File size: 16,394 Bytes
9c37045 |
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 |
#!/usr/bin/env python3
"""
Qwen2.5-Omni Complete Multimodal Demo
A comprehensive Gradio interface for the Qwen2.5-Omni-3B multimodal AI model
Optimized for Apple Silicon (MPS) with efficient memory management
"""
import os
import gc
import sys
import time
import signal
import warnings
from typing import List, Dict, Any, Optional, Tuple, Union
import tempfile
import soundfile as sf
# Suppress warnings for cleaner output
warnings.filterwarnings("ignore", category=FutureWarning)
warnings.filterwarnings("ignore", category=UserWarning)
import torch
import numpy as np
import gradio as gr
from PIL import Image
# Global variables for model and processor
model = None
processor = None
device = None
def cleanup_resources():
"""Clean up model and free memory"""
global model, processor
try:
if model is not None:
del model
model = None
if processor is not None:
del processor
processor = None
# Force garbage collection
gc.collect()
# Clear CUDA/MPS cache if available
if torch.cuda.is_available():
torch.cuda.empty_cache()
elif hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
torch.mps.empty_cache()
print("β
Resources cleaned up successfully")
except Exception as e:
print(f"β οΈ Warning during cleanup: {e}")
def signal_handler(signum, frame):
"""Handle interrupt signals gracefully"""
print("\nπ Interrupt received, cleaning up...")
cleanup_resources()
sys.exit(0)
# Register signal handlers
signal.signal(signal.SIGINT, signal_handler)
signal.signal(signal.SIGTERM, signal_handler)
def load_model():
"""Load the Qwen2.5-Omni model and processor"""
global model, processor, device
if model is not None:
return "β
Model already loaded!"
try:
# Check device
if torch.backends.mps.is_available():
device = torch.device("mps")
device_info = "π Using Apple Silicon MPS acceleration"
else:
device = torch.device("cpu")
device_info = "β οΈ Using CPU (MPS not available)"
# Import the specific Qwen2.5-Omni classes
from transformers import Qwen2_5OmniForConditionalGeneration, Qwen2_5OmniProcessor
# Load processor with optimizations
processor = Qwen2_5OmniProcessor.from_pretrained(
"Qwen/Qwen2.5-Omni-3B",
trust_remote_code=True,
use_fast=True # Use fast tokenizer if available
)
# Load model with memory-efficient settings - keep bfloat16 for all functionalities
model = Qwen2_5OmniForConditionalGeneration.from_pretrained(
"Qwen/Qwen2.5-Omni-3B",
torch_dtype=torch.bfloat16,
trust_remote_code=True,
device_map="auto" if device.type != "mps" else None,
low_cpu_mem_usage=True,
use_safetensors=True,
attn_implementation="sdpa"
)
# Immediately disable the audio generation module to prevent any initialization overhead
model.disable_talker()
print("π€ Talker module disabled immediately after loading to optimize performance")
# Explicitly move to device for MPS while keeping bfloat16
if device.type == "mps":
model = model.to(device=device, dtype=torch.bfloat16)
print(f"π§ Model loaded with dtype: bfloat16 (memory efficient)")
# Clear any cached memory after loading
gc.collect()
gc.collect() # Run twice for good measure
if hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
torch.mps.empty_cache()
return f"β
Model loaded successfully!\n{device_info}\nDevice: {device}"
except Exception as e:
return f"β Error loading model: {str(e)}"
def text_chat(message, history, system_prompt, temperature, max_tokens):
"""Handle text-only conversations correctly."""
if model is None or processor is None:
history.append((message, "β Error: Model is not loaded. Please load the model first."))
return history, ""
if not message or not message.strip():
return history, ""
try:
conversation = []
if system_prompt and system_prompt.strip():
conversation.append({"role": "system", "content": [{"type": "text", "text": system_prompt}]})
# Correctly process history for the model
for user_msg, assistant_msg in history:
if user_msg:
conversation.append({"role": "user", "content": [{"type": "text", "text": user_msg}]})
if assistant_msg:
# Avoid adding error messages to the model's context
if not assistant_msg.startswith("β Error:"):
conversation.append({"role": "assistant", "content": [{"type": "text", "text": assistant_msg}]})
conversation.append({"role": "user", "content": [{"type": "text", "text": message}]})
text = processor.apply_chat_template(conversation, add_generation_prompt=True, tokenize=False)
inputs = processor(text=text, return_tensors="pt", padding=True).to(device)
with torch.no_grad():
generated_ids = model.generate(
**inputs,
max_new_tokens=max_tokens,
temperature=temperature,
do_sample=True,
pad_token_id=processor.tokenizer.eos_token_id
)
input_token_len = inputs["input_ids"].shape[1]
response_ids = generated_ids[:, input_token_len:]
response = processor.batch_decode(response_ids, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
history.append((message, response))
return history, ""
except Exception as e:
import traceback
traceback.print_exc()
error_message = f"β Error in text chat: {str(e)}"
history.append((message, error_message))
return history, ""
def multimodal_chat(message, image, audio, history, system_prompt, temperature, max_tokens):
"""
Handle multimodal conversations (text, image, and audio) using the correct
processor.apply_chat_template method as per the official documentation.
"""
global model, processor, device
if model is None or processor is None:
history.append((message, "β Error: Model is not loaded. Please load the model first."))
return history, ""
if not message.strip() and image is None and audio is None:
history.append(("", "Please provide an input (text, image, or audio)."))
return history, ""
# --- Create a temporary directory for media files ---
temp_dir = tempfile.mkdtemp()
try:
# --- Build the conversation history in the required format ---
conversation = []
if system_prompt and system_prompt.strip():
conversation.append({"role": "system", "content": [{"type": "text", "text": system_prompt}]})
# Process Gradio history into the conversation format
for user_turn, bot_turn in history:
# For simplicity, we only process the text part of the history.
# A more robust solution would parse the [Image] and [Audio] tags
# and reconstruct the full multimodal history.
if user_turn:
conversation.append({"role": "user", "content": [{"type": "text", "text": user_turn.replace("[Image]", "").replace("[Audio]", "").strip()}]})
if bot_turn and not bot_turn.startswith("β Error:"):
conversation.append({"role": "assistant", "content": [{"type": "text", "text": bot_turn}]})
# --- Prepare the current user's turn ---
current_content = []
user_message_for_history = ""
# Process text
if message and message.strip():
current_content.append({"type": "text", "text": message})
user_message_for_history += message
# Process image
if image is not None:
# --- FIX: Resize large images to prevent OOM errors ---
MAX_PIXELS = 1024 * 1024 # 1 megapixel
if image.width * image.height > MAX_PIXELS:
image.thumbnail((1024, 1024), Image.Resampling.LANCZOS)
temp_image_path = os.path.join(temp_dir, "temp_image.png")
image.save(temp_image_path)
current_content.append({"type": "image", "image": temp_image_path})
user_message_for_history += " [Image]"
# Process audio
if audio is not None:
sample_rate, audio_data = audio
temp_audio_path = os.path.join(temp_dir, "temp_audio.wav")
sf.write(temp_audio_path, audio_data, sample_rate)
current_content.append({"type": "audio", "audio": temp_audio_path})
user_message_for_history += " [Audio]"
if not current_content:
history.append(("", "Please provide some input."))
return history, ""
conversation.append({"role": "user", "content": current_content})
# --- Use `apply_chat_template` as per the documentation ---
# This is the single, correct way to process all modalities.
inputs = processor.apply_chat_template(
conversation,
add_generation_prompt=True,
tokenize=True,
return_dict=True,
return_tensors="pt",
padding=True,
).to(device)
# --- Generation ---
with torch.no_grad():
# Note: The model's generate function does not return audio directly in this setup
# We are focusing on getting the text response right first.
generated_ids = model.generate(
**inputs,
max_new_tokens=max_tokens,
temperature=temperature,
do_sample=True,
pad_token_id=processor.tokenizer.eos_token_id,
# return_audio=False # This might be needed if audio output is enabled by default
)
# The generate call for the full Omni model might return a tuple (text_ids, audio_wav)
# We handle both cases to be safe.
if isinstance(generated_ids, tuple):
response_ids = generated_ids[0]
else:
response_ids = generated_ids
input_token_len = inputs["input_ids"].shape[1]
response_ids_decoded = response_ids[:, input_token_len:]
response = processor.batch_decode(response_ids_decoded, skip_special_tokens=True, clean_up_tokenization_spaces=False)[0]
history.append((user_message_for_history.strip(), response))
return history, ""
except Exception as e:
import traceback
error_message = f"β Multimodal chat error: {traceback.format_exc()}"
print(error_message) # Print full traceback to console for debugging
history.append((message, f"β Error: {e}"))
return history, ""
finally:
# --- Clean up temporary files ---
if os.path.exists(temp_dir):
import shutil
shutil.rmtree(temp_dir)
def clear_history():
"""Clear chat history"""
return []
def clear_model_cache():
"""Clear model cache and free memory"""
global model, processor
try:
cleanup_resources()
# Clear additional caches
if torch.cuda.is_available():
torch.cuda.empty_cache()
elif hasattr(torch.backends, 'mps') and torch.backends.mps.is_available():
torch.mps.empty_cache()
return "β
Cache cleared successfully! Click 'Load Model' to reload."
except Exception as e:
return f"β Error clearing cache: {str(e)}"
def create_interface():
"""Create the complete Gradio interface with the fix."""
with gr.Blocks(title="Qwen2.5-Omni Multimodal Demo", theme=gr.themes.Soft()) as demo:
gr.Markdown("""
# π€ Qwen2.5-Omni Complete Multimodal Demo
A comprehensive and corrected Gradio interface for the Qwen2.5-Omni-3B model.
""")
with gr.Row():
with gr.Column(scale=2):
load_btn = gr.Button("π Load Model", variant="primary")
with gr.Column(scale=2):
cache_clear_btn = gr.Button("π§Ή Clear Cache", variant="secondary")
with gr.Column(scale=3):
model_status = gr.Textbox(label="Model Status", value="Model not loaded", interactive=False)
load_btn.click(load_model, outputs=model_status)
cache_clear_btn.click(clear_model_cache, outputs=model_status)
with gr.Tabs():
with gr.Tab("π¬ Text Chat"):
text_chatbot = gr.Chatbot(label="Conversation", height=450)
with gr.Row():
text_msg = gr.Textbox(label="Your message", placeholder="Type your message...", scale=4, container=False)
text_send = gr.Button("Send", variant="primary", scale=1)
with gr.Row():
text_clear = gr.Button("Clear History")
with gr.Accordion("Settings", open=False):
text_system = gr.Textbox(label="System Prompt", value="You are a helpful AI assistant.")
text_temp = gr.Slider(0.1, 1.5, value=0.7, label="Temperature")
text_max_tokens = gr.Slider(50, 1000, value=500, label="Max New Tokens", step=50)
text_send.click(text_chat, inputs=[text_msg, text_chatbot, text_system, text_temp, text_max_tokens], outputs=[text_chatbot, text_msg])
text_msg.submit(text_chat, inputs=[text_msg, text_chatbot, text_system, text_temp, text_max_tokens], outputs=[text_chatbot, text_msg])
text_clear.click(clear_history, outputs=text_chatbot)
with gr.Tab("π Multimodal Chat"):
multi_chatbot = gr.Chatbot(label="Multimodal Conversation", height=450)
multi_text = gr.Textbox(label="Text Message (optional)", placeholder="Describe what you want to know...", scale=4, container=False)
with gr.Row():
multi_image = gr.Image(label="Upload Image (optional)", type="pil")
multi_audio = gr.Audio(label="Upload Audio (optional)", type="numpy")
with gr.Row():
multi_send = gr.Button("Send Multimodal Input", variant="primary")
multi_clear = gr.Button("Clear History")
with gr.Accordion("Settings", open=False):
multi_system = gr.Textbox(label="System Prompt", value="You are Qwen, capable of understanding images, audio, and text.")
multi_temp = gr.Slider(0.1, 1.5, value=0.7, label="Temperature")
multi_max_tokens = gr.Slider(50, 1000, value=500, label="Max New Tokens", step=50)
multi_send.click(multimodal_chat, inputs=[multi_text, multi_image, multi_audio, multi_chatbot, multi_system, multi_temp, multi_max_tokens], outputs=[multi_chatbot, multi_text])
multi_clear.click(clear_history, outputs=multi_chatbot)
with gr.Tab("βΉοΈ Model Info"):
# Placeholder for model info content
gr.Markdown("Model information will be displayed here.")
return demo
if __name__ == "__main__":
try:
os.environ["TOKENIZERS_PARALLELISM"] = "false"
os.environ["OMP_NUM_THREADS"] = "1"
demo = create_interface()
print("π Starting Qwen2.5-Omni Gradio Demo...")
print("π Memory management optimizations enabled")
print("π Access the interface at: http://localhost:7860")
demo.launch(
server_name="0.0.0.0",
server_port=7860,
share=False,
show_error=True,
quiet=False
)
except KeyboardInterrupt:
print("\nπ Shutting down gracefully...")
cleanup_resources()
except Exception as e:
print(f"β Error starting demo: {e}")
cleanup_resources() |