Spaces:
Running
on
Zero
Running
on
Zero
File size: 11,003 Bytes
9933c70 1a1eaff 9933c70 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 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 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 |
import torch
import gradio as gr
from transformers import AutoTokenizer, AutoModel, AutoImageProcessor
from PIL import Image
import gc
import os
import spaces
# Model configuration
MODEL_PATH = "nvidia/Llama-Nemotron-Nano-VL-8B-V1"
# Load model globally
print("Loading model...")
model = AutoModel.from_pretrained(
MODEL_PATH,
torch_dtype=torch.bfloat16,
low_cpu_mem_usage=True,
trust_remote_code=True,
).eval()
tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH)
image_processor = AutoImageProcessor.from_pretrained(
MODEL_PATH,
trust_remote_code=True
)
print("Model loaded successfully!")
def move_to_device(obj, device):
"""Recursively move tensors to device"""
if torch.is_tensor(obj):
return obj.to(device)
elif isinstance(obj, dict):
return {k: move_to_device(v, device) for k, v in obj.items()}
elif isinstance(obj, list):
return [move_to_device(v, device) for v in obj]
elif isinstance(obj, tuple):
return tuple(move_to_device(v, device) for v in obj)
elif hasattr(obj, 'to'):
return obj.to(device)
else:
return obj
@spaces.GPU(duration=60)
def chat_text_only(message):
try:
device = "cuda"
# Move entire model to GPU
model.to(device)
generation_config = dict(
max_new_tokens=512,
do_sample=True,
temperature=0.7,
eos_token_id=tokenizer.eos_token_id
)
# Tokenize on CPU then move to GPU
inputs = tokenizer(message, return_tensors="pt")
inputs = move_to_device(inputs, device)
# Generate
with torch.no_grad():
response, _ = model.chat(
tokenizer,
None,
message,
generation_config,
history=None,
return_history=True
)
# Move model back to CPU
model.to("cpu")
torch.cuda.empty_cache()
gc.collect()
return response
except Exception as e:
# Ensure model is back on CPU even if error occurs
model.to("cpu")
torch.cuda.empty_cache()
gc.collect()
return f"Error: {str(e)}"
@spaces.GPU(duration=60)
def chat_with_image(image, message):
if image is None:
return "Please upload an image."
try:
device = "cuda"
# Move entire model to GPU
model.to(device)
generation_config = dict(
max_new_tokens=512,
do_sample=True,
temperature=0.7,
eos_token_id=tokenizer.eos_token_id
)
# Process image
image_features = image_processor(image)
# Move all image features to GPU
image_features = move_to_device(image_features, device)
# Add image token to message if not present
if "<image>" not in message:
message = f"<image>\n{message}"
# Generate
with torch.no_grad():
response = model.chat(
tokenizer=tokenizer,
question=message,
generation_config=generation_config,
**image_features
)
# Move model back to CPU
model.to("cpu")
torch.cuda.empty_cache()
gc.collect()
return response
except Exception as e:
# Ensure model is back on CPU even if error occurs
model.to("cpu")
torch.cuda.empty_cache()
gc.collect()
return f"Error: {str(e)}"
@spaces.GPU(duration=60)
def chat_with_two_images(image1, image2, message):
if image1 is None or image2 is None:
return "Please upload both images."
try:
device = "cuda"
# Move entire model to GPU
model.to(device)
generation_config = dict(
max_new_tokens=512,
do_sample=True,
temperature=0.7,
eos_token_id=tokenizer.eos_token_id
)
# Process both images
image_features = image_processor([image1, image2])
# Move all image features to GPU
image_features = move_to_device(image_features, device)
# Format message for two images
if "<image-1>" not in message and "<image-2>" not in message:
message = f"<image-1>: <image>\n<image-2>: <image>\n{message}"
# Generate
with torch.no_grad():
response = model.chat(
tokenizer=tokenizer,
question=message,
generation_config=generation_config,
**image_features
)
# Move model back to CPU
model.to("cpu")
torch.cuda.empty_cache()
gc.collect()
return response
except Exception as e:
# Ensure model is back on CPU even if error occurs
model.to("cpu")
torch.cuda.empty_cache()
gc.collect()
return f"Error: {str(e)}"
# Create Gradio interface
def create_interface():
with gr.Blocks(title="Llama Nemotron Nano VL 8B", theme=gr.themes.Soft()) as demo:
gr.Markdown("# 🦙 Llama Nemotron Nano VL 8B Vision-Language Model")
gr.Markdown("Chat with a powerful vision-language model that can understand both text and images!")
with gr.Tabs():
# Text-only chat tab
with gr.TabItem("💬 Text Chat"):
gr.Markdown("### Chat with the model using text only")
with gr.Row():
with gr.Column():
text_input = gr.Textbox(
label="Your message",
placeholder="Ask me anything...",
lines=3
)
text_submit = gr.Button("Send", variant="primary")
with gr.Column():
text_output = gr.Textbox(
label="Model Response",
lines=10,
max_lines=20
)
text_submit.click(
chat_text_only,
inputs=[text_input],
outputs=[text_output]
)
# Example questions
gr.Examples(
examples=[
["What is artificial intelligence?"],
["Explain quantum computing in simple terms."],
["What happened in 1969?"],
["Write a short story about a robot."]
],
inputs=[text_input]
)
# Single image chat tab
with gr.TabItem("🖼️ Image + Text Chat"):
gr.Markdown("### Upload an image and ask questions about it")
with gr.Row():
with gr.Column():
image_input = gr.Image(
label="Upload Image",
type="pil"
)
image_text_input = gr.Textbox(
label="Your question about the image",
placeholder="What do you see in this image?",
lines=3
)
image_submit = gr.Button("Analyze", variant="primary")
with gr.Column():
image_output = gr.Textbox(
label="Model Response",
lines=10,
max_lines=20
)
image_submit.click(
chat_with_image,
inputs=[image_input, image_text_input],
outputs=[image_output]
)
# Example prompts
gr.Examples(
examples=[
["Describe what you see in this image."],
["What objects are in this image?"],
["Extract any text from this image."],
["What is the main subject of this image?"]
],
inputs=[image_text_input]
)
# Two images comparison tab
with gr.TabItem("🖼️🖼️ Compare Two Images"):
gr.Markdown("### Upload two images and ask the model to compare them")
with gr.Row():
with gr.Column():
image1_input = gr.Image(
label="First Image",
type="pil"
)
image2_input = gr.Image(
label="Second Image",
type="pil"
)
two_images_text_input = gr.Textbox(
label="Your question about both images",
placeholder="Compare these two images...",
lines=3
)
two_images_submit = gr.Button("Compare", variant="primary")
with gr.Column():
two_images_output = gr.Textbox(
label="Model Response",
lines=10,
max_lines=20
)
two_images_submit.click(
chat_with_two_images,
inputs=[image1_input, image2_input, two_images_text_input],
outputs=[two_images_output]
)
# Example prompts
gr.Examples(
examples=[
["What are the main differences between these two images?"],
["Describe both images briefly."],
["Which image is more colorful?"],
["Compare the subjects in these images."]
],
inputs=[two_images_text_input]
)
# Footer
gr.Markdown("---")
gr.Markdown("⚡ Powered by NVIDIA Llama Nemotron Nano VL 8B")
return demo
# Create and launch the interface
if __name__ == "__main__":
demo = create_interface()
demo.queue() # Enable queuing for Zero GPU
demo.launch(
server_name="0.0.0.0",
server_port=7860,
ssr_mode=False
) |