Update app.py
Browse files
app.py
CHANGED
|
@@ -11,10 +11,18 @@ from threading import Thread
|
|
| 11 |
import torch
|
| 12 |
import gradio as gr
|
| 13 |
from transformers import AutoModelForCausalLM, TextIteratorStreamer
|
|
|
|
|
|
|
| 14 |
|
| 15 |
model_name = 'AIDC-AI/Ovis2-16B'
|
|
|
|
| 16 |
use_thread = False
|
| 17 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
# load model
|
| 19 |
model = AutoModelForCausalLM.from_pretrained(model_name,
|
| 20 |
torch_dtype=torch.bfloat16,
|
|
@@ -49,8 +57,8 @@ def submit_chat(chatbot, text_input):
|
|
| 49 |
return chatbot ,''
|
| 50 |
|
| 51 |
@spaces.GPU
|
| 52 |
-
def ovis_chat(chatbot: List[List[str]], image_input: Any):
|
| 53 |
-
conversations, model_inputs = prepare_inputs(chatbot, image_input)
|
| 54 |
gen_kwargs = initialize_gen_kwargs()
|
| 55 |
|
| 56 |
with torch.inference_mode():
|
|
@@ -74,7 +82,7 @@ def ovis_chat(chatbot: List[List[str]], image_input: Any):
|
|
| 74 |
log_conversation(chatbot)
|
| 75 |
|
| 76 |
|
| 77 |
-
def prepare_inputs(chatbot: List[List[str]], image_input: Any):
|
| 78 |
# conversations = [{
|
| 79 |
# "from": "system",
|
| 80 |
# "value": "You are a helpful assistant, and your task is to provide reliable and structured responses to users."
|
|
@@ -95,10 +103,30 @@ def prepare_inputs(chatbot: List[List[str]], image_input: Any):
|
|
| 95 |
if conv["from"] == "human":
|
| 96 |
conv["value"] = f'{image_placeholder}\n{conv["value"]}'
|
| 97 |
break
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 98 |
|
| 99 |
logger.info(conversations)
|
| 100 |
|
| 101 |
-
prompt, input_ids, pixel_values = model.preprocess_inputs(conversations,
|
| 102 |
attention_mask = torch.ne(input_ids, text_tokenizer.pad_token_id)
|
| 103 |
|
| 104 |
model_inputs = {
|
|
@@ -115,7 +143,7 @@ def log_conversation(chatbot):
|
|
| 115 |
logger.info("[OVIS_CONV_END]")
|
| 116 |
|
| 117 |
def clear_chat():
|
| 118 |
-
return [], None, ""
|
| 119 |
|
| 120 |
with open(f"{cur_dir}/resource/logo.svg", "r", encoding="utf-8") as svg_file:
|
| 121 |
svg_content = svg_file.read()
|
|
@@ -164,26 +192,58 @@ with gr.Blocks(title=model_name.split('/')[-1], theme=gr.themes.Ocean()) as demo
|
|
| 164 |
gr.HTML(html)
|
| 165 |
with gr.Row():
|
| 166 |
with gr.Column(scale=3):
|
| 167 |
-
|
| 168 |
-
|
| 169 |
-
|
| 170 |
-
|
| 171 |
-
|
| 172 |
-
|
| 173 |
-
[
|
| 174 |
-
|
| 175 |
-
|
| 176 |
-
|
| 177 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 178 |
with gr.Column(scale=7):
|
| 179 |
chatbot = gr.Chatbot(label="Ovis", layout="panel", height=600, show_copy_button=True, latex_delimiters=latex_delimiters_set)
|
| 180 |
text_input.render()
|
| 181 |
with gr.Row():
|
| 182 |
send_btn = gr.Button("Send", variant="primary")
|
| 183 |
clear_btn = gr.Button("Clear", variant="secondary")
|
| 184 |
-
|
| 185 |
-
|
| 186 |
-
|
| 187 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 188 |
|
| 189 |
demo.launch()
|
|
|
|
| 11 |
import torch
|
| 12 |
import gradio as gr
|
| 13 |
from transformers import AutoModelForCausalLM, TextIteratorStreamer
|
| 14 |
+
from moviepy.editor import VideoFileClip
|
| 15 |
+
from PIL import Image
|
| 16 |
|
| 17 |
model_name = 'AIDC-AI/Ovis2-16B'
|
| 18 |
+
|
| 19 |
use_thread = False
|
| 20 |
|
| 21 |
+
IMAGE_MAX_PARTITION = 16
|
| 22 |
+
|
| 23 |
+
VIDEO_FRAME_NUMS = 32
|
| 24 |
+
VIDEO_MAX_PARTITION = 1
|
| 25 |
+
|
| 26 |
# load model
|
| 27 |
model = AutoModelForCausalLM.from_pretrained(model_name,
|
| 28 |
torch_dtype=torch.bfloat16,
|
|
|
|
| 57 |
return chatbot ,''
|
| 58 |
|
| 59 |
@spaces.GPU
|
| 60 |
+
def ovis_chat(chatbot: List[List[str]], image_input: Any, video_input: Any):
|
| 61 |
+
conversations, model_inputs = prepare_inputs(chatbot, image_input, video_input)
|
| 62 |
gen_kwargs = initialize_gen_kwargs()
|
| 63 |
|
| 64 |
with torch.inference_mode():
|
|
|
|
| 82 |
log_conversation(chatbot)
|
| 83 |
|
| 84 |
|
| 85 |
+
def prepare_inputs(chatbot: List[List[str]], image_input: Any, video_input: Any):
|
| 86 |
# conversations = [{
|
| 87 |
# "from": "system",
|
| 88 |
# "value": "You are a helpful assistant, and your task is to provide reliable and structured responses to users."
|
|
|
|
| 103 |
if conv["from"] == "human":
|
| 104 |
conv["value"] = f'{image_placeholder}\n{conv["value"]}'
|
| 105 |
break
|
| 106 |
+
max_partition = IMAGE_MAX_PARTITION
|
| 107 |
+
image_input = [image_input]
|
| 108 |
+
|
| 109 |
+
if video_input is not None:
|
| 110 |
+
for conv in conversations:
|
| 111 |
+
if conv["from"] == "human":
|
| 112 |
+
conv["value"] = f'{image_placeholder}\n' * VIDEO_FRAME_NUMS + f'{conv["value"]}'
|
| 113 |
+
break
|
| 114 |
+
# extract video frames here
|
| 115 |
+
with VideoFileClip(video_input) as clip:
|
| 116 |
+
total_frames = int(clip.fps * clip.duration)
|
| 117 |
+
if total_frames <= VIDEO_FRAME_NUMS:
|
| 118 |
+
sampled_indices = range(total_frames)
|
| 119 |
+
else:
|
| 120 |
+
stride = total_frames / VIDEO_FRAME_NUMS
|
| 121 |
+
sampled_indices = [min(total_frames - 1, int((stride * i + stride * (i + 1)) / 2)) for i in range(VIDEO_FRAME_NUMS)]
|
| 122 |
+
frames = [clip.get_frame(index / clip.fps) for index in sampled_indices]
|
| 123 |
+
frames = [Image.fromarray(frame, mode='RGB') for frame in frames]
|
| 124 |
+
image_input = frames
|
| 125 |
+
max_partition = VIDEO_MAX_PARTITION
|
| 126 |
|
| 127 |
logger.info(conversations)
|
| 128 |
|
| 129 |
+
prompt, input_ids, pixel_values = model.preprocess_inputs(conversations, image_input, max_partition=max_partition)
|
| 130 |
attention_mask = torch.ne(input_ids, text_tokenizer.pad_token_id)
|
| 131 |
|
| 132 |
model_inputs = {
|
|
|
|
| 143 |
logger.info("[OVIS_CONV_END]")
|
| 144 |
|
| 145 |
def clear_chat():
|
| 146 |
+
return [], None, "", None
|
| 147 |
|
| 148 |
with open(f"{cur_dir}/resource/logo.svg", "r", encoding="utf-8") as svg_file:
|
| 149 |
svg_content = svg_file.read()
|
|
|
|
| 192 |
gr.HTML(html)
|
| 193 |
with gr.Row():
|
| 194 |
with gr.Column(scale=3):
|
| 195 |
+
input_type = gr.Radio(choices=["image + prompt", "video + prompt"], label="Select input type:", value="image + prompt", elem_classes="my_radio")
|
| 196 |
+
|
| 197 |
+
image_input = gr.Image(label="image", height=350, type="pil", visible=True)
|
| 198 |
+
video_input = gr.Video(label="video", height=350, format='mp4', visible=False)
|
| 199 |
+
with gr.Column(visible=True) as image_examples_col:
|
| 200 |
+
image_examples = gr.Examples(
|
| 201 |
+
examples=[
|
| 202 |
+
[f"{cur_dir}/examples/ovis2_math0.jpg", "Each face of the polyhedron shown is either a triangle or a square. Each square borders 4 triangles, and each triangle borders 3 squares. The polyhedron has 6 squares. How many triangles does it have?\n\nProvide a step-by-step solution to the problem, and conclude with 'the answer is' followed by the final solution."],
|
| 203 |
+
[f"{cur_dir}/examples/ovis2_math1.jpg", "A large square touches another two squares, as shown in the picture. The numbers inside the smaller squares indicate their areas. What is the area of the largest square?\n\nProvide a step-by-step solution to the problem, and conclude with 'the answer is' followed by the final solution."],
|
| 204 |
+
[f"{cur_dir}/examples/ovis2_figure0.png", "Explain this model."],
|
| 205 |
+
[f"{cur_dir}/examples/ovis2_figure1.png", "Organize the notes about GRPO in the figure."],
|
| 206 |
+
[f"{cur_dir}/examples/ovis2_multi0.jpg", "Posso avere un frappuccino e un caffè americano di taglia M? Quanto costa in totale?"],
|
| 207 |
+
],
|
| 208 |
+
inputs=[image_input, text_input]
|
| 209 |
+
)
|
| 210 |
+
|
| 211 |
+
def update_visibility_on_example(video_input, text_input):
|
| 212 |
+
return (gr.update(visible=True), text_input)
|
| 213 |
+
|
| 214 |
+
with gr.Column(visible=False) as video_examples_col:
|
| 215 |
+
video_examples = gr.Examples(
|
| 216 |
+
examples=[
|
| 217 |
+
[f"{cur_dir}/examples/video_demo_1.mp4", "Describe the video."]
|
| 218 |
+
],
|
| 219 |
+
inputs=[video_input, text_input],
|
| 220 |
+
fn = update_visibility_on_example,
|
| 221 |
+
run_on_click = True,
|
| 222 |
+
outputs=[video_input, text_input]
|
| 223 |
+
)
|
| 224 |
+
|
| 225 |
with gr.Column(scale=7):
|
| 226 |
chatbot = gr.Chatbot(label="Ovis", layout="panel", height=600, show_copy_button=True, latex_delimiters=latex_delimiters_set)
|
| 227 |
text_input.render()
|
| 228 |
with gr.Row():
|
| 229 |
send_btn = gr.Button("Send", variant="primary")
|
| 230 |
clear_btn = gr.Button("Clear", variant="secondary")
|
| 231 |
+
|
| 232 |
+
def update_input_and_clear(selected):
|
| 233 |
+
if selected == "image + prompt":
|
| 234 |
+
visibility_updates = (gr.update(visible=True), gr.update(visible=False),
|
| 235 |
+
gr.update(visible=True), gr.update(visible=False))
|
| 236 |
+
else:
|
| 237 |
+
visibility_updates = (gr.update(visible=False), gr.update(visible=True),
|
| 238 |
+
gr.update(visible=False), gr.update(visible=True))
|
| 239 |
+
clear_chat_outputs = clear_chat()
|
| 240 |
+
return visibility_updates + clear_chat_outputs
|
| 241 |
+
|
| 242 |
+
input_type.change(fn=update_input_and_clear, inputs=input_type,
|
| 243 |
+
outputs=[image_input, video_input, image_examples_col, video_examples_col, chatbot, image_input, text_input, video_input])
|
| 244 |
+
|
| 245 |
+
send_click_event = send_btn.click(submit_chat, [chatbot, text_input], [chatbot, text_input]).then(ovis_chat,[chatbot, image_input, video_input],chatbot)
|
| 246 |
+
submit_event = text_input.submit(submit_chat, [chatbot, text_input], [chatbot, text_input]).then(ovis_chat,[chatbot, image_input, video_input],chatbot)
|
| 247 |
+
clear_btn.click(clear_chat, outputs=[chatbot, image_input, text_input, video_input])
|
| 248 |
|
| 249 |
demo.launch()
|