Update app.py
Browse files
app.py
CHANGED
@@ -9,21 +9,39 @@ import spaces
|
|
9 |
import torch
|
10 |
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
11 |
|
|
|
|
|
|
|
12 |
HF_TOKEN = os.environ.get("HF_TOKEN")
|
|
|
|
|
13 |
|
14 |
DESCRIPTION = "# Mistral-7B v0.2"
|
15 |
|
16 |
if not torch.cuda.is_available():
|
17 |
DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
|
|
|
|
|
|
|
18 |
|
19 |
MAX_MAX_NEW_TOKENS = 2048
|
20 |
DEFAULT_MAX_NEW_TOKENS = 1024
|
21 |
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
22 |
|
|
|
23 |
if torch.cuda.is_available():
|
24 |
model_id = "mistralai/Mistral-7B-Instruct-v0.2"
|
25 |
-
|
26 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
27 |
|
28 |
|
29 |
@spaces.GPU
|
@@ -36,36 +54,54 @@ def generate(
|
|
36 |
top_k: int = 50,
|
37 |
repetition_penalty: float = 1.2,
|
38 |
) -> Iterator[str]:
|
|
|
|
|
|
|
39 |
conversation = []
|
40 |
for user, assistant in chat_history:
|
41 |
conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
|
42 |
conversation.append({"role": "user", "content": message})
|
43 |
|
44 |
-
|
45 |
-
|
46 |
-
input_ids =
|
47 |
-
|
48 |
-
|
49 |
-
|
50 |
-
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
|
63 |
-
|
64 |
-
|
65 |
-
|
66 |
-
|
67 |
-
|
68 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
69 |
|
70 |
|
71 |
chat_interface = gr.ChatInterface(
|
@@ -117,6 +153,9 @@ chat_interface = gr.ChatInterface(
|
|
117 |
],
|
118 |
)
|
119 |
|
|
|
|
|
|
|
120 |
with gr.Blocks(css="style.css") as demo:
|
121 |
gr.Markdown(DESCRIPTION)
|
122 |
gr.DuplicateButton(
|
@@ -126,57 +165,193 @@ with gr.Blocks(css="style.css") as demo:
|
|
126 |
)
|
127 |
chat_interface.render()
|
128 |
|
|
|
|
|
|
|
129 |
if __name__ == "__main__":
|
130 |
demo.queue(max_size=20).launch(share=True)
|
131 |
|
132 |
-
|
133 |
-
|
134 |
-
|
135 |
-
|
136 |
-
|
137 |
-
|
138 |
-
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
143 |
-
|
144 |
-
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
|
149 |
-
|
150 |
-
|
151 |
-
|
152 |
-
|
153 |
-
|
154 |
-
|
155 |
-
|
156 |
-
|
157 |
-
|
158 |
-
|
159 |
-
|
160 |
-
|
161 |
-
|
162 |
-
|
163 |
-
|
164 |
-
|
165 |
-
|
166 |
-
|
167 |
-
|
168 |
-
|
169 |
-
|
170 |
-
|
171 |
-
|
172 |
-
|
173 |
-
|
174 |
-
|
175 |
-
|
176 |
-
|
177 |
-
|
178 |
-
|
179 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
180 |
|
181 |
|
182 |
|
|
|
9 |
import torch
|
10 |
from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
11 |
|
12 |
+
# Debugging: Start script
|
13 |
+
print("Starting script...")
|
14 |
+
|
15 |
HF_TOKEN = os.environ.get("HF_TOKEN")
|
16 |
+
if HF_TOKEN is None:
|
17 |
+
print("Warning: HF_TOKEN is not set!")
|
18 |
|
19 |
DESCRIPTION = "# Mistral-7B v0.2"
|
20 |
|
21 |
if not torch.cuda.is_available():
|
22 |
DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
|
23 |
+
print("Warning: No GPU available. This model cannot run on CPU.")
|
24 |
+
else:
|
25 |
+
print("GPU is available!")
|
26 |
|
27 |
MAX_MAX_NEW_TOKENS = 2048
|
28 |
DEFAULT_MAX_NEW_TOKENS = 1024
|
29 |
MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
30 |
|
31 |
+
# Debugging: GPU check passed, loading model
|
32 |
if torch.cuda.is_available():
|
33 |
model_id = "mistralai/Mistral-7B-Instruct-v0.2"
|
34 |
+
try:
|
35 |
+
print("Loading model...")
|
36 |
+
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto", token=HF_TOKEN)
|
37 |
+
print("Model loaded successfully!")
|
38 |
+
|
39 |
+
print("Loading tokenizer...")
|
40 |
+
tokenizer = AutoTokenizer.from_pretrained(model_id, token=HF_TOKEN)
|
41 |
+
print("Tokenizer loaded successfully!")
|
42 |
+
except Exception as e:
|
43 |
+
print(f"Error loading model or tokenizer: {e}")
|
44 |
+
raise e # Re-raise the error after logging it
|
45 |
|
46 |
|
47 |
@spaces.GPU
|
|
|
54 |
top_k: int = 50,
|
55 |
repetition_penalty: float = 1.2,
|
56 |
) -> Iterator[str]:
|
57 |
+
print(f"Received message: {message}")
|
58 |
+
print(f"Chat history: {chat_history}")
|
59 |
+
|
60 |
conversation = []
|
61 |
for user, assistant in chat_history:
|
62 |
conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
|
63 |
conversation.append({"role": "user", "content": message})
|
64 |
|
65 |
+
try:
|
66 |
+
print("Tokenizing input...")
|
67 |
+
input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt")
|
68 |
+
print(f"Input tokenized: {input_ids.shape}")
|
69 |
+
|
70 |
+
if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
|
71 |
+
input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
|
72 |
+
gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
|
73 |
+
print("Trimmed input tokens due to length.")
|
74 |
+
|
75 |
+
input_ids = input_ids.to(model.device)
|
76 |
+
print("Input moved to the model's device.")
|
77 |
+
|
78 |
+
streamer = TextIteratorStreamer(tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True)
|
79 |
+
generate_kwargs = dict(
|
80 |
+
{"input_ids": input_ids},
|
81 |
+
streamer=streamer,
|
82 |
+
max_new_tokens=max_new_tokens,
|
83 |
+
do_sample=True,
|
84 |
+
top_p=top_p,
|
85 |
+
top_k=top_k,
|
86 |
+
temperature=temperature,
|
87 |
+
num_beams=1,
|
88 |
+
repetition_penalty=repetition_penalty,
|
89 |
+
)
|
90 |
+
|
91 |
+
print("Starting generation...")
|
92 |
+
t = Thread(target=model.generate, kwargs=generate_kwargs)
|
93 |
+
t.start()
|
94 |
+
print("Thread started for model generation.")
|
95 |
+
|
96 |
+
outputs = []
|
97 |
+
for text in streamer:
|
98 |
+
outputs.append(text)
|
99 |
+
print(f"Generated text so far: {''.join(outputs)}")
|
100 |
+
yield "".join(outputs)
|
101 |
+
|
102 |
+
except Exception as e:
|
103 |
+
print(f"Error during generation: {e}")
|
104 |
+
raise e # Re-raise the error after logging it
|
105 |
|
106 |
|
107 |
chat_interface = gr.ChatInterface(
|
|
|
153 |
],
|
154 |
)
|
155 |
|
156 |
+
# Debugging: Interface setup
|
157 |
+
print("Setting up interface...")
|
158 |
+
|
159 |
with gr.Blocks(css="style.css") as demo:
|
160 |
gr.Markdown(DESCRIPTION)
|
161 |
gr.DuplicateButton(
|
|
|
165 |
)
|
166 |
chat_interface.render()
|
167 |
|
168 |
+
# Debugging: Starting queue and launching the demo
|
169 |
+
print("Launching demo...")
|
170 |
+
|
171 |
if __name__ == "__main__":
|
172 |
demo.queue(max_size=20).launch(share=True)
|
173 |
|
174 |
+
|
175 |
+
|
176 |
+
#!/usr/bin/env python
|
177 |
+
|
178 |
+
# import os
|
179 |
+
# from threading import Thread
|
180 |
+
# from typing import Iterator
|
181 |
+
|
182 |
+
# import gradio as gr
|
183 |
+
# import spaces
|
184 |
+
# import torch
|
185 |
+
# from transformers import AutoModelForCausalLM, AutoTokenizer, TextIteratorStreamer
|
186 |
+
|
187 |
+
# HF_TOKEN = os.environ.get("HF_TOKEN")
|
188 |
+
|
189 |
+
# DESCRIPTION = "# Mistral-7B v0.2"
|
190 |
+
|
191 |
+
# if not torch.cuda.is_available():
|
192 |
+
# DESCRIPTION += "\n<p>Running on CPU 🥶 This demo does not work on CPU.</p>"
|
193 |
+
|
194 |
+
# MAX_MAX_NEW_TOKENS = 2048
|
195 |
+
# DEFAULT_MAX_NEW_TOKENS = 1024
|
196 |
+
# MAX_INPUT_TOKEN_LENGTH = int(os.getenv("MAX_INPUT_TOKEN_LENGTH", "4096"))
|
197 |
+
|
198 |
+
# if torch.cuda.is_available():
|
199 |
+
# model_id = "mistralai/Mistral-7B-Instruct-v0.2"
|
200 |
+
# model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype=torch.float16, device_map="auto", token=HF_TOKEN)
|
201 |
+
# tokenizer = AutoTokenizer.from_pretrained(model_id, token=HF_TOKEN)
|
202 |
+
|
203 |
+
|
204 |
+
# @spaces.GPU
|
205 |
+
# def generate(
|
206 |
+
# message: str,
|
207 |
+
# chat_history: list[tuple[str, str]],
|
208 |
+
# max_new_tokens: int = 1024,
|
209 |
+
# temperature: float = 0.6,
|
210 |
+
# top_p: float = 0.9,
|
211 |
+
# top_k: int = 50,
|
212 |
+
# repetition_penalty: float = 1.2,
|
213 |
+
# ) -> Iterator[str]:
|
214 |
+
# conversation = []
|
215 |
+
# for user, assistant in chat_history:
|
216 |
+
# conversation.extend([{"role": "user", "content": user}, {"role": "assistant", "content": assistant}])
|
217 |
+
# conversation.append({"role": "user", "content": message})
|
218 |
+
|
219 |
+
# input_ids = tokenizer.apply_chat_template(conversation, return_tensors="pt")
|
220 |
+
# if input_ids.shape[1] > MAX_INPUT_TOKEN_LENGTH:
|
221 |
+
# input_ids = input_ids[:, -MAX_INPUT_TOKEN_LENGTH:]
|
222 |
+
# gr.Warning(f"Trimmed input from conversation as it was longer than {MAX_INPUT_TOKEN_LENGTH} tokens.")
|
223 |
+
# input_ids = input_ids.to(model.device)
|
224 |
+
|
225 |
+
# streamer = TextIteratorStreamer(tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True)
|
226 |
+
# generate_kwargs = dict(
|
227 |
+
# {"input_ids": input_ids},
|
228 |
+
# streamer=streamer,
|
229 |
+
# max_new_tokens=max_new_tokens,
|
230 |
+
# do_sample=True,
|
231 |
+
# top_p=top_p,
|
232 |
+
# top_k=top_k,
|
233 |
+
# temperature=temperature,
|
234 |
+
# num_beams=1,
|
235 |
+
# repetition_penalty=repetition_penalty,
|
236 |
+
# )
|
237 |
+
# t = Thread(target=model.generate, kwargs=generate_kwargs)
|
238 |
+
# t.start()
|
239 |
+
|
240 |
+
# outputs = []
|
241 |
+
# for text in streamer:
|
242 |
+
# outputs.append(text)
|
243 |
+
# yield "".join(outputs)
|
244 |
+
|
245 |
+
|
246 |
+
# chat_interface = gr.ChatInterface(
|
247 |
+
# fn=generate,
|
248 |
+
# additional_inputs=[
|
249 |
+
# gr.Slider(
|
250 |
+
# label="Max new tokens",
|
251 |
+
# minimum=1,
|
252 |
+
# maximum=MAX_MAX_NEW_TOKENS,
|
253 |
+
# step=1,
|
254 |
+
# value=DEFAULT_MAX_NEW_TOKENS,
|
255 |
+
# ),
|
256 |
+
# gr.Slider(
|
257 |
+
# label="Temperature",
|
258 |
+
# minimum=0.1,
|
259 |
+
# maximum=4.0,
|
260 |
+
# step=0.1,
|
261 |
+
# value=0.6,
|
262 |
+
# ),
|
263 |
+
# gr.Slider(
|
264 |
+
# label="Top-p (nucleus sampling)",
|
265 |
+
# minimum=0.05,
|
266 |
+
# maximum=1.0,
|
267 |
+
# step=0.05,
|
268 |
+
# value=0.9,
|
269 |
+
# ),
|
270 |
+
# gr.Slider(
|
271 |
+
# label="Top-k",
|
272 |
+
# minimum=1,
|
273 |
+
# maximum=1000,
|
274 |
+
# step=1,
|
275 |
+
# value=50,
|
276 |
+
# ),
|
277 |
+
# gr.Slider(
|
278 |
+
# label="Repetition penalty",
|
279 |
+
# minimum=1.0,
|
280 |
+
# maximum=2.0,
|
281 |
+
# step=0.05,
|
282 |
+
# value=1.2,
|
283 |
+
# ),
|
284 |
+
# ],
|
285 |
+
# stop_btn=None,
|
286 |
+
# examples=[
|
287 |
+
# ["Hello there! How are you doing?"],
|
288 |
+
# ["Can you explain briefly to me what is the Python programming language?"],
|
289 |
+
# ["Explain the plot of Cinderella in a sentence."],
|
290 |
+
# ["How many hours does it take a man to eat a Helicopter?"],
|
291 |
+
# ["Write a 100-word article on 'Benefits of Open-Source in AI research'"],
|
292 |
+
# ],
|
293 |
+
# )
|
294 |
+
|
295 |
+
# with gr.Blocks(css="style.css") as demo:
|
296 |
+
# gr.Markdown(DESCRIPTION)
|
297 |
+
# gr.DuplicateButton(
|
298 |
+
# value="Duplicate Space for private use",
|
299 |
+
# elem_id="duplicate-button",
|
300 |
+
# visible=os.getenv("SHOW_DUPLICATE_BUTTON") == "1",
|
301 |
+
# )
|
302 |
+
# chat_interface.render()
|
303 |
+
|
304 |
+
# if __name__ == "__main__":
|
305 |
+
# demo.queue(max_size=20).launch(share=True)
|
306 |
+
|
307 |
+
# gr.ChatInterface(
|
308 |
+
# fn=generate,
|
309 |
+
# additional_inputs=[
|
310 |
+
# gr.Slider(
|
311 |
+
# label="Max new tokens",
|
312 |
+
# minimum=1,
|
313 |
+
# maximum=MAX_MAX_NEW_TOKENS,
|
314 |
+
# step=1,
|
315 |
+
# value=DEFAULT_MAX_NEW_TOKENS,
|
316 |
+
# ),
|
317 |
+
# gr.Slider(
|
318 |
+
# label="Temperature",
|
319 |
+
# minimum=0.1,
|
320 |
+
# maximum=4.0,
|
321 |
+
# step=0.1,
|
322 |
+
# value=0.6,
|
323 |
+
# ),
|
324 |
+
# gr.Slider(
|
325 |
+
# label="Top-p (nucleus sampling)",
|
326 |
+
# minimum=0.05,
|
327 |
+
# maximum=1.0,
|
328 |
+
# step=0.05,
|
329 |
+
# value=0.9,
|
330 |
+
# ),
|
331 |
+
# gr.Slider(
|
332 |
+
# label="Top-k",
|
333 |
+
# minimum=1,
|
334 |
+
# maximum=1000,
|
335 |
+
# step=1,
|
336 |
+
# value=50,
|
337 |
+
# ),
|
338 |
+
# gr.Slider(
|
339 |
+
# label="Repetition penalty",
|
340 |
+
# minimum=1.0,
|
341 |
+
# maximum=2.0,
|
342 |
+
# step=0.05,
|
343 |
+
# value=1.2,
|
344 |
+
# ),
|
345 |
+
# ],
|
346 |
+
# stop_btn=None,
|
347 |
+
# examples=[
|
348 |
+
# ["Hello there! How are you doing?"],
|
349 |
+
# ["Can you explain briefly to me what is the Python programming language?"],
|
350 |
+
# ["Explain the plot of Cinderella in a sentence."],
|
351 |
+
# ["How many hours does it take a man to eat a Helicopter?"],
|
352 |
+
# ["Write a 100-word article on 'Benefits of Open-Source in AI research'"],
|
353 |
+
# ],
|
354 |
+
# ).launch(share=True)
|
355 |
|
356 |
|
357 |
|