Update app.py

app.py

@@ -1,22 +1,20 @@
-# from .demo_modelpart import InferenceDemo
 import gradio as gr
 import os
 from threading import Thread
-
-# import time
+from queue import Queue
+import time
 import cv2
-
 import datetime
-# import copy
 import torch
-
 import spaces
 import numpy as np
+import json
+import hashlib
+import PIL
+from typing import Iterator
 
 from llava import conversation as conversation_lib
 from llava.constants import DEFAULT_IMAGE_TOKEN
-
-
 from llava.constants import (
     IMAGE_TOKEN_INDEX,
     DEFAULT_IMAGE_TOKEN,
@@ -37,16 +35,7 @@ from serve_constants import html_header
 import requests
 from PIL import Image
 from io import BytesIO
-from transformers import TextStreamer, TextIteratorStreamer
-
-import hashlib
-import PIL
-import base64
-
-import gradio as gr
-import gradio_client
-import subprocess
-import sys
+from transformers import TextIteratorStreamer
 
 external_log_dir = "./logs"
 LOGDIR = external_log_dir
@@ -61,13 +50,9 @@ def install_gradio_4_35_0():
     else:
         print("Gradio 4.35.0 is already installed.")
 
-# Call the function to install Gradio 4.35.0 if needed
 install_gradio_4_35_0()
 
-import gradio as gr
-import gradio_client
 print(f"Gradio version: {gr.__version__}")
-print(f"Gradio-client version: {gradio_client.__version__}")
 
 def get_conv_log_filename():
     t = datetime.datetime.now()
@@ -80,12 +65,12 @@ class InferenceDemo(object):
     ) -> None:
         disable_torch_init()
 
-        self.tokenizer, self.model, self.image_processor, self.context_len = (
-            tokenizer,
-            model,
-            image_processor,
-            context_len,
-        )
+        self.tokenizer = tokenizer
+        self.model = model
+        self.image_processor = image_processor
+        self.context_len = context_len
+        
+        model_name = get_model_name_from_path(model_path)
 
         if "llama-2" in model_name.lower():
             conv_mode = "llava_llama_2"
@@ -108,31 +93,43 @@ class InferenceDemo(object):
             )
         else:
             args.conv_mode = conv_mode
+        
         self.conv_mode = conv_mode
         self.conversation = conv_templates[args.conv_mode].copy()
         self.num_frames = args.num_frames
 
+def process_stream(streamer: TextIteratorStreamer, history: list, q: Queue):
+    """Process the output stream and put partial text into a queue"""
+    try:
+        current_message = ""
+        for new_text in streamer:
+            current_message += new_text
+            history[-1][1] = current_message
+            q.put(history.copy())
+            time.sleep(0.02)  # Add a small delay to prevent overloading
+    except Exception as e:
+        print(f"Error in process_stream: {e}")
+    finally:
+        q.put(None)  # Signal that we're done
+
+def stream_output(history: list, q: Queue) -> Iterator[list]:
+    """Yield updated history as it comes through the queue"""
+    while True:
+        val = q.get()
+        if val is None:
+            break
+        yield val
+        q.task_done()
 
 def is_valid_video_filename(name):
     video_extensions = ["avi", "mp4", "mov", "mkv", "flv", "wmv", "mjpeg"]
-
     ext = name.split(".")[-1].lower()
-
-    if ext in video_extensions:
-        return True
-    else:
-        return False
+    return ext in video_extensions
 
 def is_valid_image_filename(name):
-    image_extensions = ["jpg", "jpeg", "png", "bmp", "gif", "tiff", "webp", "heic", "heif", "jfif", "svg", "eps", "raw"] 
-
+    image_extensions = ["jpg", "jpeg", "png", "bmp", "gif", "tiff", "webp", "heic", "heif", "jfif", "svg", "eps", "raw"]
     ext = name.split(".")[-1].lower()
-
-    if ext in image_extensions:
-        return True
-    else:
-        return False
-
+    return ext in image_extensions
 
 def sample_frames(video_file, num_frames):
     video = cv2.VideoCapture(video_file)
@@ -141,54 +138,33 @@ def sample_frames(video_file, num_frames):
     frames = []
     for i in range(total_frames):
         ret, frame = video.read()
-        pil_img = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
         if not ret:
             continue
+        pil_img = Image.fromarray(cv2.cvtColor(frame, cv2.COLOR_BGR2RGB))
         if i % interval == 0:
             frames.append(pil_img)
     video.release()
     return frames
 
-
 def load_image(image_file):
-    if image_file.startswith("http") or image_file.startswith("https"):
+    if image_file.startswith(("http://", "https://")):
         response = requests.get(image_file)
         if response.status_code == 200:
             image = Image.open(BytesIO(response.content)).convert("RGB")
         else:
-            print("failed to load the image")
+            print("Failed to load the image")
+            return None
     else:
-        print("Load image from local file")
-        print(image_file)
+        print("Load image from local file:", image_file)
         image = Image.open(image_file).convert("RGB")
-
     return image
 
-
 def clear_history(history):
-
+    global our_chatbot
     our_chatbot.conversation = conv_templates[our_chatbot.conv_mode].copy()
-
     return None
 
-
-def clear_response(history):
-    for index_conv in range(1, len(history)):
-        # loop until get a text response from our model.
-        conv = history[-index_conv]
-        if not (conv[0] is None):
-            break
-    question = history[-index_conv][0]
-    history = history[:-index_conv]
-    return history, question
-
-
-# def print_like_dislike(x: gr.LikeData):
-#     print(x.index, x.value, x.liked)
-
-
 def add_message(history, message):
-    # history=[]
     global our_chatbot
     if len(history) == 0:
         our_chatbot = InferenceDemo(
@@ -201,36 +177,46 @@ def add_message(history, message):
         history.append((message["text"], None))
     return history, gr.MultimodalTextbox(value=None, interactive=False)
 
-
 @spaces.GPU
 def bot(history):
+    global start_tstamp, finish_tstamp
+    
+    start_tstamp = time.time()
     text = history[-1][0]
     images_this_term = []
-    text_this_term = ""
-    # import pdb;pdb.set_trace()
     num_new_images = 0
+    
     for i, message in enumerate(history[:-1]):
-        if type(message[0]) is tuple:
+        if isinstance(message[0], tuple):
             images_this_term.append(message[0][0])
             if is_valid_video_filename(message[0][0]):
-                # 不接受视频
                 raise ValueError("Video is not supported")
-                num_new_images += our_chatbot.num_frames
             elif is_valid_image_filename(message[0][0]):
-                print("#### Load image from local file",message[0][0])
                 num_new_images += 1
             else:
                 raise ValueError("Invalid image file")
         else:
             num_new_images = 0
 
-    # for message in history[-i-1:]:
-    #     images_this_term.append(message[0][0])
-
-    assert len(images_this_term) > 0, "must have an image"
-    # image_files = (args.image_file).split(',')
-    # image = [load_image(f) for f in images_this_term if f]
-
+    assert len(images_this_term) > 0, "Must have an image"
+    
+    image_list = []
+    for f in images_this_term:
+        if is_valid_video_filename(f):
+            image_list += sample_frames(f, our_chatbot.num_frames)
+        elif is_valid_image_filename(f):
+            image_list.append(load_image(f))
+        else:
+            raise ValueError("Invalid image file")
+    
+    image_tensor = [
+        our_chatbot.image_processor.preprocess(f, return_tensors="pt")["pixel_values"][0]
+        .half()
+        .to(our_chatbot.model.device)
+        for f in image_list
+    ]
+    
+    # Process image hashes
     all_image_hash = []
     for image_path in images_this_term:
         with open(image_path, "rb") as image_file:
@@ -248,35 +234,12 @@ def bot(history):
             if not os.path.isfile(filename):
                 os.makedirs(os.path.dirname(filename), exist_ok=True)
                 image.save(filename)
-    
-    image_list = []
-    for f in images_this_term:
-        if is_valid_video_filename(f):
-            image_list += sample_frames(f, our_chatbot.num_frames)
-        elif is_valid_image_filename(f):
-            image_list.append(load_image(f))
-        else:
-            raise ValueError("Invalid image file")
-    
-    image_tensor = [
-        our_chatbot.image_processor.preprocess(f, return_tensors="pt")["pixel_values"][
-            0
-        ]
-        .half()
-        .to(our_chatbot.model.device)
-        for f in image_list
-    ]
-
 
     image_tensor = torch.stack(image_tensor)
     image_token = DEFAULT_IMAGE_TOKEN * num_new_images
-    # if our_chatbot.model.config.mm_use_im_start_end:
-    #     inp = DEFAULT_IM_START_TOKEN + image_token + DEFAULT_IM_END_TOKEN + "\n" + inp
-    # else:
-    inp = text
-    inp = image_token + "\n" + inp
+    inp = image_token + "\n" + text
+    
     our_chatbot.conversation.append_message(our_chatbot.conversation.roles[0], inp)
-    # image = None
     our_chatbot.conversation.append_message(our_chatbot.conversation.roles[1], None)
     prompt = our_chatbot.conversation.get_prompt()
 
@@ -287,10 +250,7 @@ def bot(history):
         .unsqueeze(0)
         .to(our_chatbot.model.device)
     )
-    # input_ids = tokenizer_image_token(
-    #         prompt, our_chatbot.tokenizer, IMAGE_TOKEN_INDEX, return_tensors="pt"
-    #     ).unsqueeze(0).to(our_chatbot.model.device)
-    # print("### input_id",input_ids)
+
     stop_str = (
         our_chatbot.conversation.sep
         if our_chatbot.conversation.sep_style != SeparatorStyle.TWO
@@ -300,16 +260,23 @@ def bot(history):
     stopping_criteria = KeywordsStoppingCriteria(
         keywords, our_chatbot.tokenizer, input_ids
     )
-    streamer = TextStreamer(
-        our_chatbot.tokenizer, skip_prompt=True, skip_special_tokens=True
+
+    # Set up streaming
+    q = Queue()
+    streamer = TextIteratorStreamer(
+        our_chatbot.tokenizer,
+        skip_prompt=True,
+        skip_special_tokens=True
+    )
+
+    # Start generation in a separate thread
+    thread = Thread(
+        target=process_stream,
+        args=(streamer, history, q)
     )
-    # streamer = TextIteratorStreamer(
-    #     our_chatbot.tokenizer, skip_prompt=True, skip_special_tokens=True
-    # )
-    print(our_chatbot.model.device)
-    print(input_ids.device)
-    print(image_tensor.device)
-    # import pdb;pdb.set_trace()
+    thread.start()
+
+    # Start the generation
     with torch.inference_mode():
         output_ids = our_chatbot.model.generate(
             input_ids,
@@ -318,86 +285,29 @@ def bot(history):
             temperature=0.2,
             max_new_tokens=1024,
             streamer=streamer,
-            use_cache=False,
+            use_cache=True,
             stopping_criteria=[stopping_criteria],
         )
 
-    outputs = our_chatbot.tokenizer.decode(output_ids[0]).strip()
-    if outputs.endswith(stop_str):
-        outputs = outputs[: -len(stop_str)]
-    our_chatbot.conversation.messages[-1][-1] = outputs
-
-    history[-1] = [text, outputs]
-    print("#### history",history)
+    finish_tstamp = time.time()
     
+    # Log conversation
     with open(get_conv_log_filename(), "a") as fout:
         data = {
+            "tstamp": round(finish_tstamp, 4),
             "type": "chat",
             "model": "Pangea-7b",
+            "start": round(start_tstamp, 4),
+            "finish": round(finish_tstamp, 4),
             "state": history,
             "images": all_image_hash,
         }
         fout.write(json.dumps(data) + "\n")
-    return history
-    # generate_kwargs = dict(
-    #     inputs=input_ids,
-    #     streamer=streamer,
-    #     images=image_tensor,
-    #     max_new_tokens=1024,
-    #     do_sample=True,
-    #     temperature=0.2,
-    #     num_beams=1,
-    #     use_cache=False,
-    #     stopping_criteria=[stopping_criteria],
-    # )
-    
-    # t = Thread(target=our_chatbot.model.generate, kwargs=generate_kwargs)
-    # t.start()
-    
-    # outputs = []
-    # for text in streamer:
-    #     outputs.append(text)
-    #     yield "".join(outputs)
-    
-    # our_chatbot.conversation.messages[-1][-1] = "".join(outputs)
-    # history[-1] = [text, "".join(outputs)]
-    
-
-txt = gr.Textbox(
-    scale=4,
-    show_label=False,
-    placeholder="Enter text and press enter.",
-    container=False,
-)
 
-with gr.Blocks(
-    css=".message-wrap.svelte-1lcyrx4>div.svelte-1lcyrx4  img {min-width: 40px}",
-) as demo:
+    # Return a generator that will yield updated history
+    return stream_output(history, q)
 
-    # Informations
-    title_markdown = """
-        # LLaVA-NeXT Interleave
-        [[Blog]](https://llava-vl.github.io/blog/2024-06-16-llava-next-interleave/)  [[Code]](https://github.com/LLaVA-VL/LLaVA-NeXT) [[Model]](https://huggingface.co/lmms-lab/llava-next-interleave-7b)
-        Note: The internleave checkpoint is updated (Date: Jul. 24, 2024), the wrong checkpiont is used before.
-    """
-    
-    
-    tos_markdown = """
-    ### TODO!. Terms of use
-    By using this service, users are required to agree to the following terms:
-    The service is a research preview intended for non-commercial use only. It only provides limited safety measures and may generate offensive content. It must not be used for any illegal, harmful, violent, racist, or sexual purposes. The service may collect user dialogue data for future research.
-    Please click the "Flag" button if you get any inappropriate answer! We will collect those to keep improving our moderator.
-    For an optimal experience, please use desktop computers for this demo, as mobile devices may compromise its quality.
-    """
-    learn_more_markdown = """
-    ### TODO!. License
-    The service is a research preview intended for non-commercial use only, subject to the model [License](https://github.com/facebookresearch/llama/blob/main/MODEL_CARD.md) of LLaMA, [Terms of Use](https://openai.com/policies/terms-of-use) of the data generated by OpenAI, and [Privacy Practices](https://chrome.google.com/webstore/detail/sharegpt-share-your-chatg/daiacboceoaocpibfodeljbdfacokfjb) of ShareGPT. Please contact us if you find any potential violation.
-    """
-    models = [
-        "LLaVA-Interleave-7B",
-    ]
-    cur_dir = os.path.dirname(os.path.abspath(__file__))
-    # gr.Markdown(title_markdown)
+with gr.Blocks(css=".message-wrap.svelte-1lcyrx4>div.svelte-1lcyrx4 img {min-width: 40px}") as demo:
     gr.HTML(html_header)
     
     with gr.Column():
@@ -408,10 +318,8 @@ with gr.Blocks(
             upvote_btn = gr.Button(value="👍  Upvote", interactive=True)
             downvote_btn = gr.Button(value="👎  Downvote", interactive=True)
             flag_btn = gr.Button(value="⚠️  Flag", interactive=True)
-            # stop_btn = gr.Button(value="⏹️  Stop Generation", interactive=True)
             regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=True)
             clear_btn = gr.Button(value="🗑️  Clear history", interactive=True)
-            
 
         chat_input = gr.MultimodalTextbox(
             interactive=True,
@@ -421,11 +329,11 @@ with gr.Blocks(
             submit_btn="🚀"
         )
 
-        print(cur_dir)
+        cur_dir = os.path.dirname(os.path.abspath(__file__))
         gr.Examples(
-                examples_per_page=20,
-                examples=[
-                    [
+            examples_per_page=20,
+            examples=[
+                [
                         {
                             "files": [
                                 f"{cur_dir}/examples/user_example_07.jpg",
@@ -449,160 +357,47 @@ with gr.Blocks(
                             "text": "Why this image funny?",
                         },
                     ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/norway.jpg",
-                            ],
-                            "text": "Analysieren, in welchem Land diese Szene höchstwahrscheinlich gedreht wurde.",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/totoro.jpg",
-                            ],
-                            "text": "¿En qué anime aparece esta escena? ¿Puedes presentarlo?",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/africa.jpg",
-                            ],
-                            "text": "इस तस्वीर में हर एक दृश्य तत्व का क्या प्रतिनिधित्व करता है?",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/hot_ballon.jpg",
-                            ],
-                            "text": "ฉากบอลลูนลมร้อนในภาพนี้อาจอยู่ที่ไหน? สถานที่นี้มีความพิเศษอย่างไร?",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/bar.jpg",
-                            ],
-                            "text": "Você pode me dar ideias de design baseadas no tema de coquetéis deste letreiro?",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/pink_lake.jpg",
-                            ],
-                            "text": "Обясни защо езерото на този остров е в този цвят.",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/hanzi.jpg",
-                            ],
-                            "text": "Can you describe in Hebrew the evolution process of these four Chinese characters from pictographs to modern characters?",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/ballon.jpg",
-                            ],
-                            "text": "இந்த காட்சியை விவரிக்கவும், மேலும் இந்த படத்தின் அடிப்படையில் துருக்கியில் இந்த காட்சியுடன் தொடர்பான சில பிரபலமான நிகழ்வுகள் என்ன?",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/pie.jpg",
-                            ],
-                            "text": "Décrivez ce graphique. Quelles informations pouvons-nous en tirer?",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/camera.jpg",
-                            ],
-                            "text": "Apa arti dari dua angka di sebelah kiri yang ditampilkan di layar kamera?",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/dog.jpg",
-                            ],
-                            "text": "이 강아지의 표정을 보고 어떤 기분이나 감정을 느끼고 있는지 설명해 주시겠어요?",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/book.jpg",
-                            ],
-                            "text": "What language is the text in, and what does the title mean in English?",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/food.jpg",
-                            ],
-                            "text": "Unaweza kunipa kichocheo cha kutengeneza hii pancake?",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/line chart.jpg",
-                            ],
-                            "text": "Hãy trình bày những xu hướng mà bạn quan sát được từ biểu đồ và hiện tượng xã hội tiềm ẩn từ đó.",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/south africa.jpg",
-                            ],
-                            "text": "Waar is hierdie plek? Help my om ’n reisroete vir hierdie land te beplan.",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/girl.jpg",
-                            ],
-                            "text": "لماذا هذه الصورة مضحكة؟",
-                        },
-                    ],
-                    [
-                        {
-                            "files": [
-                                f"{cur_dir}/examples/eagles.jpg",
-                            ],
-                            "text": "Какой креатив должен быть в этом логотипе?",
-                        },
-                    ],
-                ],
-                inputs=[chat_input],
-                label="Image",
-            )
+            ],
+            inputs=[chat_input],
+            label="Image",
+        )
 
     chat_msg = chat_input.submit(
-        add_message, [chatbot, chat_input], [chatbot, chat_input]
+        add_message, 
+        [chatbot, chat_input], 
+        [chatbot, chat_input],
+        queue=False
+    ).then(
+        bot,
+        chatbot,
+        chatbot,
+        api_name="bot_response"
+    ).then(
+        lambda: gr.MultimodalTextbox(interactive=True), 
+        None, 
+        [chat_input]
     )
-    bot_msg = chat_msg.then(bot, chatbot, chatbot, api_name="bot_response")
-    bot_msg.then(lambda: gr.MultimodalTextbox(interactive=True), None, [chat_input])
 
-    # chatbot.like(print_like_dislike, None, None)
     clear_btn.click(
-        fn=clear_history, inputs=[chatbot], outputs=[chatbot], api_name="clear_all"
+        fn=clear_history, 
+        inputs=[chatbot], 
+        outputs=[chatbot], 
+        api_name="clear_all",
+        queue=False
     )
 
+    regenerate_btn.click(
+        fn=lambda history: history[:-1],
+        inputs=[chatbot],
+        outputs=[chatbot],
+        queue=False
+    ).then(
+        bot,
+        chatbot,
+        chatbot
+    )
 
-demo.queue()
+demo.queue(concurrency_count=5)
     
 if __name__ == "__main__":
     import argparse