Update app.py

app.py

@@ -1,79 +1,94 @@
 import gradio as gr
-from huggingface_hub import InferenceClient
-from transformers import LlavaProcessor, LlavaForConditionalGeneration, TextIteratorStreamer
-from PIL import Image
-from threading import Thread
+import torch
+from llava.model.builder import load_pretrained_model
+from llava.mm_utils import get_model_name_from_path, process_images, tokenizer_image_token
+from llava.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN
+from llava.conversation import conv_templates
+import copy
+from decord import VideoReader, cpu
+import numpy as np
 
-# Initialize model and processor
-model_id = "llava-hf/llava-interleave-qwen-0.5b-hf"
-processor = LlavaProcessor.from_pretrained(model_id)
-model = LlavaForConditionalGeneration.from_pretrained(model_id).to("cpu")
+# Load the model
+pretrained = "lmms-lab/LLaVA-Video-7B-Qwen2"
+model_name = "llava_qwen"
+device = "cuda" if torch.cuda.is_available() else "cpu"
+device_map = "auto"
 
-# Initialize inference clients
-client_mistral = InferenceClient("mistralai/Mistral-7B-Instruct-v0.3")
+print("Loading model...")
+tokenizer, model, image_processor, max_length = load_pretrained_model(pretrained, None, model_name, torch_dtype="bfloat16", device_map=device_map)
+model.eval()
+print("Model loaded successfully!")
 
-def llava(inputs):
-    """Processes an image and text input using Llava."""
-    try:
-        image = Image.open(inputs["files"][0]).convert("RGB")
-        prompt = f"<|im_start|>user <image>\n{inputs['text']}<|im_end|>"
-        processed = processor(prompt, image, return_tensors="pt").to("cpu")
-        return processed
-    except Exception as e:
-        print(f"Error in llava function: {e}")
-        return None
+def load_video(video_path, max_frames_num, fps=1, force_sample=False):
+    if max_frames_num == 0:
+        return np.zeros((1, 336, 336, 3))
+    vr = VideoReader(video_path, ctx=cpu(0), num_threads=1)
+    total_frame_num = len(vr)
+    video_time = total_frame_num / vr.get_avg_fps()
+    fps = round(vr.get_avg_fps()/fps)
+    frame_idx = [i for i in range(0, len(vr), fps)]
+    frame_time = [i/fps for i in frame_idx]
+    if len(frame_idx) > max_frames_num or force_sample:
+        sample_fps = max_frames_num
+        uniform_sampled_frames = np.linspace(0, total_frame_num - 1, sample_fps, dtype=int)
+        frame_idx = uniform_sampled_frames.tolist()
+        frame_time = [i/vr.get_avg_fps() for i in frame_idx]
+    frame_time = ",".join([f"{i:.2f}s" for i in frame_time])
+    spare_frames = vr.get_batch(frame_idx).asnumpy()
+    return spare_frames, frame_time, video_time
 
-def respond(message, history):
-    """Generate a response based on text or image input."""
-    try:
-        if "files" in message and message["files"]:
-            # Handle image + text input
-            inputs = llava(message)
-            if inputs is None:
-                raise ValueError("Failed to process image input")
-            
-            streamer = TextIteratorStreamer(skip_prompt=True, skip_special_tokens=True)
-            thread = Thread(target=model.generate, kwargs=dict(inputs=inputs, max_new_tokens=512, streamer=streamer))
-            thread.start()
-            
-            buffer = ""
-            for new_text in streamer:
-                buffer += new_text
-                history[-1][1] = buffer
-                yield history, history
-        else:
-            # Handle text-only input
-            user_message = message["text"]
-            history.append([user_message, None])
-            prompt = [{"role": "user", "content": msg[0]} for msg in history if msg[0]]
-            response = client_mistral.chat_completion(prompt, max_tokens=200)
-            bot_message = response["choices"][0]["message"]["content"]
-            history[-1][1] = bot_message
-            yield history, history
-    except Exception as e:
-        print(f"Error in respond function: {e}")
-        history[-1][1] = f"An error occurred: {str(e)}"
-        yield history, history
+def process_video(video_path, question):
+    max_frames_num = 64
+    video, frame_time, video_time = load_video(video_path, max_frames_num, 1, force_sample=True)
+    video = image_processor.preprocess(video, return_tensors="pt")["pixel_values"].to(device).bfloat16()
+    video = [video]
+
+    conv_template = "qwen_1_5"
+    time_instruction = f"The video lasts for {video_time:.2f} seconds, and {len(video[0])} frames are uniformly sampled from it. These frames are located at {frame_time}. Please answer the following questions related to this video."
+    
+    full_question = DEFAULT_IMAGE_TOKEN + f"{time_instruction}\n{question}"
+    
+    conv = copy.deepcopy(conv_templates[conv_template])
+    conv.append_message(conv.roles[0], full_question)
+    conv.append_message(conv.roles[1], None)
+    prompt_question = conv.get_prompt()
+    
+    input_ids = tokenizer_image_token(prompt_question, tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt").unsqueeze(0).to(device)
+    
+    with torch.no_grad():
+        output = model.generate(
+            input_ids,
+            images=video,
+            modalities=["video"],
+            do_sample=False,
+            temperature=0,
+            max_new_tokens=4096,
+        )
+    
+    response = tokenizer.batch_decode(output, skip_special_tokens=True)[0].strip()
+    return response
+
+def gradio_interface(video_file, question):
+    if video_file is None:
+        return "Please upload a video file."
+    response = process_video(video_file, question)
+    return response
 
 # Set up Gradio interface
 with gr.Blocks() as demo:
-    chatbot = gr.Chatbot()
+    gr.Markdown("# 🌋📹 LLaVA-Video Chatbot")
     with gr.Row():
         with gr.Column():
-            text_input = gr.Textbox(placeholder="Enter your message...")
-            file_input = gr.File(label="Upload an image")
-
-    def handle_text(text, history=[]):
-        """Handle text input and generate responses."""
-        return respond({"text": text}, history)
-
-    def handle_file_upload(files, history=[]):
-        """Handle file uploads and generate responses."""
-        return respond({"files": files, "text": "Describe this image."}, history)
-
-    # Connect components to callbacks
-    text_input.submit(handle_text, [text_input, chatbot], [chatbot, chatbot])
-    file_input.change(handle_file_upload, [file_input, chatbot], [chatbot, chatbot])
+            video_input = gr.Video()
+            question_input = gr.Textbox(label="User Question", placeholder="Ask a question about the video...")
+            submit_button = gr.Button("Ask LLaVA-Video")
+        output = gr.Textbox(label="LLaVA-Video Response")
+    
+    submit_button.click(
+        fn=gradio_interface,
+        inputs=[video_input, question_input],
+        outputs=output
+    )
 
-# Launch the Gradio app
-demo.launch()
+if __name__ == "__main__":
+    demo.launch(show_error=True)