content': turn[1]})
|
190 |
-
chat_sample = tokenizer.apply_chat_template(sample, tokenize=False)
|
191 |
-
input_ids = tokenizer(chat_sample, return_tensors='pt', add_special_tokens=False).to(model.device)
|
192 |
-
|
193 |
-
max_new_tokens = int(max_new_tokens)
|
194 |
-
max_input_length = max_context_length - max_new_tokens
|
195 |
-
if input_ids['input_ids'].size(1) > max_input_length:
|
196 |
-
input_ids['input_ids'] = input_ids['input_ids'][:, -max_input_length:]
|
197 |
-
if 'attention_mask' in input_ids:
|
198 |
-
input_ids['attention_mask'] = input_ids['attention_mask'][:, -max_input_length:]
|
199 |
-
|
200 |
-
streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True)
|
201 |
-
generation_kwargs = dict(**input_ids, max_new_tokens=int(max_new_tokens), streamer=streamer)
|
202 |
-
|
203 |
-
thread = threading.Thread(target=model.generate, kwargs=generation_kwargs)
|
204 |
-
thread.start()
|
205 |
-
|
206 |
-
assistant_response = ""
|
207 |
-
|
208 |
-
for new_text in streamer:
|
209 |
-
new_text = re.sub(r'^\s*(?i:assistant)[:\s]*', '', new_text)
|
210 |
-
assistant_response += new_text
|
211 |
-
yield assistant_response
|
212 |
-
|
213 |
-
thread.join()
|
214 |
-
del input_ids
|
215 |
-
torch.cuda.empty_cache()
|
216 |
-
|
217 |
-
with gr.Blocks() as demo:
|
218 |
-
gr.Markdown("# Zamba2 Model Selector")
|
219 |
-
with gr.Tabs():
|
220 |
-
with gr.TabItem("7B Instruct Model"):
|
221 |
-
gr.Markdown("### Zamba2-7B Instruct Model")
|
222 |
-
with gr.Column():
|
223 |
-
chat_history_7B_instruct = gr.State([])
|
224 |
-
chatbot_7B_instruct = gr.Chatbot()
|
225 |
-
message_7B_instruct = gr.Textbox(lines=2, placeholder="Enter your message...", label="Your Message")
|
226 |
-
with gr.Accordion("Generation Parameters", open=False):
|
227 |
-
max_new_tokens_7B_instruct = gr.Slider(50, 1000, step=50, value=500, label="Max New Tokens")
|
228 |
-
|
229 |
-
def user_message_7B_instruct(message, chat_history):
|
230 |
-
chat_history = chat_history + [[message, None]]
|
231 |
-
return gr.update(value=""), chat_history, chat_history
|
232 |
-
|
233 |
-
def bot_response_7B_instruct(chat_history, max_new_tokens):
|
234 |
-
assistant_response_generator = generate_response(chat_history, max_new_tokens, model_7B_instruct, tokenizer_7B_instruct)
|
235 |
-
for assistant_response in assistant_response_generator:
|
236 |
-
chat_history[-1][1] = assistant_response
|
237 |
-
yield chat_history
|
238 |
-
|
239 |
-
send_button_7B_instruct = gr.Button("Send")
|
240 |
-
send_button_7B_instruct.click(
|
241 |
-
fn=user_message_7B_instruct,
|
242 |
-
inputs=[message_7B_instruct, chat_history_7B_instruct],
|
243 |
-
outputs=[message_7B_instruct, chat_history_7B_instruct, chatbot_7B_instruct]
|
244 |
-
).then(
|
245 |
-
fn=bot_response_7B_instruct,
|
246 |
-
inputs=[chat_history_7B_instruct, max_new_tokens_7B_instruct],
|
247 |
-
outputs=chatbot_7B_instruct,
|
248 |
-
)
|
249 |
-
|
250 |
-
with gr.TabItem("2.7B Instruct Model"):
|
251 |
-
gr.Markdown("### Zamba2-2.7B Instruct Model")
|
252 |
-
with gr.Column():
|
253 |
-
chat_history_2_7B_instruct = gr.State([])
|
254 |
-
chatbot_2_7B_instruct = gr.Chatbot()
|
255 |
-
message_2_7B_instruct = gr.Textbox(lines=2, placeholder="Enter your message...", label="Your Message")
|
256 |
-
with gr.Accordion("Generation Parameters", open=False):
|
257 |
-
max_new_tokens_2_7B_instruct = gr.Slider(50, 1000, step=50, value=500, label="Max New Tokens")
|
258 |
-
|
259 |
-
def user_message_2_7B_instruct(message, chat_history):
|
260 |
-
chat_history = chat_history + [[message, None]]
|
261 |
-
return gr.update(value=""), chat_history, chat_history
|
262 |
-
|
263 |
-
def bot_response_2_7B_instruct(chat_history, max_new_tokens):
|
264 |
-
assistant_response_generator = generate_response(chat_history, max_new_tokens, model_2_7B_instruct, tokenizer_2_7B_instruct)
|
265 |
-
for assistant_response in assistant_response_generator:
|
266 |
-
chat_history[-1][1] = assistant_response
|
267 |
-
yield chat_history
|
268 |
-
|
269 |
-
send_button_2_7B_instruct = gr.Button("Send")
|
270 |
-
send_button_2_7B_instruct.click(
|
271 |
-
fn=user_message_2_7B_instruct,
|
272 |
-
inputs=[message_2_7B_instruct, chat_history_2_7B_instruct],
|
273 |
-
outputs=[message_2_7B_instruct, chat_history_2_7B_instruct, chatbot_2_7B_instruct]
|
274 |
-
).then(
|
275 |
-
fn=bot_response_2_7B_instruct,
|
276 |
-
inputs=[chat_history_2_7B_instruct, max_new_tokens_2_7B_instruct],
|
277 |
-
outputs=chatbot_2_7B_instruct,
|
278 |
-
)
|
279 |
-
|
280 |
-
if __name__ == "__main__":
|
281 |
-
demo.queue().launch(max_threads=1)
|
282 |
-
'''
|
283 |
-
|
284 |
import os
|
285 |
import gradio as gr
|
286 |
from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
import os
|
2 |
import gradio as gr
|
3 |
from transformers import AutoTokenizer, AutoModelForCausalLM, TextIteratorStreamer
|