Upload HauauCS_Abliterated_Qwen_3_5_4B_LLM_model.ipynb

HauauCS_Abliterated_Qwen_3_5_4B_LLM_model.ipynb

@@ -0,0 +1,808 @@
+{
+  "nbformat": 4,
+  "nbformat_minor": 0,
+  "metadata": {
+    "colab": {
+      "provenance": [],
+      "gpuType": "T4"
+    },
+    "kernelspec": {
+      "name": "python3",
+      "display_name": "Python 3"
+    },
+    "language_info": {
+      "name": "python"
+    },
+    "accelerator": "GPU"
+  },
+  "cells": [
+    {
+      "cell_type": "markdown",
+      "source": [
+        "# HauauCS Abliterated Qwen 3.5 -4B LLM model (image and/or text input)"
+      ],
+      "metadata": {
+        "id": "ewVsp7wRAH2j"
+      }
+    },
+    {
+      "cell_type": "code",
+      "execution_count": null,
+      "metadata": {
+        "id": "lC55DC9m_4Zz"
+      },
+      "outputs": [],
+      "source": [
+        "\n",
+        "# ============================================================\n",
+        "# QWEN3.5-4B UNCENSORED - GOOGLE COLAB T4\n",
+        "#\n",
+        "# Model:\n",
+        "# DreamFast/Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Safetensor-Benchmark\n",
+        "#\n",
+        "# GPU: NVIDIA T4 16GB\n",
+        "# Quantization: BitsAndBytes NF4 4-bit\n",
+        "# Compute: FP16\n",
+        "# Attention: SDPA\n",
+        "#\n",
+        "# Features:\n",
+        "# - Text and image conversations\n",
+        "# - Conversation history\n",
+        "# - Separate THINKS and SAYS sections\n",
+        "# - Larger Gradio chat interface\n",
+        "# - Expandable latest-response panels\n",
+        "# - Adjustable generation settings\n",
+        "# - HF_TOKEN from Colab Secrets\n",
+        "# ============================================================\n",
+        "\n",
+        "\n",
+        "# ============================================================\n",
+        "# 1. INSTALL DEPENDENCIES\n",
+        "# ============================================================\n",
+        "\n",
+        "import sys\n",
+        "import subprocess\n",
+        "\n",
+        "subprocess.run(\n",
+        "    [\n",
+        "        sys.executable, \"-m\", \"pip\", \"install\", \"-q\", \"-U\",\n",
+        "        \"bitsandbytes\",\n",
+        "        \"accelerate\",\n",
+        "        \"transformers\",\n",
+        "        \"gradio>=6.0\",\n",
+        "        \"huggingface_hub\",\n",
+        "        \"pillow\"\n",
+        "    ],\n",
+        "    check=True\n",
+        ")\n",
+        "\n",
+        "\n",
+        "# ============================================================\n",
+        "# 2. IMPORTS\n",
+        "# ============================================================\n",
+        "\n",
+        "import os\n",
+        "import re\n",
+        "import gc\n",
+        "import torch\n",
+        "import gradio as gr\n",
+        "\n",
+        "from PIL import Image\n",
+        "from pathlib import Path\n",
+        "\n",
+        "from google.colab import userdata\n",
+        "\n",
+        "from huggingface_hub import hf_hub_download\n",
+        "\n",
+        "from transformers import (\n",
+        "    AutoProcessor,\n",
+        "    AutoTokenizer,\n",
+        "    AutoModelForMultimodalLM,\n",
+        "    BitsAndBytesConfig\n",
+        ")\n",
+        "\n",
+        "\n",
+        "# ============================================================\n",
+        "# 3. CONFIGURATION\n",
+        "# ============================================================\n",
+        "\n",
+        "MODEL_ID = (\n",
+        "    \"DreamFast/\"\n",
+        "    \"Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Safetensor-Benchmark\"\n",
+        ")\n",
+        "\n",
+        "BASE_MODEL_ID = \"Qwen/Qwen3.5-4B\"\n",
+        "\n",
+        "HF_TOKEN = userdata.get(\"HF_TOKEN\")\n",
+        "\n",
+        "if not HF_TOKEN:\n",
+        "    raise ValueError(\n",
+        "        \"HF_TOKEN not found in Colab Secrets. \"\n",
+        "        \"Please add your Hugging Face token.\"\n",
+        "    )\n",
+        "\n",
+        "if not torch.cuda.is_available():\n",
+        "    raise RuntimeError(\"No CUDA GPU detected.\")\n",
+        "\n",
+        "print(\"GPU:\", torch.cuda.get_device_name(0))\n",
+        "print(\"VRAM:\", round(\n",
+        "    torch.cuda.get_device_properties(0).total_memory / 1024**3, 2\n",
+        "), \"GB\")\n",
+        "\n",
+        "torch.backends.cuda.matmul.allow_tf32 = True\n",
+        "\n",
+        "\n",
+        "# ============================================================\n",
+        "# 4. LOAD CHAT TEMPLATE\n",
+        "# ============================================================\n",
+        "\n",
+        "print(\"\\nLoading Qwen3.5 tokenizer...\")\n",
+        "\n",
+        "tokenizer = AutoTokenizer.from_pretrained(\n",
+        "    BASE_MODEL_ID,\n",
+        "    token=HF_TOKEN,\n",
+        "    trust_remote_code=True\n",
+        ")\n",
+        "\n",
+        "if not tokenizer.chat_template:\n",
+        "\n",
+        "    print(\"Downloading original Qwen3.5 chat template...\")\n",
+        "\n",
+        "    template_path = hf_hub_download(\n",
+        "        repo_id=BASE_MODEL_ID,\n",
+        "        filename=\"chat_template.jinja\",\n",
+        "        token=HF_TOKEN\n",
+        "    )\n",
+        "\n",
+        "    with open(template_path, \"r\", encoding=\"utf-8\") as f:\n",
+        "        chat_template = f.read()\n",
+        "\n",
+        "    tokenizer.chat_template = chat_template\n",
+        "\n",
+        "else:\n",
+        "    chat_template = tokenizer.chat_template\n",
+        "\n",
+        "print(\"Chat template loaded.\")\n",
+        "\n",
+        "\n",
+        "# ============================================================\n",
+        "# 5. LOAD PROCESSOR\n",
+        "# ============================================================\n",
+        "\n",
+        "print(\"\\nLoading multimodal processor...\")\n",
+        "\n",
+        "processor = AutoProcessor.from_pretrained(\n",
+        "    MODEL_ID,\n",
+        "    token=HF_TOKEN,\n",
+        "    trust_remote_code=True\n",
+        ")\n",
+        "\n",
+        "# Use the compatible original Qwen3.5 tokenizer/template.\n",
+        "processor.tokenizer = tokenizer\n",
+        "processor.chat_template = chat_template\n",
+        "\n",
+        "print(\"Processor loaded.\")\n",
+        "\n",
+        "\n",
+        "# ============================================================\n",
+        "# 6. 4-BIT QUANTIZATION\n",
+        "# ============================================================\n",
+        "\n",
+        "quantization_config = BitsAndBytesConfig(\n",
+        "    load_in_4bit=True,\n",
+        "    bnb_4bit_quant_type=\"nf4\",\n",
+        "    bnb_4bit_compute_dtype=torch.float16,\n",
+        "    bnb_4bit_use_double_quant=True\n",
+        ")\n",
+        "\n",
+        "\n",
+        "# ============================================================\n",
+        "# 7. LOAD MODEL\n",
+        "# ============================================================\n",
+        "\n",
+        "print(\"\\nLoading model in 4-bit...\")\n",
+        "\n",
+        "model = AutoModelForMultimodalLM.from_pretrained(\n",
+        "    MODEL_ID,\n",
+        "    token=HF_TOKEN,\n",
+        "    quantization_config=quantization_config,\n",
+        "    dtype=torch.float16,\n",
+        "    device_map=\"auto\",\n",
+        "    low_cpu_mem_usage=True,\n",
+        "    attn_implementation=\"sdpa\",\n",
+        "    trust_remote_code=True\n",
+        ")\n",
+        "\n",
+        "model.eval()\n",
+        "\n",
+        "print(\"\\nModel loaded successfully.\")\n",
+        "\n",
+        "print(\"Model device:\", model.get_input_embeddings().weight.device)\n",
+        "\n",
+        "# ============================================================\n",
+        "# 8. PREPARE MULTIMODAL INPUTS\n",
+        "# ============================================================\n",
+        "\n",
+        "def prepare_multimodal_inputs(messages):\n",
+        "\n",
+        "    # Apply the chat template using the tokenizer directly.\n",
+        "    prompt = tokenizer.apply_chat_template(\n",
+        "        messages,\n",
+        "        tokenize=False,\n",
+        "        add_generation_prompt=True\n",
+        "    )\n",
+        "\n",
+        "    # Collect images in conversation order.\n",
+        "    images = []\n",
+        "\n",
+        "    for msg in messages:\n",
+        "\n",
+        "        content = msg.get(\"content\", [])\n",
+        "\n",
+        "        if isinstance(content, str):\n",
+        "            continue\n",
+        "\n",
+        "        if not isinstance(content, list):\n",
+        "            continue\n",
+        "\n",
+        "        for item in content:\n",
+        "\n",
+        "            if (\n",
+        "                isinstance(item, dict)\n",
+        "                and item.get(\"type\") == \"image\"\n",
+        "            ):\n",
+        "\n",
+        "                image = item.get(\"image\")\n",
+        "\n",
+        "                if image is not None:\n",
+        "                    images.append(image)\n",
+        "\n",
+        "    # Process text and images.\n",
+        "    if images:\n",
+        "\n",
+        "        formatted_inputs = processor(\n",
+        "            text=prompt,\n",
+        "            images=images,\n",
+        "            return_tensors=\"pt\"\n",
+        "        )\n",
+        "\n",
+        "    else:\n",
+        "\n",
+        "        formatted_inputs = processor(\n",
+        "            text=prompt,\n",
+        "            return_tensors=\"pt\"\n",
+        "        )\n",
+        "\n",
+        "    return formatted_inputs\n",
+        "\n",
+        "\n",
+        "# ============================================================\n",
+        "# 9. SPLIT THINKING AND FINAL ANSWER\n",
+        "# ============================================================\n",
+        "\n",
+        "def split_thinking_and_answer(raw_text):\n",
+        "\n",
+        "    \"\"\"\n",
+        "    Separates model-generated <think>...</think> content\n",
+        "    from the final answer.\n",
+        "\n",
+        "    If the model does not return a think block, its complete\n",
+        "    response is treated as the final answer.\n",
+        "    \"\"\"\n",
+        "\n",
+        "    raw_text = raw_text.strip()\n",
+        "\n",
+        "    # Remove common trailing generation tokens.\n",
+        "    raw_text = re.sub(\n",
+        "        r\"<\\|im_end\\|>|<\\|endoftext\\|>|</s>\",\n",
+        "        \"\",\n",
+        "        raw_text\n",
+        "    ).strip()\n",
+        "\n",
+        "    # Find a complete think block.\n",
+        "    match = re.search(\n",
+        "        r\"<think>(.*?)</think>\",\n",
+        "        raw_text,\n",
+        "        flags=re.DOTALL | re.IGNORECASE\n",
+        "    )\n",
+        "\n",
+        "    if match:\n",
+        "\n",
+        "        thinking = match.group(1).strip()\n",
+        "\n",
+        "        # Everything outside the think block is the answer.\n",
+        "        answer = (\n",
+        "            raw_text[:match.start()]\n",
+        "            + raw_text[match.end():]\n",
+        "        ).strip()\n",
+        "\n",
+        "    else:\n",
+        "\n",
+        "        # Handle an unclosed think block.\n",
+        "        open_match = re.search(\n",
+        "            r\"<think>\",\n",
+        "            raw_text,\n",
+        "            flags=re.IGNORECASE\n",
+        "        )\n",
+        "\n",
+        "        if open_match:\n",
+        "\n",
+        "            thinking = raw_text[open_match.end():].strip()\n",
+        "            answer = \"\"\n",
+        "\n",
+        "        else:\n",
+        "\n",
+        "            thinking = \"The model did not return a separate <think> block.\"\n",
+        "            answer = raw_text\n",
+        "\n",
+        "    # Remove any remaining template artifacts.\n",
+        "    answer = re.sub(\n",
+        "        r\"<\\|im_start\\|>|<\\|im_sep\\|>|<\\|im_end\\|>\",\n",
+        "        \"\",\n",
+        "        answer\n",
+        "    ).strip()\n",
+        "\n",
+        "    thinking = re.sub(\n",
+        "        r\"<\\|im_start\\|>|<\\|im_sep\\|>|<\\|im_end\\|>\",\n",
+        "        \"\",\n",
+        "        thinking\n",
+        "    ).strip()\n",
+        "\n",
+        "    if not answer:\n",
+        "        answer = \"(The model did not produce a separate final answer.)\"\n",
+        "\n",
+        "    if not thinking:\n",
+        "        thinking = \"(No thinking text was returned.)\"\n",
+        "\n",
+        "    return thinking, answer\n",
+        "\n",
+        "\n",
+        "# ============================================================\n",
+        "# 10. GENERATION FUNCTION\n",
+        "# ============================================================\n",
+        "\n",
+        "def generate_response(\n",
+        "    messages,\n",
+        "    max_new_tokens=1024,\n",
+        "    temperature=0.7,\n",
+        "    top_p=0.9\n",
+        "):\n",
+        "\n",
+        "    inputs = prepare_multimodal_inputs(messages)\n",
+        "\n",
+        "    # Move inputs to the model's embedding device.\n",
+        "    input_device = model.get_input_embeddings().weight.device\n",
+        "\n",
+        "    inputs = inputs.to(input_device)\n",
+        "\n",
+        "    input_length = inputs[\"input_ids\"].shape[-1]\n",
+        "\n",
+        "    do_sample = float(temperature) > 0\n",
+        "\n",
+        "    generation_kwargs = {\n",
+        "        \"max_new_tokens\": int(max_new_tokens),\n",
+        "        \"do_sample\": do_sample,\n",
+        "        \"repetition_penalty\": 1.05,\n",
+        "        \"pad_token_id\": tokenizer.pad_token_id\n",
+        "            if tokenizer.pad_token_id is not None\n",
+        "            else tokenizer.eos_token_id,\n",
+        "        \"eos_token_id\": tokenizer.eos_token_id,\n",
+        "        \"use_cache\": True\n",
+        "    }\n",
+        "\n",
+        "    if do_sample:\n",
+        "\n",
+        "        generation_kwargs.update({\n",
+        "            \"temperature\": float(temperature),\n",
+        "            \"top_p\": float(top_p)\n",
+        "        })\n",
+        "\n",
+        "    with torch.inference_mode():\n",
+        "\n",
+        "        output_ids = model.generate(\n",
+        "            **inputs,\n",
+        "            **generation_kwargs\n",
+        "        )\n",
+        "\n",
+        "    # Decode only newly generated tokens.\n",
+        "    generated_ids = output_ids[0][input_length:]\n",
+        "\n",
+        "    raw_response = tokenizer.decode(\n",
+        "        generated_ids,\n",
+        "        skip_special_tokens=False,\n",
+        "        clean_up_tokenization_spaces=False\n",
+        "    )\n",
+        "\n",
+        "    thinking, answer = split_thinking_and_answer(raw_response)\n",
+        "\n",
+        "    # Keep the original model response in the conversation history.\n",
+        "    return raw_response.strip(), thinking, answer\n",
+        "\n",
+        "\n",
+        "# ============================================================\n",
+        "# 11. GRADIO CHAT FUNCTION\n",
+        "# ============================================================\n",
+        "\n",
+        "def chat(\n",
+        "    message,\n",
+        "    image,\n",
+        "    chat_history,\n",
+        "    conversation_history,\n",
+        "    max_new_tokens,\n",
+        "    temperature,\n",
+        "    top_p\n",
+        "):\n",
+        "\n",
+        "    if chat_history is None:\n",
+        "        chat_history = []\n",
+        "\n",
+        "    if conversation_history is None:\n",
+        "        conversation_history = []\n",
+        "\n",
+        "    message = (message or \"\").strip()\n",
+        "\n",
+        "    if not message and image is None:\n",
+        "\n",
+        "        return (\n",
+        "            chat_history,\n",
+        "            conversation_history,\n",
+        "            \"\",\n",
+        "            None,\n",
+        "            \"\",\n",
+        "            \"Please enter a message or upload an image.\"\n",
+        "        )\n",
+        "\n",
+        "    # Build user content.\n",
+        "    user_content = []\n",
+        "\n",
+        "    if message:\n",
+        "\n",
+        "        user_content.append({\n",
+        "            \"type\": \"text\",\n",
+        "            \"text\": message\n",
+        "        })\n",
+        "\n",
+        "    if image is not None:\n",
+        "\n",
+        "        # Convert to RGB PIL image.\n",
+        "        if isinstance(image, Image.Image):\n",
+        "            image = image.convert(\"RGB\")\n",
+        "\n",
+        "        user_content.append({\n",
+        "            \"type\": \"image\",\n",
+        "            \"image\": image\n",
+        "        })\n",
+        "\n",
+        "    # Add user turn to actual model conversation.\n",
+        "    conversation_history = list(conversation_history)\n",
+        "\n",
+        "    conversation_history.append({\n",
+        "        \"role\": \"user\",\n",
+        "        \"content\": user_content\n",
+        "    })\n",
+        "\n",
+        "    # Display user message in the chat window.\n",
+        "    display_message = message if message else \"Please describe this image.\"\n",
+        "\n",
+        "    if image is not None:\n",
+        "        display_message += \"\\n\\n📷 *Image attached*\"\n",
+        "\n",
+        "    chat_history = list(chat_history)\n",
+        "\n",
+        "    chat_history.append({\n",
+        "        \"role\": \"user\",\n",
+        "        \"content\": display_message\n",
+        "    })\n",
+        "\n",
+        "    try:\n",
+        "\n",
+        "        raw_response, thinking, answer = generate_response(\n",
+        "            conversation_history,\n",
+        "            max_new_tokens=max_new_tokens,\n",
+        "            temperature=temperature,\n",
+        "            top_p=top_p\n",
+        "        )\n",
+        "\n",
+        "        # Store the complete response for subsequent turns.\n",
+        "        conversation_history.append({\n",
+        "            \"role\": \"assistant\",\n",
+        "            \"content\": raw_response\n",
+        "        })\n",
+        "\n",
+        "        # Add two separate assistant messages to the chat.\n",
+        "        chat_history.append({\n",
+        "            \"role\": \"assistant\",\n",
+        "            \"content\": \"### 💭 WHAT THE MODEL THINKS\\n\\n\" + thinking\n",
+        "        })\n",
+        "\n",
+        "        chat_history.append({\n",
+        "            \"role\": \"assistant\",\n",
+        "            \"content\": \"### 🗣️ WHAT THE MODEL SAYS\\n\\n\" + answer\n",
+        "        })\n",
+        "\n",
+        "        return (\n",
+        "            chat_history,\n",
+        "            conversation_history,\n",
+        "            \"\",\n",
+        "            None,\n",
+        "            thinking,\n",
+        "            answer\n",
+        "        )\n",
+        "\n",
+        "    except Exception as e:\n",
+        "\n",
+        "        error_message = f\"Generation error:\\n\\n{str(e)}\"\n",
+        "\n",
+        "        chat_history.append({\n",
+        "            \"role\": \"assistant\",\n",
+        "            \"content\": error_message\n",
+        "        })\n",
+        "\n",
+        "        return (\n",
+        "            chat_history,\n",
+        "            conversation_history,\n",
+        "            message,\n",
+        "            image,\n",
+        "            \"\",\n",
+        "            error_message\n",
+        "        )\n",
+        "\n",
+        "\n",
+        "# ============================================================\n",
+        "# 12. CLEAR CONVERSATION\n",
+        "# ============================================================\n",
+        "\n",
+        "def clear_chat():\n",
+        "\n",
+        "    return (\n",
+        "        [],\n",
+        "        [],\n",
+        "        \"\",\n",
+        "        None,\n",
+        "        \"\",\n",
+        "        \"\"\n",
+        "    )\n",
+        "\n",
+        "\n",
+        "# ============================================================\n",
+        "# 13. GRADIO INTERFACE\n",
+        "# ============================================================\n",
+        "\n",
+        "CUSTOM_CSS = \"\"\"\n",
+        "body {\n",
+        "    background: #181818 !important;\n",
+        "}\n",
+        "\n",
+        ".gradio-container {\n",
+        "    max-width: 1500px !important;\n",
+        "    margin: auto !important;\n",
+        "}\n",
+        "\n",
+        "#main-title {\n",
+        "    text-align: center;\n",
+        "    margin-bottom: 5px;\n",
+        "}\n",
+        "\n",
+        "#chat-window {\n",
+        "    min-height: 650px !important;\n",
+        "}\n",
+        "\n",
+        "#chat-window .message {\n",
+        "    font-size: 16px !important;\n",
+        "    line-height: 1.65 !important;\n",
+        "}\n",
+        "\n",
+        "#message-input textarea {\n",
+        "    font-size: 16px !important;\n",
+        "    line-height: 1.6 !important;\n",
+        "}\n",
+        "\n",
+        "#thinking-panel textarea,\n",
+        "#answer-panel textarea {\n",
+        "    font-size: 15px !important;\n",
+        "    line-height: 1.7 !important;\n",
+        "}\n",
+        "\n",
+        "#send-button {\n",
+        "    background: #a78bfa !important;\n",
+        "    color: #181818 !important;\n",
+        "    font-weight: bold !important;\n",
+        "}\n",
+        "\n",
+        "#clear-button {\n",
+        "    background: #383344 !important;\n",
+        "}\n",
+        "\"\"\"\n",
+        "\n",
+        "\n",
+        "with gr.Blocks(\n",
+        "    theme=gr.themes.Base(\n",
+        "        primary_hue=\"violet\",\n",
+        "        secondary_hue=\"orange\",\n",
+        "        neutral_hue=\"slate\"\n",
+        "    ),\n",
+        "    css=CUSTOM_CSS,\n",
+        "    title=\"Qwen3.5-4B Uncensored Chat\"\n",
+        ") as demo:\n",
+        "\n",
+        "    gr.Markdown(\n",
+        "        \"\"\"\n",
+        "        # Qwen3.5-4B Uncensored Chat\n",
+        "\n",
+        "        **DreamFast / HauhauCS — 4-bit NF4 — NVIDIA T4**\n",
+        "\n",
+        "        Model responses are separated into their generated thinking\n",
+        "        section and their final answer.\n",
+        "        \"\"\",\n",
+        "        elem_id=\"main-title\"\n",
+        "    )\n",
+        "\n",
+        "    # Actual model conversation and displayed chat are separate.\n",
+        "    conversation_state = gr.State([])\n",
+        "\n",
+        "    with gr.Row():\n",
+        "\n",
+        "        with gr.Column(scale=7):\n",
+        "\n",
+        "            chatbot = gr.Chatbot(\n",
+        "                label=\"Conversation\",\n",
+        "                height=650,\n",
+        "                autoscroll=True,\n",
+        "                elem_id=\"chat-window\"\n",
+        "            )\n",
+        "\n",
+        "            with gr.Row():\n",
+        "\n",
+        "                message_input = gr.Textbox(\n",
+        "                    label=\"Your message\",\n",
+        "                    placeholder=\"Write your message here...\",\n",
+        "                    lines=4,\n",
+        "                    max_lines=10,\n",
+        "                    scale=5,\n",
+        "                    elem_id=\"message-input\"\n",
+        "                )\n",
+        "\n",
+        "                image_input = gr.Image(\n",
+        "                    label=\"Optional image\",\n",
+        "                    type=\"pil\",\n",
+        "                    height=150,\n",
+        "                    sources=[\"upload\", \"clipboard\"],\n",
+        "                    scale=2\n",
+        "                )\n",
+        "\n",
+        "            with gr.Row():\n",
+        "\n",
+        "                send_button = gr.Button(\n",
+        "                    \"Send\",\n",
+        "                    variant=\"primary\",\n",
+        "                    elem_id=\"send-button\",\n",
+        "                    scale=3\n",
+        "                )\n",
+        "\n",
+        "                clear_button = gr.Button(\n",
+        "                    \"Clear conversation\",\n",
+        "                    variant=\"secondary\",\n",
+        "                    elem_id=\"clear-button\",\n",
+        "                    scale=1\n",
+        "                )\n",
+        "\n",
+        "        with gr.Column(scale=3):\n",
+        "\n",
+        "            gr.Markdown(\"## Generation settings\")\n",
+        "\n",
+        "            max_new_tokens = gr.Slider(\n",
+        "                minimum=128,\n",
+        "                maximum=4096,\n",
+        "                value=1024,\n",
+        "                step=128,\n",
+        "                label=\"Maximum new tokens\"\n",
+        "            )\n",
+        "\n",
+        "            temperature = gr.Slider(\n",
+        "                minimum=0.0,\n",
+        "                maximum=1.5,\n",
+        "                value=0.7,\n",
+        "                step=0.05,\n",
+        "                label=\"Temperature\"\n",
+        "            )\n",
+        "\n",
+        "            top_p = gr.Slider(\n",
+        "                minimum=0.1,\n",
+        "                maximum=1.0,\n",
+        "                value=0.9,\n",
+        "                step=0.05,\n",
+        "                label=\"Top-p\"\n",
+        "            )\n",
+        "\n",
+        "            gr.Markdown(\n",
+        "                \"\"\"\n",
+        "                **Tip:** Set temperature to `0` for deterministic\n",
+        "                generation. Increase maximum tokens for longer responses.\n",
+        "                \"\"\"\n",
+        "            )\n",
+        "\n",
+        "    gr.Markdown(\"---\")\n",
+        "\n",
+        "    gr.Markdown(\"## Latest model response\")\n",
+        "\n",
+        "    with gr.Row():\n",
+        "\n",
+        "        with gr.Column():\n",
+        "\n",
+        "            thinking_output = gr.Textbox(\n",
+        "                label=\"💭 What the model thinks\",\n",
+        "                lines=12,\n",
+        "                max_lines=25,\n",
+        "                interactive=False,\n",
+        "                placeholder=\"Generated thinking will appear here...\",\n",
+        "                elem_id=\"thinking-panel\"\n",
+        "            )\n",
+        "\n",
+        "        with gr.Column():\n",
+        "\n",
+        "            answer_output = gr.Textbox(\n",
+        "                label=\"🗣️ What the model says\",\n",
+        "                lines=12,\n",
+        "                max_lines=25,\n",
+        "                interactive=False,\n",
+        "                placeholder=\"The final answer will appear here...\",\n",
+        "                elem_id=\"answer-panel\"\n",
+        "            )\n",
+        "\n",
+        "    # ========================================================\n",
+        "    # EVENTS\n",
+        "    # ========================================================\n",
+        "\n",
+        "    inputs = [\n",
+        "        message_input,\n",
+        "        image_input,\n",
+        "        chatbot,\n",
+        "        conversation_state,\n",
+        "        max_new_tokens,\n",
+        "        temperature,\n",
+        "        top_p\n",
+        "    ]\n",
+        "\n",
+        "    outputs = [\n",
+        "        chatbot,\n",
+        "        conversation_state,\n",
+        "        message_input,\n",
+        "        image_input,\n",
+        "        thinking_output,\n",
+        "        answer_output\n",
+        "    ]\n",
+        "\n",
+        "    send_button.click(\n",
+        "        fn=chat,\n",
+        "        inputs=inputs,\n",
+        "        outputs=outputs\n",
+        "    )\n",
+        "\n",
+        "    message_input.submit(\n",
+        "        fn=chat,\n",
+        "        inputs=inputs,\n",
+        "        outputs=outputs\n",
+        "    )\n",
+        "\n",
+        "    clear_button.click(\n",
+        "        fn=clear_chat,\n",
+        "        inputs=[],\n",
+        "        outputs=outputs\n",
+        "    )\n",
+        "\n",
+        "\n",
+        "# ============================================================\n",
+        "# 14. LAUNCH\n",
+        "# ============================================================\n",
+        "\n",
+        "print(\"\\nStarting Gradio interface...\")\n",
+        "\n",
+        "demo.queue().launch(\n",
+        "    share=True,\n",
+        "    debug=True,\n",
+        "    show_error=True\n",
+        ")\n"
+      ]
+    }
+  ]
+}