Spaces:
				
			
			
	
			
			
		Running
		
			on 
			
			Zero
	
	
	
			
			
	
	
	
	
		
		
		Running
		
			on 
			
			Zero
	Update app.py
Browse files
    	
        app.py
    CHANGED
    
    | @@ -13,6 +13,7 @@ import torch | |
| 13 | 
             
            import numpy as np
         | 
| 14 | 
             
            from PIL import Image
         | 
| 15 | 
             
            import edge_tts
         | 
|  | |
| 16 |  | 
| 17 | 
             
            from transformers import (
         | 
| 18 | 
             
                AutoModelForCausalLM,
         | 
| @@ -113,7 +114,6 @@ ENABLE_CPU_OFFLOAD = os.getenv("ENABLE_CPU_OFFLOAD", "0") == "1" | |
| 113 |  | 
| 114 | 
             
            dtype = torch.float16 if device.type == "cuda" else torch.float32
         | 
| 115 |  | 
| 116 | 
            -
             | 
| 117 | 
             
            # STABLE DIFFUSION IMAGE GENERATION MODELS
         | 
| 118 |  | 
| 119 | 
             
            if torch.cuda.is_available():
         | 
| @@ -201,7 +201,6 @@ def save_image(img: Image.Image) -> str: | |
| 201 | 
             
                img.save(unique_name)
         | 
| 202 | 
             
                return unique_name
         | 
| 203 |  | 
| 204 | 
            -
             | 
| 205 | 
             
            # GEMMA3-4B MULTIMODAL MODEL
         | 
| 206 |  | 
| 207 | 
             
            gemma3_model_id = "google/gemma-3-4b-it"
         | 
| @@ -210,6 +209,25 @@ gemma3_model = Gemma3ForConditionalGeneration.from_pretrained( | |
| 210 | 
             
            ).eval()
         | 
| 211 | 
             
            gemma3_processor = AutoProcessor.from_pretrained(gemma3_model_id)
         | 
| 212 |  | 
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
| 213 |  | 
| 214 | 
             
            # MAIN GENERATION FUNCTION
         | 
| 215 |  | 
| @@ -228,7 +246,7 @@ def generate( | |
| 228 |  | 
| 229 | 
             
                lower_text = text.lower().strip()
         | 
| 230 |  | 
| 231 | 
            -
                #  | 
| 232 | 
             
                if (lower_text.startswith("@lightningv5") or 
         | 
| 233 | 
             
                    lower_text.startswith("@lightningv4") or 
         | 
| 234 | 
             
                    lower_text.startswith("@turbov3")):
         | 
| @@ -280,20 +298,76 @@ def generate( | |
| 280 | 
             
                    yield gr.Image(image_path)
         | 
| 281 | 
             
                    return
         | 
| 282 |  | 
| 283 | 
            -
                # GEMMA3-4B  | 
| 284 | 
             
                if lower_text.startswith("@gemma3-4b"):
         | 
| 285 | 
            -
                    #  | 
| 286 | 
            -
                     | 
| 287 | 
            -
             | 
| 288 | 
            -
             | 
| 289 | 
            -
                         | 
| 290 | 
            -
                         | 
| 291 | 
            -
             | 
| 292 | 
            -
                             | 
| 293 | 
            -
             | 
| 294 | 
            -
             | 
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
| 295 | 
             
                            ]
         | 
| 296 | 
            -
                         | 
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
|  | |
| 297 | 
             
                    else:
         | 
| 298 | 
             
                        messages = [
         | 
| 299 | 
             
                            {"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}]},
         | 
| @@ -319,7 +393,7 @@ def generate( | |
| 319 | 
             
                    thread = Thread(target=gemma3_model.generate, kwargs=generation_kwargs)
         | 
| 320 | 
             
                    thread.start()
         | 
| 321 | 
             
                    buffer = ""
         | 
| 322 | 
            -
                    yield progress_bar_html("Processing with Gemma3-4b")
         | 
| 323 | 
             
                    for new_text in streamer:
         | 
| 324 | 
             
                        buffer += new_text
         | 
| 325 | 
             
                        time.sleep(0.01)
         | 
| @@ -408,7 +482,9 @@ demo = gr.ChatInterface( | |
| 408 | 
             
                ],
         | 
| 409 | 
             
                examples=[
         | 
| 410 | 
             
                    [{"text": "@gemma3-4b Explain the Image", "files": ["examples/3.jpg"]}],
         | 
| 411 | 
            -
                    [{"text": "@gemma3-4b  | 
|  | |
|  | |
| 412 | 
             
                    [{"text": "@gemma3-4b Where do the major drought happen?", "files": ["examples/111.png"]}],
         | 
| 413 | 
             
                    [{"text": "@gemma3-4b Transcription of the letter", "files": ["examples/222.png"]}],
         | 
| 414 | 
             
                    ['@lightningv5 Chocolate dripping from a donut'],
         | 
| @@ -420,9 +496,9 @@ demo = gr.ChatInterface( | |
| 420 | 
             
                ],
         | 
| 421 | 
             
                cache_examples=False,
         | 
| 422 | 
             
                type="messages",
         | 
| 423 | 
            -
                description="# **Imagineo Chat `@gemma3-4b 'prompt..', @lightningv5, etc..`**",
         | 
| 424 | 
             
                fill_height=True,
         | 
| 425 | 
            -
                textbox=gr.MultimodalTextbox(label="Query Input", file_types=["image"], file_count="multiple", placeholder="use the tags | 
| 426 | 
             
                stop_btn="Stop Generation",
         | 
| 427 | 
             
                multimodal=True,
         | 
| 428 | 
             
            )
         | 
|  | |
| 13 | 
             
            import numpy as np
         | 
| 14 | 
             
            from PIL import Image
         | 
| 15 | 
             
            import edge_tts
         | 
| 16 | 
            +
            import cv2  # New import for video processing
         | 
| 17 |  | 
| 18 | 
             
            from transformers import (
         | 
| 19 | 
             
                AutoModelForCausalLM,
         | 
|  | |
| 114 |  | 
| 115 | 
             
            dtype = torch.float16 if device.type == "cuda" else torch.float32
         | 
| 116 |  | 
|  | |
| 117 | 
             
            # STABLE DIFFUSION IMAGE GENERATION MODELS
         | 
| 118 |  | 
| 119 | 
             
            if torch.cuda.is_available():
         | 
|  | |
| 201 | 
             
                img.save(unique_name)
         | 
| 202 | 
             
                return unique_name
         | 
| 203 |  | 
|  | |
| 204 | 
             
            # GEMMA3-4B MULTIMODAL MODEL
         | 
| 205 |  | 
| 206 | 
             
            gemma3_model_id = "google/gemma-3-4b-it"
         | 
|  | |
| 209 | 
             
            ).eval()
         | 
| 210 | 
             
            gemma3_processor = AutoProcessor.from_pretrained(gemma3_model_id)
         | 
| 211 |  | 
| 212 | 
            +
            # VIDEO PROCESSING HELPER
         | 
| 213 | 
            +
            def downsample_video(video_path):
         | 
| 214 | 
            +
                vidcap = cv2.VideoCapture(video_path)
         | 
| 215 | 
            +
                total_frames = int(vidcap.get(cv2.CAP_PROP_FRAME_COUNT))
         | 
| 216 | 
            +
                fps = vidcap.get(cv2.CAP_PROP_FPS)
         | 
| 217 | 
            +
                frames = []
         | 
| 218 | 
            +
                # Sample 10 evenly spaced frames.
         | 
| 219 | 
            +
                frame_indices = np.linspace(0, total_frames - 1, 10, dtype=int)
         | 
| 220 | 
            +
                for i in frame_indices:
         | 
| 221 | 
            +
                    vidcap.set(cv2.CAP_PROP_POS_FRAMES, i)
         | 
| 222 | 
            +
                    success, image = vidcap.read()
         | 
| 223 | 
            +
                    if success:
         | 
| 224 | 
            +
                        # Convert from BGR to RGB and then to PIL Image.
         | 
| 225 | 
            +
                        image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
         | 
| 226 | 
            +
                        pil_image = Image.fromarray(image)
         | 
| 227 | 
            +
                        timestamp = round(i / fps, 2)
         | 
| 228 | 
            +
                        frames.append((pil_image, timestamp))
         | 
| 229 | 
            +
                vidcap.release()
         | 
| 230 | 
            +
                return frames
         | 
| 231 |  | 
| 232 | 
             
            # MAIN GENERATION FUNCTION
         | 
| 233 |  | 
|  | |
| 246 |  | 
| 247 | 
             
                lower_text = text.lower().strip()
         | 
| 248 |  | 
| 249 | 
            +
                # IMAGE GENERATION BRANCH (Stable Diffusion models)
         | 
| 250 | 
             
                if (lower_text.startswith("@lightningv5") or 
         | 
| 251 | 
             
                    lower_text.startswith("@lightningv4") or 
         | 
| 252 | 
             
                    lower_text.startswith("@turbov3")):
         | 
|  | |
| 298 | 
             
                    yield gr.Image(image_path)
         | 
| 299 | 
             
                    return
         | 
| 300 |  | 
| 301 | 
            +
                # GEMMA3-4B TEXT & MULTIMODAL (image) Branch
         | 
| 302 | 
             
                if lower_text.startswith("@gemma3-4b"):
         | 
| 303 | 
            +
                    # If it is video, let the dedicated branch handle it.
         | 
| 304 | 
            +
                    if lower_text.startswith("@gemma3-4b-video"):
         | 
| 305 | 
            +
                        pass  # video branch is handled below.
         | 
| 306 | 
            +
                    else:
         | 
| 307 | 
            +
                        # Remove the gemma3 flag from the prompt.
         | 
| 308 | 
            +
                        prompt_clean = re.sub(r"@gemma3-4b", "", text, flags=re.IGNORECASE).strip().strip('"')
         | 
| 309 | 
            +
                        if files:
         | 
| 310 | 
            +
                            # If image files are provided, load them.
         | 
| 311 | 
            +
                            images = [load_image(f) for f in files]
         | 
| 312 | 
            +
                            messages = [{
         | 
| 313 | 
            +
                                "role": "user",
         | 
| 314 | 
            +
                                "content": [
         | 
| 315 | 
            +
                                    *[{"type": "image", "image": image} for image in images],
         | 
| 316 | 
            +
                                    {"type": "text", "text": prompt_clean},
         | 
| 317 | 
            +
                                ]
         | 
| 318 | 
            +
                            }]
         | 
| 319 | 
            +
                        else:
         | 
| 320 | 
            +
                            messages = [
         | 
| 321 | 
            +
                                {"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}]},
         | 
| 322 | 
            +
                                {"role": "user", "content": [{"type": "text", "text": prompt_clean}]}
         | 
| 323 | 
             
                            ]
         | 
| 324 | 
            +
                        inputs = gemma3_processor.apply_chat_template(
         | 
| 325 | 
            +
                            messages, add_generation_prompt=True, tokenize=True,
         | 
| 326 | 
            +
                            return_dict=True, return_tensors="pt"
         | 
| 327 | 
            +
                        ).to(gemma3_model.device, dtype=torch.bfloat16)
         | 
| 328 | 
            +
                        streamer = TextIteratorStreamer(
         | 
| 329 | 
            +
                            gemma3_processor.tokenizer, timeout=20.0, skip_prompt=True, skip_special_tokens=True
         | 
| 330 | 
            +
                        )
         | 
| 331 | 
            +
                        generation_kwargs = {
         | 
| 332 | 
            +
                            **inputs,
         | 
| 333 | 
            +
                            "streamer": streamer,
         | 
| 334 | 
            +
                            "max_new_tokens": max_new_tokens,
         | 
| 335 | 
            +
                            "do_sample": True,
         | 
| 336 | 
            +
                            "temperature": temperature,
         | 
| 337 | 
            +
                            "top_p": top_p,
         | 
| 338 | 
            +
                            "top_k": top_k,
         | 
| 339 | 
            +
                            "repetition_penalty": repetition_penalty,
         | 
| 340 | 
            +
                        }
         | 
| 341 | 
            +
                        thread = Thread(target=gemma3_model.generate, kwargs=generation_kwargs)
         | 
| 342 | 
            +
                        thread.start()
         | 
| 343 | 
            +
                        buffer = ""
         | 
| 344 | 
            +
                        yield progress_bar_html("Processing with Gemma3-4b")
         | 
| 345 | 
            +
                        for new_text in streamer:
         | 
| 346 | 
            +
                            buffer += new_text
         | 
| 347 | 
            +
                            time.sleep(0.01)
         | 
| 348 | 
            +
                            yield buffer
         | 
| 349 | 
            +
                        return
         | 
| 350 | 
            +
             | 
| 351 | 
            +
                # NEW: GEMMA3-4B VIDEO Branch
         | 
| 352 | 
            +
                if lower_text.startswith("@gemma3-4b-video"):
         | 
| 353 | 
            +
                    # Remove the video flag from the prompt.
         | 
| 354 | 
            +
                    prompt_clean = re.sub(r"@gemma3-4b-video", "", text, flags=re.IGNORECASE).strip().strip('"')
         | 
| 355 | 
            +
                    if files:
         | 
| 356 | 
            +
                        # Assume the first file is a video.
         | 
| 357 | 
            +
                        video_path = files[0]
         | 
| 358 | 
            +
                        frames = downsample_video(video_path)
         | 
| 359 | 
            +
                        messages = [
         | 
| 360 | 
            +
                            {"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}]},
         | 
| 361 | 
            +
                            {"role": "user", "content": [{"type": "text", "text": prompt_clean}]}
         | 
| 362 | 
            +
                        ]
         | 
| 363 | 
            +
                        # Append each frame as an image with a timestamp label.
         | 
| 364 | 
            +
                        for frame in frames:
         | 
| 365 | 
            +
                            image, timestamp = frame
         | 
| 366 | 
            +
                            # Save the frame image to a temporary unique filename.
         | 
| 367 | 
            +
                            image_path = f"video_frame_{uuid.uuid4().hex}.png"
         | 
| 368 | 
            +
                            image.save(image_path)
         | 
| 369 | 
            +
                            messages[1]["content"].append({"type": "text", "text": f"Frame {timestamp}:"})
         | 
| 370 | 
            +
                            messages[1]["content"].append({"type": "image", "url": image_path})
         | 
| 371 | 
             
                    else:
         | 
| 372 | 
             
                        messages = [
         | 
| 373 | 
             
                            {"role": "system", "content": [{"type": "text", "text": "You are a helpful assistant."}]},
         | 
|  | |
| 393 | 
             
                    thread = Thread(target=gemma3_model.generate, kwargs=generation_kwargs)
         | 
| 394 | 
             
                    thread.start()
         | 
| 395 | 
             
                    buffer = ""
         | 
| 396 | 
            +
                    yield progress_bar_html("Processing with Gemma3-4b Video")
         | 
| 397 | 
             
                    for new_text in streamer:
         | 
| 398 | 
             
                        buffer += new_text
         | 
| 399 | 
             
                        time.sleep(0.01)
         | 
|  | |
| 482 | 
             
                ],
         | 
| 483 | 
             
                examples=[
         | 
| 484 | 
             
                    [{"text": "@gemma3-4b Explain the Image", "files": ["examples/3.jpg"]}],
         | 
| 485 | 
            +
                    [{"text": "@gemma3-4b-video Explain what is happening in this video ?", "files": ["examples/oreo.mp4"]}],
         | 
| 486 | 
            +
                    [{"text": "@gemma3-4b-video Summarize the events in this video", "files": ["examples/sky.mp4"]}],
         | 
| 487 | 
            +
                    [{"text": "@gemma3-4b-video What is in the video ?", "files": ["examples/redlight.mp4"]}],
         | 
| 488 | 
             
                    [{"text": "@gemma3-4b Where do the major drought happen?", "files": ["examples/111.png"]}],
         | 
| 489 | 
             
                    [{"text": "@gemma3-4b Transcription of the letter", "files": ["examples/222.png"]}],
         | 
| 490 | 
             
                    ['@lightningv5 Chocolate dripping from a donut'],
         | 
|  | |
| 496 | 
             
                ],
         | 
| 497 | 
             
                cache_examples=False,
         | 
| 498 | 
             
                type="messages",
         | 
| 499 | 
            +
                description="# **Imagineo Chat `@gemma3-4b 'prompt..', @gemma3-4b-video, @lightningv5, etc..`**",
         | 
| 500 | 
             
                fill_height=True,
         | 
| 501 | 
            +
                textbox=gr.MultimodalTextbox(label="Query Input", file_types=["image", "video"], file_count="multiple", placeholder="use the tags @gemma3-4b for multimodal, @gemma3-4b-video for video, @lightningv5, @lightningv4, @turbov3 for image gen !"),
         | 
| 502 | 
             
                stop_btn="Stop Generation",
         | 
| 503 | 
             
                multimodal=True,
         | 
| 504 | 
             
            )
         | 
