Delete app.py

app.py

@@ -1,203 +0,0 @@
-# --- Imports bleiben unverändert ---
-import subprocess
-import sys
-import os
-from transformers import TextIteratorStreamer
-import argparse
-import time
-import subprocess
-import spaces
-import cumo.serve.gradio_web_server as gws
-from transformers import AutoProcessor, AutoTokenizer, AutoImageProcessor
-import datetime
-import json
-import gradio as gr
-import requests
-from PIL import Image
-from cumo.conversation import (default_conversation, conv_templates, SeparatorStyle)
-from cumo.constants import LOGDIR
-from cumo.model.language_model.llava_mistral import LlavaMistralForCausalLM
-from cumo.utils import (build_logger, server_error_msg, violates_moderation, moderation_msg)
-import hashlib
-import torch
-import io
-from cumo.constants import WORKER_HEART_BEAT_INTERVAL
-from cumo.utils import (build_logger, server_error_msg, pretty_print_semaphore)
-from cumo.model.builder import load_pretrained_model
-from cumo.mm_utils import process_images, load_image_from_base64, tokenizer_image_token
-from cumo.constants import IMAGE_TOKEN_INDEX, DEFAULT_IMAGE_TOKEN, DEFAULT_IM_START_TOKEN, DEFAULT_IM_END_TOKEN
-from threading import Thread
-
-# --- Model Setup ---
-headers = {"User-Agent": "CuMo"}
-no_change_btn = gr.Button()
-enable_btn = gr.Button(interactive=True)
-disable_btn = gr.Button(interactive=False)
-
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model_path = 'BenkHel/CumoThesis'
-model_base = 'mistralai/Mistral-7B-Instruct-v0.2'
-model_name = 'CuMo-mistral-7b'
-conv_mode = 'mistral_instruct_system'
-load_8bit = False
-load_4bit = False
-
-tokenizer, model, image_processor, context_len = load_pretrained_model(
-    model_path, model_base, model_name, load_8bit, load_4bit, device=device, use_flash_attn=False
-)
-model.config.training = False
-
-# --- Prompt ---
-FIXED_PROMPT = "<image>\nWhat material is this item and how to dispose of it?"
-
-# --- Functions ---
-def clear_history():
-    state = default_conversation.copy()
-    return (state, state.to_gradio_chatbot(), "", None) + (disable_btn,) * 5
-
-def add_text(state, imagebox, textbox, image_process_mode):
-    if state is None:
-        state = conv_templates[conv_mode].copy()
-    if imagebox is not None:
-        image = Image.open(imagebox).convert('RGB')
-        textbox = (FIXED_PROMPT, image, image_process_mode)
-        state.append_message(state.roles[0], textbox)
-        state.append_message(state.roles[1], None)
-        yield (state, state.to_gradio_chatbot(), "", None) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
-
-def delete_text(state, image_process_mode):
-    state.messages[-1][-1] = None
-    prev_human_msg = state.messages[-2]
-    if type(prev_human_msg[1]) in (tuple, list):
-        prev_human_msg[1] = (*prev_human_msg[1][:2], image_process_mode)
-    yield (state, state.to_gradio_chatbot(), "", None) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
-
-@spaces.GPU
-def generate(state, imagebox, textbox, image_process_mode, temperature, top_p, max_output_tokens):
-    prompt = FIXED_PROMPT
-    images = state.get_images(return_pil=True)
-    ori_prompt = prompt
-    num_image_tokens = 0
-
-    if images and len(images) > 0:
-        if len(images) != prompt.count(DEFAULT_IMAGE_TOKEN):
-            raise ValueError("Number of images does not match number of <image> tokens in prompt")
-        image_sizes = [image.size for image in images]
-        images = process_images(images, image_processor, model.config)
-        if isinstance(images, list):
-            images = [image.to(model.device, dtype=torch.float16) for image in images]
-        else:
-            images = images.to(model.device, dtype=torch.float16)
-        replace_token = DEFAULT_IMAGE_TOKEN
-        if getattr(model.config, 'mm_use_im_start_end', False):
-            replace_token = DEFAULT_IM_START_TOKEN + replace_token + DEFAULT_IM_END_TOKEN
-        prompt = prompt.replace(DEFAULT_IMAGE_TOKEN, replace_token)
-        num_image_tokens = prompt.count(replace_token) * model.get_vision_tower().num_patches
-        image_args = {"images": images, "image_sizes": image_sizes}
-    else:
-        image_args = {}
-
-    max_context_length = getattr(model.config, 'max_position_embeddings', 2048)
-    input_ids = tokenizer_image_token(prompt, tokenizer, IMAGE_TOKEN_INDEX, return_tensors='pt').unsqueeze(0).to(model.device)
-    max_new_tokens = min(512, max_context_length - input_ids.shape[-1] - num_image_tokens)
-    if max_new_tokens < 1:
-        yield json.dumps({"text": ori_prompt + "Exceeds max token length. Please start a new conversation.", "error_code": 0}).encode() + b"\0"
-        return
-
-    streamer = TextIteratorStreamer(tokenizer, skip_prompt=True, skip_special_tokens=True, timeout=15)
-    thread = Thread(target=model.generate, kwargs=dict(
-        inputs=input_ids,
-        do_sample=(temperature > 0.001),
-        temperature=temperature,
-        top_p=top_p,
-        max_new_tokens=max_new_tokens,
-        streamer=streamer,
-        use_cache=True,
-        pad_token_id=tokenizer.eos_token_id,
-        **image_args
-    ))
-    thread.start()
-    generated_text = ''
-    stop_str = state.sep if state.sep_style in [SeparatorStyle.SINGLE, SeparatorStyle.MPT] else state.sep2
-
-    for new_text in streamer:
-        generated_text += new_text
-        if generated_text.endswith(stop_str):
-            generated_text = generated_text[:-len(stop_str)]
-        state.messages[-1][-1] = generated_text
-        yield (state, state.to_gradio_chatbot(), "", None) + (disable_btn, disable_btn, disable_btn, enable_btn, enable_btn)
-    yield (state, state.to_gradio_chatbot(), "", None) + (enable_btn,) * 5
-    torch.cuda.empty_cache()
-
-# --- UI Setup ---
-textbox = gr.Textbox(
-    show_label=False, 
-    placeholder="Prompt is fixed: What material is this item and how to dispose of it.", 
-    container=False,
-    interactive=False
-)
-
-with gr.Blocks(title="CuMo", theme=gr.themes.Default(), css="""
-#buttons button {
-    min-width: min(120px,100%);
-}
-""") as demo:
-    state = gr.State()
-
-    gr.Markdown("# CuMo: Trained for waste management")
-    gr.Markdown(f"**Prompt:** `{FIXED_PROMPT}`")
-
-    with gr.Row():
-        with gr.Column(scale=3):
-            imagebox = gr.Image(label="Input Image", type="filepath")
-            image_process_mode = gr.Radio(
-                ["Crop", "Resize", "Pad", "Default"],
-                value="Default",
-                label="Preprocess for non-square image", visible=False)
-
-            cur_dir = './cumo/serve'
-            gr.Examples(examples=[
-                [f"{cur_dir}/examples/0165 CB.jpg"],
-                [f"{cur_dir}/examples/0225 PA.jpg"],
-                [f"{cur_dir}/examples/0787 GM.jpg"],
-                [f"{cur_dir}/examples/1396 A.jpg"],
-                [f"{cur_dir}/examples/2001 P.jpg"],
-                [f"{cur_dir}/examples/2658 PE.jpg"],
-                [f"{cur_dir}/examples/3113 R.jpg"],
-                [f"{cur_dir}/examples/3750 RPC.jpg"],
-                [f"{cur_dir}/examples/5033 CC.jpg"],
-                [f"{cur_dir}/examples/5307 B.jpg"],
-            ], inputs=[imagebox], cache_examples=False)
-
-            with gr.Accordion("Parameters", open=False):
-                temperature = gr.Slider(0.0, 1.0, value=0.2, step=0.1, interactive=True, label="Temperature")
-                top_p = gr.Slider(0.0, 1.0, value=0.7, step=0.1, interactive=True, label="Top P")
-                max_output_tokens = gr.Slider(0, 1024, value=512, step=64, interactive=True, label="Max output tokens")
-
-        with gr.Column(scale=8):
-            chatbot = gr.Chatbot(elem_id="chatbot", label="CuMo Chatbot", height=650, layout="panel")
-            with gr.Row():
-                with gr.Column(scale=8):
-                    textbox.render()
-                with gr.Column(scale=1, min_width=50):
-                    submit_btn = gr.Button(value="Send", variant="primary")
-            with gr.Row(elem_id="buttons") as button_row:
-                stop_btn = gr.Button(value="⏹️  Stop Generation", interactive=False)
-                regenerate_btn = gr.Button(value="🔄  Regenerate", interactive=False)
-                clear_btn = gr.Button(value="🗑️  Clear", interactive=False)
-
-    gr.Markdown(tos_markdown)
-    gr.Markdown(learn_more_markdown)
-    url_params = gr.JSON(visible=False)
-
-    # --- Event Bindings ---
-    btn_list = [regenerate_btn, clear_btn]
-    clear_btn.click(clear_history, None, [state, chatbot, textbox, imagebox] + btn_list, queue=False)
-    regenerate_btn.click(delete_text, [state, image_process_mode], [state, chatbot, textbox, imagebox] + btn_list
-    ).then(generate, [state, imagebox, textbox, image_process_mode, temperature, top_p, max_output_tokens], [state, chatbot, textbox, imagebox] + btn_list)
-    textbox.submit(add_text, [state, imagebox, textbox, image_process_mode], [state, chatbot, textbox, imagebox] + btn_list
-    ).then(generate, [state, imagebox, textbox, image_process_mode, temperature, top_p, max_output_tokens], [state, chatbot, textbox, imagebox] + btn_list)
-    submit_btn.click(add_text, [state, imagebox, textbox, image_process_mode], [state, chatbot, textbox, imagebox] + btn_list
-    ).then(generate, [state, imagebox, textbox, image_process_mode, temperature, top_p, max_output_tokens], [state, chatbot, textbox, imagebox] + btn_list)
-
-demo.queue(status_update_rate=10, api_open=False).launch()