Upload 70 files
Browse filesThis view is limited to 50 files because it contains too many changes.
See raw diff
- .gitattributes +7 -0
- README.md +8 -10
- app.py +535 -0
- examples/.DS_Store +0 -0
- examples/172197131626056_P7966202.png +0 -0
- examples/A-17-processors-1024x576.jpg +0 -0
- examples/Iphone-15-Usb-c-charger-1024x576.jpg +0 -0
- examples/Iphone-15-specs-1024x576.jpg +0 -0
- examples/africa.jpg +0 -0
- examples/ballon.jpg +0 -0
- examples/bar.jpg +0 -0
- examples/bigcompany.png +3 -0
- examples/bijiasuo2.jpeg +0 -0
- examples/book.jpg +0 -0
- examples/camera.jpg +0 -0
- examples/changed_bench.jpeg +0 -0
- examples/code.mp4 +0 -0
- examples/code1.jpeg +0 -0
- examples/code2.jpeg +0 -0
- examples/dog.jpg +0 -0
- examples/dog1.jpg +0 -0
- examples/dog6.jpeg +0 -0
- examples/dog9.jpeg +0 -0
- examples/dog_to_monkey1.png +3 -0
- examples/dog_to_monkey2.png +3 -0
- examples/dynamic-island-1024x576.jpg +0 -0
- examples/eagles.jpg +0 -0
- examples/ecg_example1.jpg +0 -0
- examples/ecg_example2.png +0 -0
- examples/examples_image12.jpg +0 -0
- examples/examples_image13.jpg +0 -0
- examples/examples_image14.jpg +0 -0
- examples/fangao1.jpeg +0 -0
- examples/fangao2.jpeg +0 -0
- examples/fangao3.jpeg +0 -0
- examples/food.jpg +0 -0
- examples/girl.jpg +0 -0
- examples/hanzi.jpg +0 -0
- examples/hot_ballon.jpg +0 -0
- examples/ice_cream.jpg +0 -0
- examples/image-00007.jpeg +0 -0
- examples/image-00053.jpeg +0 -0
- examples/iphone-15-colors-1024x576.jpg +0 -0
- examples/iphone-15-price-1024x576.jpg +0 -0
- examples/iphone-15-pricing-1024x576.jpg +0 -0
- examples/line chart.jpg +0 -0
- examples/norway.jpg +0 -0
- examples/oprah-winfrey-resume.png +0 -0
- examples/orange.png +0 -0
- examples/original_bench.jpeg +0 -0
.gitattributes
CHANGED
@@ -33,3 +33,10 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
33 |
*.zip filter=lfs diff=lfs merge=lfs -text
|
34 |
*.zst filter=lfs diff=lfs merge=lfs -text
|
35 |
*tfevents* filter=lfs diff=lfs merge=lfs -text
|
36 |
+
examples/bigcompany.png filter=lfs diff=lfs merge=lfs -text
|
37 |
+
examples/dog_to_monkey1.png filter=lfs diff=lfs merge=lfs -text
|
38 |
+
examples/dog_to_monkey2.png filter=lfs diff=lfs merge=lfs -text
|
39 |
+
examples/twitter2.jpeg filter=lfs diff=lfs merge=lfs -text
|
40 |
+
examples/twitter3.jpeg filter=lfs diff=lfs merge=lfs -text
|
41 |
+
examples/twitter4.jpeg filter=lfs diff=lfs merge=lfs -text
|
42 |
+
examples/user_example_07.jpg filter=lfs diff=lfs merge=lfs -text
|
README.md
CHANGED
@@ -1,14 +1,12 @@
|
|
1 |
---
|
2 |
-
title:
|
3 |
-
emoji:
|
4 |
-
colorFrom:
|
5 |
-
colorTo:
|
6 |
sdk: gradio
|
7 |
-
sdk_version:
|
8 |
app_file: app.py
|
9 |
-
pinned:
|
10 |
-
|
11 |
-
short_description: ECG
|
12 |
---
|
13 |
-
|
14 |
-
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
|
|
1 |
---
|
2 |
+
title: Pangea
|
3 |
+
emoji: 🚀
|
4 |
+
colorFrom: green
|
5 |
+
colorTo: red
|
6 |
sdk: gradio
|
7 |
+
sdk_version: 4.37.2
|
8 |
app_file: app.py
|
9 |
+
pinned: true
|
10 |
+
short_description: A Fully Open Multilingual Multimodal LLM for 39 Languages
|
|
|
11 |
---
|
12 |
+
Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference
|
|
app.py
ADDED
@@ -0,0 +1,535 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# from .demo_modelpart import InferenceDemo
|
2 |
+
import gradio as gr
|
3 |
+
import os
|
4 |
+
from threading import Thread
|
5 |
+
|
6 |
+
# import time
|
7 |
+
import cv2
|
8 |
+
|
9 |
+
import datetime
|
10 |
+
# import copy
|
11 |
+
import torch
|
12 |
+
|
13 |
+
import spaces
|
14 |
+
import numpy as np
|
15 |
+
|
16 |
+
from llava import conversation as conversation_lib
|
17 |
+
from llava.constants import DEFAULT_IMAGE_TOKEN
|
18 |
+
|
19 |
+
|
20 |
+
from llava.constants import (
|
21 |
+
IMAGE_TOKEN_INDEX,
|
22 |
+
DEFAULT_IMAGE_TOKEN,
|
23 |
+
DEFAULT_IM_START_TOKEN,
|
24 |
+
DEFAULT_IM_END_TOKEN,
|
25 |
+
)
|
26 |
+
from llava.conversation import conv_templates, SeparatorStyle
|
27 |
+
from llava.model.builder import load_pretrained_model
|
28 |
+
from llava.utils import disable_torch_init
|
29 |
+
from llava.mm_utils import (
|
30 |
+
tokenizer_image_token,
|
31 |
+
process_images,
|
32 |
+
get_model_name_from_path,
|
33 |
+
KeywordsStoppingCriteria,
|
34 |
+
)
|
35 |
+
|
36 |
+
from serve_constants import html_header, bibtext, learn_more_markdown, tos_markdown
|
37 |
+
|
38 |
+
import requests
|
39 |
+
from PIL import Image
|
40 |
+
from io import BytesIO
|
41 |
+
from transformers import TextStreamer, TextIteratorStreamer
|
42 |
+
|
43 |
+
import hashlib
|
44 |
+
import PIL
|
45 |
+
import base64
|
46 |
+
import json
|
47 |
+
|
48 |
+
import datetime
|
49 |
+
import gradio as gr
|
50 |
+
import gradio_client
|
51 |
+
import subprocess
|
52 |
+
import sys
|
53 |
+
|
54 |
+
from huggingface_hub import HfApi
|
55 |
+
from huggingface_hub import login
|
56 |
+
from huggingface_hub import revision_exists
|
57 |
+
|
58 |
+
login(token=os.environ["HF_TOKEN"],
|
59 |
+
write_permission=True)
|
60 |
+
|
61 |
+
api = HfApi()
|
62 |
+
repo_name = os.environ["LOG_REPO"]
|
63 |
+
|
64 |
+
external_log_dir = "./logs"
|
65 |
+
LOGDIR = external_log_dir
|
66 |
+
|
67 |
+
|
68 |
+
def install_gradio_4_35_0():
|
69 |
+
current_version = gr.__version__
|
70 |
+
if current_version != "4.35.0":
|
71 |
+
print(f"Current Gradio version: {current_version}")
|
72 |
+
print("Installing Gradio 4.35.0...")
|
73 |
+
subprocess.check_call([sys.executable, "-m", "pip", "install", "gradio==4.35.0", "--force-reinstall"])
|
74 |
+
print("Gradio 4.35.0 installed successfully.")
|
75 |
+
else:
|
76 |
+
print("Gradio 4.35.0 is already installed.")
|
77 |
+
|
78 |
+
# Call the function to install Gradio 4.35.0 if needed
|
79 |
+
install_gradio_4_35_0()
|
80 |
+
|
81 |
+
import gradio as gr
|
82 |
+
import gradio_client
|
83 |
+
print(f"Gradio version: {gr.__version__}")
|
84 |
+
print(f"Gradio-client version: {gradio_client.__version__}")
|
85 |
+
|
86 |
+
def get_conv_log_filename():
|
87 |
+
t = datetime.datetime.now()
|
88 |
+
name = os.path.join(LOGDIR, f"{t.year}-{t.month:02d}-{t.day:02d}-user_conv.json")
|
89 |
+
return name
|
90 |
+
|
91 |
+
class InferenceDemo(object):
|
92 |
+
def __init__(
|
93 |
+
self, args, model_path, tokenizer, model, image_processor, context_len
|
94 |
+
) -> None:
|
95 |
+
disable_torch_init()
|
96 |
+
|
97 |
+
self.tokenizer, self.model, self.image_processor, self.context_len = (
|
98 |
+
tokenizer,
|
99 |
+
model,
|
100 |
+
image_processor,
|
101 |
+
context_len,
|
102 |
+
)
|
103 |
+
|
104 |
+
if "llama-2" in model_name.lower():
|
105 |
+
conv_mode = "llava_llama_2"
|
106 |
+
elif "v1" in model_name.lower() or "pulse" in model_name.lower():
|
107 |
+
conv_mode = "llava_v1"
|
108 |
+
elif "mpt" in model_name.lower():
|
109 |
+
conv_mode = "mpt"
|
110 |
+
elif "qwen" in model_name.lower():
|
111 |
+
conv_mode = "qwen_1_5"
|
112 |
+
else:
|
113 |
+
conv_mode = "llava_v0"
|
114 |
+
|
115 |
+
if args.conv_mode is not None and conv_mode != args.conv_mode:
|
116 |
+
print(
|
117 |
+
"[WARNING] the auto inferred conversation mode is {}, while `--conv-mode` is {}, using {}".format(
|
118 |
+
conv_mode, args.conv_mode, args.conv_mode
|
119 |
+
)
|
120 |
+
)
|
121 |
+
else:
|
122 |
+
args.conv_mode = conv_mode
|
123 |
+
self.conv_mode = conv_mode
|
124 |
+
self.conversation = conv_templates[args.conv_mode].copy()
|
125 |
+
self.num_frames = args.num_frames
|
126 |
+
|
127 |
+
|
128 |
+
def is_valid_video_filename(name):
|
129 |
+
video_extensions = ["avi", "mp4", "mov", "mkv", "flv", "wmv", "mjpeg"]
|
130 |
+
|
131 |
+
ext = name.split(".")[-1].lower()
|
132 |
+
|
133 |
+
if ext in video_extensions:
|
134 |
+
return True
|
135 |
+
else:
|
136 |
+
return False
|
137 |
+
|
138 |
+
def is_valid_image_filename(name):
|
139 |
+
image_extensions = ["jpg", "jpeg", "png", "bmp", "gif", "tiff", "webp", "heic", "heif", "jfif", "svg", "eps", "raw"]
|
140 |
+
|
141 |
+
ext = name.split(".")[-1].lower()
|
142 |
+
|
143 |
+
if ext in image_extensions:
|
144 |
+
return True
|
145 |
+
else:
|
146 |
+
return False
|
147 |
+
|
148 |
+
|
149 |
+
def sample_frames(video_file, num_frames):
|
150 |
+
video = cv2.VideoCapture(video_file)
|
151 |
+
total_frames = int(video.get(cv2.CAP_PROP_FRAME_COUNT))
|
152 |
+
interval = total_frames // num_frames
|
153 |
+
frames = []
|
154 |
+
for i in range(total_frames):
|
155 |
+
ret, frame = video.read()
|
156 |
+
pil_img = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
|
157 |
+
if not ret:
|
158 |
+
continue
|
159 |
+
if i % interval == 0:
|
160 |
+
frames.append(pil_img)
|
161 |
+
video.release()
|
162 |
+
return frames
|
163 |
+
|
164 |
+
|
165 |
+
def load_image(image_file):
|
166 |
+
if image_file.startswith("http") or image_file.startswith("https"):
|
167 |
+
response = requests.get(image_file)
|
168 |
+
if response.status_code == 200:
|
169 |
+
image = Image.open(BytesIO(response.content)).convert("RGB")
|
170 |
+
else:
|
171 |
+
print("failed to load the image")
|
172 |
+
else:
|
173 |
+
print("Load image from local file")
|
174 |
+
print(image_file)
|
175 |
+
image = Image.open(image_file).convert("RGB")
|
176 |
+
|
177 |
+
return image
|
178 |
+
|
179 |
+
|
180 |
+
def clear_history(history):
|
181 |
+
|
182 |
+
our_chatbot.conversation = conv_templates[our_chatbot.conv_mode].copy()
|
183 |
+
|
184 |
+
return None
|
185 |
+
|
186 |
+
|
187 |
+
def clear_response(history):
|
188 |
+
for index_conv in range(1, len(history)):
|
189 |
+
# loop until get a text response from our model.
|
190 |
+
conv = history[-index_conv]
|
191 |
+
if not (conv[0] is None):
|
192 |
+
break
|
193 |
+
question = history[-index_conv][0]
|
194 |
+
history = history[:-index_conv]
|
195 |
+
return history, question
|
196 |
+
|
197 |
+
|
198 |
+
# def print_like_dislike(x: gr.LikeData):
|
199 |
+
# print(x.index, x.value, x.liked)
|
200 |
+
|
201 |
+
|
202 |
+
def add_message(history, message):
|
203 |
+
# history=[]
|
204 |
+
global our_chatbot
|
205 |
+
if len(history) == 0:
|
206 |
+
our_chatbot = InferenceDemo(
|
207 |
+
args, model_path, tokenizer, model, image_processor, context_len
|
208 |
+
)
|
209 |
+
|
210 |
+
for x in message["files"]:
|
211 |
+
history.append(((x,), None))
|
212 |
+
if message["text"] is not None:
|
213 |
+
history.append((message["text"], None))
|
214 |
+
return history, gr.MultimodalTextbox(value=None, interactive=False)
|
215 |
+
|
216 |
+
|
217 |
+
@spaces.GPU
|
218 |
+
def bot(history, temperature, top_p, max_output_tokens):
|
219 |
+
print("### turn start history",history)
|
220 |
+
print("### turn start conv",our_chatbot.conversation)
|
221 |
+
text = history[-1][0]
|
222 |
+
images_this_term = []
|
223 |
+
text_this_term = ""
|
224 |
+
# import pdb;pdb.set_trace()
|
225 |
+
num_new_images = 0
|
226 |
+
for i, message in enumerate(history[:-1]):
|
227 |
+
if type(message[0]) is tuple:
|
228 |
+
images_this_term.append(message[0][0])
|
229 |
+
if is_valid_video_filename(message[0][0]):
|
230 |
+
# 不接受视频
|
231 |
+
raise ValueError("Video is not supported")
|
232 |
+
num_new_images += our_chatbot.num_frames
|
233 |
+
elif is_valid_image_filename(message[0][0]):
|
234 |
+
print("#### Load image from local file",message[0][0])
|
235 |
+
num_new_images += 1
|
236 |
+
else:
|
237 |
+
raise ValueError("Invalid image file")
|
238 |
+
else:
|
239 |
+
num_new_images = 0
|
240 |
+
|
241 |
+
# for message in history[-i-1:]:
|
242 |
+
# images_this_term.append(message[0][0])
|
243 |
+
|
244 |
+
assert len(images_this_term) > 0, "must have an image"
|
245 |
+
# image_files = (args.image_file).split(',')
|
246 |
+
# image = [load_image(f) for f in images_this_term if f]
|
247 |
+
|
248 |
+
all_image_hash = []
|
249 |
+
all_image_path = []
|
250 |
+
for image_path in images_this_term:
|
251 |
+
with open(image_path, "rb") as image_file:
|
252 |
+
image_data = image_file.read()
|
253 |
+
image_hash = hashlib.md5(image_data).hexdigest()
|
254 |
+
all_image_hash.append(image_hash)
|
255 |
+
image = PIL.Image.open(image_path).convert("RGB")
|
256 |
+
t = datetime.datetime.now()
|
257 |
+
filename = os.path.join(
|
258 |
+
LOGDIR,
|
259 |
+
"serve_images",
|
260 |
+
f"{t.year}-{t.month:02d}-{t.day:02d}",
|
261 |
+
f"{image_hash}.jpg",
|
262 |
+
)
|
263 |
+
all_image_path.append(filename)
|
264 |
+
if not os.path.isfile(filename):
|
265 |
+
os.makedirs(os.path.dirname(filename), exist_ok=True)
|
266 |
+
print("image save to",filename)
|
267 |
+
image.save(filename)
|
268 |
+
|
269 |
+
image_list = []
|
270 |
+
for f in images_this_term:
|
271 |
+
if is_valid_video_filename(f):
|
272 |
+
image_list += sample_frames(f, our_chatbot.num_frames)
|
273 |
+
elif is_valid_image_filename(f):
|
274 |
+
image_list.append(load_image(f))
|
275 |
+
else:
|
276 |
+
raise ValueError("Invalid image file")
|
277 |
+
|
278 |
+
image_tensor = [
|
279 |
+
process_images([f], our_chatbot.image_processor, our_chatbot.model.config)[0]
|
280 |
+
.to(our_chatbot.model.device)
|
281 |
+
for f in image_list
|
282 |
+
]
|
283 |
+
|
284 |
+
|
285 |
+
image_tensor = torch.stack(image_tensor)
|
286 |
+
image_token = DEFAULT_IMAGE_TOKEN * num_new_images
|
287 |
+
# if our_chatbot.model.config.mm_use_im_start_end:
|
288 |
+
# inp = DEFAULT_IM_START_TOKEN + image_token + DEFAULT_IM_END_TOKEN + "\n" + inp
|
289 |
+
# else:
|
290 |
+
inp = text
|
291 |
+
inp = image_token + "\n" + inp
|
292 |
+
our_chatbot.conversation.append_message(our_chatbot.conversation.roles[0], inp)
|
293 |
+
# image = None
|
294 |
+
our_chatbot.conversation.append_message(our_chatbot.conversation.roles[1], None)
|
295 |
+
prompt = our_chatbot.conversation.get_prompt()
|
296 |
+
|
297 |
+
# input_ids = (
|
298 |
+
# tokenizer_image_token(
|
299 |
+
# prompt, our_chatbot.tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt"
|
300 |
+
# )
|
301 |
+
# .unsqueeze(0)
|
302 |
+
# .to(our_chatbot.model.device)
|
303 |
+
# )
|
304 |
+
input_ids = tokenizer_image_token(
|
305 |
+
prompt, our_chatbot.tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt"
|
306 |
+
).unsqueeze(0).to(our_chatbot.model.device)
|
307 |
+
# print("### input_id",input_ids)
|
308 |
+
stop_str = (
|
309 |
+
our_chatbot.conversation.sep
|
310 |
+
if our_chatbot.conversation.sep_style != SeparatorStyle.TWO
|
311 |
+
else our_chatbot.conversation.sep2
|
312 |
+
)
|
313 |
+
keywords = [stop_str]
|
314 |
+
stopping_criteria = KeywordsStoppingCriteria(
|
315 |
+
keywords, our_chatbot.tokenizer, input_ids
|
316 |
+
)
|
317 |
+
# streamer = TextStreamer(
|
318 |
+
# our_chatbot.tokenizer, skip_prompt=True, skip_special_tokens=True
|
319 |
+
# )
|
320 |
+
streamer = TextIteratorStreamer(
|
321 |
+
our_chatbot.tokenizer, skip_prompt=True, skip_special_tokens=True
|
322 |
+
)
|
323 |
+
print(our_chatbot.model.device)
|
324 |
+
print(input_ids.device)
|
325 |
+
print(image_tensor.device)
|
326 |
+
|
327 |
+
# with torch.inference_mode():
|
328 |
+
# output_ids = our_chatbot.model.generate(
|
329 |
+
# input_ids,
|
330 |
+
# images=image_tensor,
|
331 |
+
# do_sample=True,
|
332 |
+
# temperature=0.7,
|
333 |
+
# top_p=1.0,
|
334 |
+
# max_new_tokens=4096,
|
335 |
+
# streamer=streamer,
|
336 |
+
# use_cache=False,
|
337 |
+
# stopping_criteria=[stopping_criteria],
|
338 |
+
# )
|
339 |
+
|
340 |
+
# outputs = our_chatbot.tokenizer.decode(output_ids[0]).strip()
|
341 |
+
# if outputs.endswith(stop_str):
|
342 |
+
# outputs = outputs[: -len(stop_str)]
|
343 |
+
# our_chatbot.conversation.messages[-1][-1] = outputs
|
344 |
+
|
345 |
+
# history[-1] = [text, outputs]
|
346 |
+
|
347 |
+
# return history
|
348 |
+
generate_kwargs = dict(
|
349 |
+
inputs=input_ids,
|
350 |
+
streamer=streamer,
|
351 |
+
images=image_tensor,
|
352 |
+
do_sample=True,
|
353 |
+
temperature=temperature,
|
354 |
+
top_p=top_p,
|
355 |
+
max_new_tokens=max_output_tokens,
|
356 |
+
use_cache=False,
|
357 |
+
stopping_criteria=[stopping_criteria],
|
358 |
+
)
|
359 |
+
|
360 |
+
t = Thread(target=our_chatbot.model.generate, kwargs=generate_kwargs)
|
361 |
+
t.start()
|
362 |
+
|
363 |
+
outputs = []
|
364 |
+
for stream_token in streamer:
|
365 |
+
outputs.append(stream_token)
|
366 |
+
# print("### stream_token",stream_token)
|
367 |
+
# our_chatbot.conversation.messages[-1][-1] = "".join(outputs)
|
368 |
+
history[-1] = [text, "".join(outputs)]
|
369 |
+
yield history
|
370 |
+
our_chatbot.conversation.messages[-1][-1] = "".join(outputs)
|
371 |
+
print("### turn end history", history)
|
372 |
+
print("### turn end conv",our_chatbot.conversation)
|
373 |
+
|
374 |
+
with open(get_conv_log_filename(), "a") as fout:
|
375 |
+
data = {
|
376 |
+
"type": "chat",
|
377 |
+
"model": "PULSE-7b",
|
378 |
+
"state": history,
|
379 |
+
"images": all_image_hash,
|
380 |
+
"images_path": all_image_path
|
381 |
+
}
|
382 |
+
print("#### conv log",data)
|
383 |
+
fout.write(json.dumps(data) + "\n")
|
384 |
+
for upload_img in all_image_path:
|
385 |
+
api.upload_file(
|
386 |
+
path_or_fileobj=upload_img,
|
387 |
+
path_in_repo=upload_img.replace("./logs/", ""),
|
388 |
+
repo_id=repo_name,
|
389 |
+
repo_type="dataset",
|
390 |
+
# revision=revision,
|
391 |
+
# ignore_patterns=["data*"]
|
392 |
+
)
|
393 |
+
# upload json
|
394 |
+
api.upload_file(
|
395 |
+
path_or_fileobj=get_conv_log_filename(),
|
396 |
+
path_in_repo=get_conv_log_filename().replace("./logs/", ""),
|
397 |
+
repo_id=repo_name,
|
398 |
+
repo_type="dataset")
|
399 |
+
|
400 |
+
|
401 |
+
|
402 |
+
txt = gr.Textbox(
|
403 |
+
scale=4,
|
404 |
+
show_label=False,
|
405 |
+
placeholder="Enter text and press enter.",
|
406 |
+
container=False,
|
407 |
+
)
|
408 |
+
|
409 |
+
with gr.Blocks(
|
410 |
+
css=".message-wrap.svelte-1lcyrx4>div.svelte-1lcyrx4 img {min-width: 40px}",
|
411 |
+
) as demo:
|
412 |
+
|
413 |
+
cur_dir = os.path.dirname(os.path.abspath(__file__))
|
414 |
+
# gr.Markdown(title_markdown)
|
415 |
+
gr.HTML(html_header)
|
416 |
+
|
417 |
+
with gr.Column():
|
418 |
+
with gr.Accordion("Parameters", open=False) as parameter_row:
|
419 |
+
temperature = gr.Slider(
|
420 |
+
minimum=0.0,
|
421 |
+
maximum=1.0,
|
422 |
+
value=0.0,
|
423 |
+
step=0.1,
|
424 |
+
interactive=True,
|
425 |
+
label="Temperature",
|
426 |
+
)
|
427 |
+
top_p = gr.Slider(
|
428 |
+
minimum=0.0,
|
429 |
+
maximum=1.0,
|
430 |
+
value=1,
|
431 |
+
step=0.1,
|
432 |
+
interactive=True,
|
433 |
+
label="Top P",
|
434 |
+
)
|
435 |
+
max_output_tokens = gr.Slider(
|
436 |
+
minimum=0,
|
437 |
+
maximum=8192,
|
438 |
+
value=4096,
|
439 |
+
step=256,
|
440 |
+
interactive=True,
|
441 |
+
label="Max output tokens",
|
442 |
+
)
|
443 |
+
with gr.Row():
|
444 |
+
chatbot = gr.Chatbot([], elem_id="PULSE", bubble_full_width=False, height=750)
|
445 |
+
|
446 |
+
with gr.Row():
|
447 |
+
upvote_btn = gr.Button(value="👍 Upvote", interactive=True)
|
448 |
+
downvote_btn = gr.Button(value="👎 Downvote", interactive=True)
|
449 |
+
flag_btn = gr.Button(value="⚠️ Flag", interactive=True)
|
450 |
+
# stop_btn = gr.Button(value="⏹️ Stop Generation", interactive=True)
|
451 |
+
regenerate_btn = gr.Button(value="🔄 Regenerate", interactive=True)
|
452 |
+
clear_btn = gr.Button(value="🗑️ Clear history", interactive=True)
|
453 |
+
|
454 |
+
|
455 |
+
chat_input = gr.MultimodalTextbox(
|
456 |
+
interactive=True,
|
457 |
+
file_types=["image"],
|
458 |
+
placeholder="Enter message or upload file...",
|
459 |
+
show_label=False,
|
460 |
+
submit_btn="🚀"
|
461 |
+
)
|
462 |
+
|
463 |
+
print(cur_dir)
|
464 |
+
gr.Examples(
|
465 |
+
examples_per_page=5,
|
466 |
+
examples=[
|
467 |
+
[
|
468 |
+
{
|
469 |
+
"files": [
|
470 |
+
f"{cur_dir}/examples/ecg_example2.png",
|
471 |
+
],
|
472 |
+
"text": "What are the main features in this ECG image?",
|
473 |
+
},
|
474 |
+
],
|
475 |
+
[
|
476 |
+
{
|
477 |
+
"files": [
|
478 |
+
f"{cur_dir}/examples/ecg_example1.jpg",
|
479 |
+
],
|
480 |
+
"text": "What can be inferred from the pattern of the qR complexes and rS complexes in the leads of this ECG image?",
|
481 |
+
},
|
482 |
+
]
|
483 |
+
],
|
484 |
+
inputs=[chat_input],
|
485 |
+
label="Image",
|
486 |
+
)
|
487 |
+
|
488 |
+
gr.Markdown(tos_markdown)
|
489 |
+
gr.Markdown(learn_more_markdown)
|
490 |
+
gr.Markdown(bibtext)
|
491 |
+
|
492 |
+
chat_msg = chat_input.submit(
|
493 |
+
add_message, [chatbot, chat_input], [chatbot, chat_input]
|
494 |
+
)
|
495 |
+
bot_msg = chat_msg.then(bot, [chatbot,temperature, top_p, max_output_tokens], chatbot, api_name="bot_response")
|
496 |
+
bot_msg.then(lambda: gr.MultimodalTextbox(interactive=True), None, [chat_input])
|
497 |
+
|
498 |
+
# chatbot.like(print_like_dislike, None, None)
|
499 |
+
clear_btn.click(
|
500 |
+
fn=clear_history, inputs=[chatbot], outputs=[chatbot], api_name="clear_all"
|
501 |
+
)
|
502 |
+
|
503 |
+
|
504 |
+
demo.queue()
|
505 |
+
|
506 |
+
if __name__ == "__main__":
|
507 |
+
import argparse
|
508 |
+
|
509 |
+
argparser = argparse.ArgumentParser()
|
510 |
+
argparser.add_argument("--server_name", default="0.0.0.0", type=str)
|
511 |
+
argparser.add_argument("--port", default="6123", type=str)
|
512 |
+
argparser.add_argument(
|
513 |
+
"--model_path", default="PULSE-ECG/PULSE-7B", type=str
|
514 |
+
)
|
515 |
+
# argparser.add_argument("--model-path", type=str, default="facebook/opt-350m")
|
516 |
+
argparser.add_argument("--model-base", type=str, default=None)
|
517 |
+
argparser.add_argument("--num-gpus", type=int, default=1)
|
518 |
+
argparser.add_argument("--conv-mode", type=str, default=None)
|
519 |
+
argparser.add_argument("--temperature", type=float, default=0.0)
|
520 |
+
argparser.add_argument("--max-new-tokens", type=int, default=1024)
|
521 |
+
argparser.add_argument("--num_frames", type=int, default=16)
|
522 |
+
argparser.add_argument("--load-8bit", action="store_true")
|
523 |
+
argparser.add_argument("--load-4bit", action="store_true")
|
524 |
+
argparser.add_argument("--debug", action="store_true")
|
525 |
+
|
526 |
+
args = argparser.parse_args()
|
527 |
+
|
528 |
+
model_path = args.model_path
|
529 |
+
filt_invalid = "cut"
|
530 |
+
model_name = get_model_name_from_path(args.model_path)
|
531 |
+
tokenizer, model, image_processor, context_len = load_pretrained_model(args.model_path, args.model_base, model_name, args.load_8bit, args.load_4bit)
|
532 |
+
print("### image_processor",image_processor)
|
533 |
+
model=model.to(torch.device('cuda'))
|
534 |
+
our_chatbot = None
|
535 |
+
demo.launch()
|
examples/.DS_Store
ADDED
Binary file (6.15 kB). View file
|
|
examples/172197131626056_P7966202.png
ADDED
examples/A-17-processors-1024x576.jpg
ADDED
examples/Iphone-15-Usb-c-charger-1024x576.jpg
ADDED
examples/Iphone-15-specs-1024x576.jpg
ADDED
examples/africa.jpg
ADDED
examples/ballon.jpg
ADDED
examples/bar.jpg
ADDED
examples/bigcompany.png
ADDED
Git LFS Details
|
examples/bijiasuo2.jpeg
ADDED
examples/book.jpg
ADDED
examples/camera.jpg
ADDED
examples/changed_bench.jpeg
ADDED
examples/code.mp4
ADDED
Binary file (91.1 kB). View file
|
|
examples/code1.jpeg
ADDED
examples/code2.jpeg
ADDED
examples/dog.jpg
ADDED
examples/dog1.jpg
ADDED
examples/dog6.jpeg
ADDED
examples/dog9.jpeg
ADDED
examples/dog_to_monkey1.png
ADDED
Git LFS Details
|
examples/dog_to_monkey2.png
ADDED
Git LFS Details
|
examples/dynamic-island-1024x576.jpg
ADDED
examples/eagles.jpg
ADDED
examples/ecg_example1.jpg
ADDED
examples/ecg_example2.png
ADDED
examples/examples_image12.jpg
ADDED
examples/examples_image13.jpg
ADDED
examples/examples_image14.jpg
ADDED
examples/fangao1.jpeg
ADDED
examples/fangao2.jpeg
ADDED
examples/fangao3.jpeg
ADDED
examples/food.jpg
ADDED
examples/girl.jpg
ADDED
examples/hanzi.jpg
ADDED
examples/hot_ballon.jpg
ADDED
examples/ice_cream.jpg
ADDED
examples/image-00007.jpeg
ADDED
examples/image-00053.jpeg
ADDED
examples/iphone-15-colors-1024x576.jpg
ADDED
examples/iphone-15-price-1024x576.jpg
ADDED
examples/iphone-15-pricing-1024x576.jpg
ADDED
examples/line chart.jpg
ADDED
examples/norway.jpg
ADDED
examples/oprah-winfrey-resume.png
ADDED
examples/orange.png
ADDED
examples/original_bench.jpeg
ADDED