Spaces:
Running
on
Zero
Running
on
Zero
yuexiang96
commited on
Commit
•
17c6e95
1
Parent(s):
70f1fe7
Update app.py
Browse files
app.py
CHANGED
@@ -62,6 +62,7 @@ repo_name = os.environ["LOG_REPO"]
|
|
62 |
|
63 |
external_log_dir = "./logs"
|
64 |
LOGDIR = external_log_dir
|
|
|
65 |
|
66 |
|
67 |
def install_gradio_4_35_0():
|
@@ -87,6 +88,38 @@ def get_conv_log_filename():
|
|
87 |
name = os.path.join(LOGDIR, f"{t.year}-{t.month:02d}-{t.day:02d}-user_conv.json")
|
88 |
return name
|
89 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
90 |
class InferenceDemo(object):
|
91 |
def __init__(
|
92 |
self, args, model_path, tokenizer, model, image_processor, context_len
|
@@ -125,6 +158,22 @@ class InferenceDemo(object):
|
|
125 |
self.conversation = conv_templates[args.conv_mode].copy()
|
126 |
self.num_frames = args.num_frames
|
127 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
128 |
|
129 |
def is_valid_video_filename(name):
|
130 |
video_extensions = ["avi", "mp4", "mov", "mkv", "flv", "wmv", "mjpeg"]
|
@@ -178,13 +227,6 @@ def load_image(image_file):
|
|
178 |
return image
|
179 |
|
180 |
|
181 |
-
def clear_history(history):
|
182 |
-
|
183 |
-
our_chatbot.conversation = conv_templates[our_chatbot.conv_mode].copy()
|
184 |
-
|
185 |
-
return None
|
186 |
-
|
187 |
-
|
188 |
def clear_response(history):
|
189 |
for index_conv in range(1, len(history)):
|
190 |
# loop until get a text response from our model.
|
@@ -195,40 +237,69 @@ def clear_response(history):
|
|
195 |
history = history[:-index_conv]
|
196 |
return history, question
|
197 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
198 |
|
199 |
-
# def print_like_dislike(x: gr.LikeData):
|
200 |
-
# print(x.index, x.value, x.liked)
|
201 |
|
202 |
|
203 |
def add_message(history, message):
|
204 |
-
|
205 |
-
|
206 |
-
|
207 |
-
our_chatbot =
|
208 |
-
|
209 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
210 |
|
211 |
-
|
212 |
-
history.append(((x,), None))
|
213 |
-
if message["text"] is not None:
|
214 |
-
history.append((message["text"], None))
|
215 |
-
return history, gr.MultimodalTextbox(value=None, interactive=False)
|
216 |
|
217 |
|
218 |
@spaces.GPU
|
219 |
def bot(history, temperature, top_p, max_output_tokens):
|
220 |
-
|
221 |
-
print("###
|
222 |
text = history[-1][0]
|
223 |
images_this_term = []
|
224 |
text_this_term = ""
|
225 |
-
|
226 |
num_new_images = 0
|
|
|
227 |
for i, message in enumerate(history[:-1]):
|
228 |
if type(message[0]) is tuple:
|
|
|
|
|
|
|
|
|
|
|
229 |
images_this_term.append(message[0][0])
|
230 |
if is_valid_video_filename(message[0][0]):
|
231 |
-
# 不接受视频
|
232 |
raise ValueError("Video is not supported")
|
233 |
num_new_images += our_chatbot.num_frames
|
234 |
elif is_valid_image_filename(message[0][0]):
|
@@ -236,15 +307,10 @@ def bot(history, temperature, top_p, max_output_tokens):
|
|
236 |
num_new_images += 1
|
237 |
else:
|
238 |
raise ValueError("Invalid image file")
|
|
|
239 |
else:
|
240 |
num_new_images = 0
|
241 |
-
|
242 |
-
# for message in history[-i-1:]:
|
243 |
-
# images_this_term.append(message[0][0])
|
244 |
-
|
245 |
-
assert len(images_this_term) > 0, "must have an image"
|
246 |
-
# image_files = (args.image_file).split(',')
|
247 |
-
# image = [load_image(f) for f in images_this_term if f]
|
248 |
|
249 |
all_image_hash = []
|
250 |
all_image_path = []
|
@@ -288,9 +354,7 @@ def bot(history, temperature, top_p, max_output_tokens):
|
|
288 |
|
289 |
image_tensor = torch.stack(image_tensor)
|
290 |
image_token = DEFAULT_IMAGE_TOKEN * num_new_images
|
291 |
-
|
292 |
-
# inp = DEFAULT_IM_START_TOKEN + image_token + DEFAULT_IM_END_TOKEN + "\n" + inp
|
293 |
-
# else:
|
294 |
inp = text
|
295 |
inp = image_token + "\n" + inp
|
296 |
our_chatbot.conversation.append_message(our_chatbot.conversation.roles[0], inp)
|
@@ -298,13 +362,6 @@ def bot(history, temperature, top_p, max_output_tokens):
|
|
298 |
our_chatbot.conversation.append_message(our_chatbot.conversation.roles[1], None)
|
299 |
prompt = our_chatbot.conversation.get_prompt()
|
300 |
|
301 |
-
# input_ids = (
|
302 |
-
# tokenizer_image_token(
|
303 |
-
# prompt, our_chatbot.tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt"
|
304 |
-
# )
|
305 |
-
# .unsqueeze(0)
|
306 |
-
# .to(our_chatbot.model.device)
|
307 |
-
# )
|
308 |
input_ids = tokenizer_image_token(
|
309 |
prompt, our_chatbot.tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt"
|
310 |
).unsqueeze(0).to(our_chatbot.model.device)
|
@@ -318,9 +375,7 @@ def bot(history, temperature, top_p, max_output_tokens):
|
|
318 |
stopping_criteria = KeywordsStoppingCriteria(
|
319 |
keywords, our_chatbot.tokenizer, input_ids
|
320 |
)
|
321 |
-
|
322 |
-
# our_chatbot.tokenizer, skip_prompt=True, skip_special_tokens=True
|
323 |
-
# )
|
324 |
streamer = TextIteratorStreamer(
|
325 |
our_chatbot.tokenizer, skip_prompt=True, skip_special_tokens=True
|
326 |
)
|
@@ -328,27 +383,6 @@ def bot(history, temperature, top_p, max_output_tokens):
|
|
328 |
print(input_ids.device)
|
329 |
print(image_tensor.device)
|
330 |
|
331 |
-
# with torch.inference_mode():
|
332 |
-
# output_ids = our_chatbot.model.generate(
|
333 |
-
# input_ids,
|
334 |
-
# images=image_tensor,
|
335 |
-
# do_sample=True,
|
336 |
-
# temperature=0.7,
|
337 |
-
# top_p=1.0,
|
338 |
-
# max_new_tokens=4096,
|
339 |
-
# streamer=streamer,
|
340 |
-
# use_cache=False,
|
341 |
-
# stopping_criteria=[stopping_criteria],
|
342 |
-
# )
|
343 |
-
|
344 |
-
# outputs = our_chatbot.tokenizer.decode(output_ids[0]).strip()
|
345 |
-
# if outputs.endswith(stop_str):
|
346 |
-
# outputs = outputs[: -len(stop_str)]
|
347 |
-
# our_chatbot.conversation.messages[-1][-1] = outputs
|
348 |
-
|
349 |
-
# history[-1] = [text, outputs]
|
350 |
-
|
351 |
-
# return history
|
352 |
generate_kwargs = dict(
|
353 |
inputs=input_ids,
|
354 |
streamer=streamer,
|
@@ -367,13 +401,12 @@ def bot(history, temperature, top_p, max_output_tokens):
|
|
367 |
outputs = []
|
368 |
for stream_token in streamer:
|
369 |
outputs.append(stream_token)
|
370 |
-
|
371 |
-
# our_chatbot.conversation.messages[-1][-1] = "".join(outputs)
|
372 |
history[-1] = [text, "".join(outputs)]
|
373 |
yield history
|
374 |
our_chatbot.conversation.messages[-1][-1] = "".join(outputs)
|
375 |
-
print("### turn end history", history)
|
376 |
-
print("### turn end conv",our_chatbot.conversation)
|
377 |
|
378 |
with open(get_conv_log_filename(), "a") as fout:
|
379 |
data = {
|
@@ -637,17 +670,25 @@ with gr.Blocks(
|
|
637 |
gr.Markdown(learn_more_markdown)
|
638 |
gr.Markdown(bibtext)
|
639 |
|
640 |
-
|
641 |
-
|
642 |
-
)
|
643 |
-
|
644 |
-
bot_msg.then(lambda: gr.MultimodalTextbox(interactive=True), None, [chat_input])
|
645 |
|
646 |
# chatbot.like(print_like_dislike, None, None)
|
647 |
clear_btn.click(
|
648 |
fn=clear_history, inputs=[chatbot], outputs=[chatbot], api_name="clear_all"
|
649 |
)
|
650 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
651 |
|
652 |
demo.queue()
|
653 |
|
@@ -678,5 +719,5 @@ if __name__ == "__main__":
|
|
678 |
model_name = get_model_name_from_path(args.model_path)
|
679 |
tokenizer, model, image_processor, context_len = load_pretrained_model(args.model_path, args.model_base, model_name, args.load_8bit, args.load_4bit)
|
680 |
model=model.to(torch.device('cuda'))
|
681 |
-
|
682 |
demo.launch()
|
|
|
62 |
|
63 |
external_log_dir = "./logs"
|
64 |
LOGDIR = external_log_dir
|
65 |
+
VOTEDIR = "./votes"
|
66 |
|
67 |
|
68 |
def install_gradio_4_35_0():
|
|
|
88 |
name = os.path.join(LOGDIR, f"{t.year}-{t.month:02d}-{t.day:02d}-user_conv.json")
|
89 |
return name
|
90 |
|
91 |
+
def get_conv_vote_filename():
|
92 |
+
t = datetime.datetime.now()
|
93 |
+
name = os.path.join(VOTEDIR, f"{t.year}-{t.month:02d}-{t.day:02d}-user_vote.json")
|
94 |
+
if not os.path.isfile(name):
|
95 |
+
os.makedirs(os.path.dirname(name), exist_ok=True)
|
96 |
+
return name
|
97 |
+
|
98 |
+
def vote_last_response(state, vote_type, model_selector):
|
99 |
+
with open(get_conv_vote_filename(), "a") as fout:
|
100 |
+
data = {
|
101 |
+
"type": vote_type,
|
102 |
+
"model": model_selector,
|
103 |
+
"state": state,
|
104 |
+
}
|
105 |
+
fout.write(json.dumps(data) + "\n")
|
106 |
+
api.upload_file(
|
107 |
+
path_or_fileobj=get_conv_vote_filename(),
|
108 |
+
path_in_repo=get_conv_vote_filename().replace("./votes/", ""),
|
109 |
+
repo_id=repo_name,
|
110 |
+
repo_type="dataset")
|
111 |
+
|
112 |
+
|
113 |
+
def upvote_last_response(state):
|
114 |
+
vote_last_response(state, "upvote", "Pangea-7b")
|
115 |
+
gr.Info("Thank you for your voting!")
|
116 |
+
return state
|
117 |
+
|
118 |
+
def downvote_last_response(state):
|
119 |
+
vote_last_response(state, "downvote", "Pangea-7b")
|
120 |
+
gr.Info("Thank you for your voting!")
|
121 |
+
return state
|
122 |
+
|
123 |
class InferenceDemo(object):
|
124 |
def __init__(
|
125 |
self, args, model_path, tokenizer, model, image_processor, context_len
|
|
|
158 |
self.conversation = conv_templates[args.conv_mode].copy()
|
159 |
self.num_frames = args.num_frames
|
160 |
|
161 |
+
class ChatSessionManager:
|
162 |
+
def __init__(self):
|
163 |
+
self.chatbot_instance = None
|
164 |
+
|
165 |
+
def initialize_chatbot(self, args, model_path, tokenizer, model, image_processor, context_len):
|
166 |
+
self.chatbot_instance = InferenceDemo(args, model_path, tokenizer, model, image_processor, context_len)
|
167 |
+
print(f"Initialized Chatbot instance with ID: {id(self.chatbot_instance)}")
|
168 |
+
|
169 |
+
def reset_chatbot(self):
|
170 |
+
self.chatbot_instance = None
|
171 |
+
|
172 |
+
def get_chatbot(self, args, model_path, tokenizer, model, image_processor, context_len):
|
173 |
+
if self.chatbot_instance is None:
|
174 |
+
self.initialize_chatbot(args, model_path, tokenizer, model, image_processor, context_len)
|
175 |
+
return self.chatbot_instance
|
176 |
+
|
177 |
|
178 |
def is_valid_video_filename(name):
|
179 |
video_extensions = ["avi", "mp4", "mov", "mkv", "flv", "wmv", "mjpeg"]
|
|
|
227 |
return image
|
228 |
|
229 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
230 |
def clear_response(history):
|
231 |
for index_conv in range(1, len(history)):
|
232 |
# loop until get a text response from our model.
|
|
|
237 |
history = history[:-index_conv]
|
238 |
return history, question
|
239 |
|
240 |
+
chat_manager = ChatSessionManager()
|
241 |
+
|
242 |
+
|
243 |
+
def clear_history(history):
|
244 |
+
chatbot_instance = chat_manager.get_chatbot(args, model_path, tokenizer, model, image_processor, context_len)
|
245 |
+
chatbot_instance.conversation = conv_templates[chatbot_instance.conv_mode].copy()
|
246 |
+
return None
|
247 |
|
|
|
|
|
248 |
|
249 |
|
250 |
def add_message(history, message):
|
251 |
+
global chat_image_num
|
252 |
+
if not history:
|
253 |
+
history = []
|
254 |
+
our_chatbot = chat_manager.get_chatbot(args, model_path, tokenizer, model, image_processor, context_len)
|
255 |
+
chat_image_num = 0
|
256 |
+
|
257 |
+
if len(message["files"]) <= 1:
|
258 |
+
for x in message["files"]:
|
259 |
+
history.append(((x,), None))
|
260 |
+
chat_image_num += 1
|
261 |
+
if chat_image_num > 1:
|
262 |
+
history = []
|
263 |
+
chat_manager.reset_chatbot()
|
264 |
+
our_chatbot = chat_manager.get_chatbot(args, model_path, tokenizer, model, image_processor, context_len)
|
265 |
+
chat_image_num = 0
|
266 |
+
for x in message["files"]:
|
267 |
+
history.append(((x,), None))
|
268 |
+
chat_image_num += 1
|
269 |
+
|
270 |
+
if message["text"] is not None:
|
271 |
+
history.append((message["text"], None))
|
272 |
+
|
273 |
+
print(f"### Chatbot instance ID: {id(our_chatbot)}")
|
274 |
+
return history, gr.MultimodalTextbox(value=None, interactive=False)
|
275 |
+
else:
|
276 |
+
for x in message["files"]:
|
277 |
+
history.append(((x,), None))
|
278 |
+
if message["text"] is not None:
|
279 |
+
history.append((message["text"], None))
|
280 |
|
281 |
+
return history, gr.MultimodalTextbox(value=None, interactive=False)
|
|
|
|
|
|
|
|
|
282 |
|
283 |
|
284 |
@spaces.GPU
|
285 |
def bot(history, temperature, top_p, max_output_tokens):
|
286 |
+
our_chatbot = chat_manager.get_chatbot(args, model_path, tokenizer, model, image_processor, context_len)
|
287 |
+
print(f"### Chatbot instance ID: {id(our_chatbot)}")
|
288 |
text = history[-1][0]
|
289 |
images_this_term = []
|
290 |
text_this_term = ""
|
291 |
+
|
292 |
num_new_images = 0
|
293 |
+
previous_image = False
|
294 |
for i, message in enumerate(history[:-1]):
|
295 |
if type(message[0]) is tuple:
|
296 |
+
if previous_image:
|
297 |
+
gr.Warning("Only one image can be uploaded in a conversation. Please reduce the number of images and start a new conversation.")
|
298 |
+
our_chatbot.conversation = conv_templates[our_chatbot.conv_mode].copy()
|
299 |
+
return None
|
300 |
+
|
301 |
images_this_term.append(message[0][0])
|
302 |
if is_valid_video_filename(message[0][0]):
|
|
|
303 |
raise ValueError("Video is not supported")
|
304 |
num_new_images += our_chatbot.num_frames
|
305 |
elif is_valid_image_filename(message[0][0]):
|
|
|
307 |
num_new_images += 1
|
308 |
else:
|
309 |
raise ValueError("Invalid image file")
|
310 |
+
previous_image = True
|
311 |
else:
|
312 |
num_new_images = 0
|
313 |
+
previous_image = False
|
|
|
|
|
|
|
|
|
|
|
|
|
314 |
|
315 |
all_image_hash = []
|
316 |
all_image_path = []
|
|
|
354 |
|
355 |
image_tensor = torch.stack(image_tensor)
|
356 |
image_token = DEFAULT_IMAGE_TOKEN * num_new_images
|
357 |
+
|
|
|
|
|
358 |
inp = text
|
359 |
inp = image_token + "\n" + inp
|
360 |
our_chatbot.conversation.append_message(our_chatbot.conversation.roles[0], inp)
|
|
|
362 |
our_chatbot.conversation.append_message(our_chatbot.conversation.roles[1], None)
|
363 |
prompt = our_chatbot.conversation.get_prompt()
|
364 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
365 |
input_ids = tokenizer_image_token(
|
366 |
prompt, our_chatbot.tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt"
|
367 |
).unsqueeze(0).to(our_chatbot.model.device)
|
|
|
375 |
stopping_criteria = KeywordsStoppingCriteria(
|
376 |
keywords, our_chatbot.tokenizer, input_ids
|
377 |
)
|
378 |
+
|
|
|
|
|
379 |
streamer = TextIteratorStreamer(
|
380 |
our_chatbot.tokenizer, skip_prompt=True, skip_special_tokens=True
|
381 |
)
|
|
|
383 |
print(input_ids.device)
|
384 |
print(image_tensor.device)
|
385 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
386 |
generate_kwargs = dict(
|
387 |
inputs=input_ids,
|
388 |
streamer=streamer,
|
|
|
401 |
outputs = []
|
402 |
for stream_token in streamer:
|
403 |
outputs.append(stream_token)
|
404 |
+
|
|
|
405 |
history[-1] = [text, "".join(outputs)]
|
406 |
yield history
|
407 |
our_chatbot.conversation.messages[-1][-1] = "".join(outputs)
|
408 |
+
# print("### turn end history", history)
|
409 |
+
# print("### turn end conv",our_chatbot.conversation)
|
410 |
|
411 |
with open(get_conv_log_filename(), "a") as fout:
|
412 |
data = {
|
|
|
670 |
gr.Markdown(learn_more_markdown)
|
671 |
gr.Markdown(bibtext)
|
672 |
|
673 |
+
chat_input.submit(
|
674 |
+
add_message, [chatbot, chat_input], [chatbot, chat_input]
|
675 |
+
).then(bot, [chatbot, temperature, top_p, max_output_tokens], chatbot, api_name="bot_response").then(lambda: gr.MultimodalTextbox(interactive=True), None, [chat_input])
|
676 |
+
|
|
|
677 |
|
678 |
# chatbot.like(print_like_dislike, None, None)
|
679 |
clear_btn.click(
|
680 |
fn=clear_history, inputs=[chatbot], outputs=[chatbot], api_name="clear_all"
|
681 |
)
|
682 |
|
683 |
+
upvote_btn.click(
|
684 |
+
fn=upvote_last_response, inputs=chatbot, outputs=chatbot, api_name="upvote_last_response"
|
685 |
+
)
|
686 |
+
|
687 |
+
|
688 |
+
downvote_btn.click(
|
689 |
+
fn=downvote_last_response, inputs=chatbot, outputs=chatbot, api_name="upvote_last_response"
|
690 |
+
)
|
691 |
+
|
692 |
|
693 |
demo.queue()
|
694 |
|
|
|
719 |
model_name = get_model_name_from_path(args.model_path)
|
720 |
tokenizer, model, image_processor, context_len = load_pretrained_model(args.model_path, args.model_base, model_name, args.load_8bit, args.load_4bit)
|
721 |
model=model.to(torch.device('cuda'))
|
722 |
+
chat_image_num = 0
|
723 |
demo.launch()
|