{ "cells": [ { "cell_type": "markdown", "metadata": {}, "source": [ "#
🦙 Файнтюнинг больших языковых моделей 🎛
\n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "## 🧑‍🎓 В этом уроке: \n", "\n", "\n", "* 🎲 Разберемся, зачем и когда использовать файнтюнинг (fine-tuning)\n", "* [🤹‍♀️ Научимся готовить датасеты для файнтюнинга](#part2)\n", "* [🚀 Поймем, как осуществить дообучение с помощью фрэймворка [unsloth](https://github.com/unslothai/unsloth)](#part3)\n", "* [📦👩‍💻 Освоим техники эффективного фантюна (**Peft**, **Lora**, **QLora** и прочие неприличные аббревиатуры)](#part4)\n", "* [📦 Произведем инференс и правильное сохранение дообученной модели ](#part4)\n", "* 🥊 [Разберём файнтюнинг на примере 2-х датасетов:](#part6)\n", " * Соберать свой\n", " * или взять готовый датасет с HF [статья на Хабр](https://habr.com/ru/articles/832984/)\n", "* [🧸 Выводы и заключения ✅](#part6)\n", "\n", " " ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "
\n", "\n", "**🚀 Дообучать проще, чем кажется!`**\n", " \n", "\n", "\n", "* Чтобы запустить процесс файнтюнинга на ресурсах Google Colab и не ждать вечность.
\n", "* Воспользуемся специальным фрэймворком [Unsloth](https://github.com/unslothai/unsloth), который позволяет ускорить процесс файнтюнинга от 2-х до 5 раз на потребительских видеокартах (и даже на CPU), и требует меньше памяти.\n", "* На странице проекта есть готовые удобные ноутбуки c разбором файнтюнинга самых популярных моделей - на базе одного из них мы сделали этот ноутбук. \n", "* Так же фреймворк позволяет запускать файнтюн без глубоких знаний нейросетей и ML." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "# Установим unsloth и зависимости." ] }, { "cell_type": "code", "execution_count": 1, "metadata": { "id": "2eSvM9zX_2d3" }, "outputs": [], "source": [ "%%capture\n", "# установка Unsloth, Xformers (Flash Attention) и все другие пакеты!\n", "!pip install \"unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git\"\n", "\n", "#Нам нужно проверить какая версия Torch для Xformers (2.3 -> 0.0.27)\n", "from torch import __version__; from packaging.version import Version as V\n", "xformers = \"xformers==0.0.27\" if V(__version__) < V(\"2.4.0\") else \"xformers\"\n", "!pip install --no-deps {xformers} trl peft accelerate bitsandbytes triton" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "
\n", " \n", "Импортируем библиотеки и модель, которую будем файнтюнить - `Llama-3.1-8B`.
\n", "В логах увидим, что **Unsloth** пропатчил нашу систему для двукратного ускорения процесса." ] }, { "cell_type": "code", "execution_count": 1, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "QmUBVEnvCDJv", "outputId": "9d79f0f1-c3c3-44f2-df15-2f15a5e77216" }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "🦥 Unsloth: Will patch your computer to enable 2x faster free finetuning.\n", "==((====))== Unsloth 2024.8: Fast Llama patching. Transformers = 4.44.2.\n", " \\\\ /| GPU: Tesla T4. Max memory: 14.748 GB. Platform = Linux.\n", "O^O/ \\_/ \\ Pytorch: 2.4.0+cu121. CUDA = 7.5. CUDA Toolkit = 12.1.\n", "\\ / Bfloat16 = FALSE. FA [Xformers = 0.0.27.post2. FA2 = False]\n", " \"-____-\" Free Apache license: http://github.com/unslothai/unsloth\n", "Unsloth: Fast downloading is enabled - ignore downloading bars which are red colored!\n" ] } ], "source": [ "from unsloth import FastLanguageModel\n", "import torch\n", "\n", "max_seq_length = 2048 # Choose any! We auto support RoPE Scaling internally!\n", "dtype = None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+\n", "load_in_4bit = True # Use 4bit quantization to reduce memory usage. Can be False.\n", "\n", "model, tokenizer = FastLanguageModel.from_pretrained(\n", " model_name = \"unsloth/Meta-Llama-3.1-8B\",\n", " max_seq_length = max_seq_length,\n", " dtype = dtype,\n", " load_in_4bit = load_in_4bit, # Применяем QLoRA\n", ")" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "
\n", " \n", "Используемые параметры:\n", "* `model_name` - имя модели для файнтюна (смотрим доступные [здесь](https://github.com/unslothai/unsloth))\n", "* `max_seq_length` - максимальная длина последовательности в токенах.\n", "* `dtype` - Тип данных для хранения весов (зависит от модели видеокарты). Для атоматического определения ставить `None`.\n", "* `load_in_4bit` - ключевой параметр для подгрузки 4-х битной версии модели. " ] }, { "cell_type": "markdown", "metadata": { "id": "SXd9bTZd1aaL" }, "source": [ "
\n", " \n", "🔥 Чтобы не переучивать полностью всю модель - будем использовать **PEFT** с **LoRA**. Так мы будем обновлять от 1 до 10% от всех параметров!" ] }, { "cell_type": "code", "execution_count": 2, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "6bZsfBuZDeCL", "outputId": "2872e6db-5237-4658-9e3f-4b991a13a9d9" }, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "Unsloth 2024.8 patched 32 layers with 32 QKV layers, 32 O layers and 32 MLP layers.\n" ] } ], "source": [ "model = FastLanguageModel.get_peft_model(\n", " model,\n", " r = 16, # Choose any number > 0 ! Suggested 8, 16, 32, 64, 128\n", " target_modules = [\"q_proj\", \"k_proj\", \"v_proj\", \"o_proj\",\n", " \"gate_proj\", \"up_proj\", \"down_proj\",],\n", " lora_alpha = 16,\n", " lora_dropout = 0, # Supports any, but = 0 is optimized\n", " bias = \"none\", # Supports any, but = \"none\" is optimized\n", " # [NEW] \"unsloth\" uses 30% less VRAM, fits 2x larger batch sizes!\n", " use_gradient_checkpointing = \"unsloth\", # True or \"unsloth\" for very long context\n", " random_state = 3407,\n", " use_rslora = False, # We support rank stabilized LoRA\n", " loftq_config = None, # And LoftQ\n", ")" ] }, { "cell_type": "markdown", "metadata": { "id": "SXd9bTZd1aaL" }, "source": [ "
\n", "\n", "Для 90% случаев, вам подойдут значения параметров по умолчанию, можно экспериментировать только с параметром `r`. Полное описание всех параметров можно найти [здесь](https://docs.unsloth.ai/basics/lora-parameters-encyclopedia).\n", "\n", "Важные параметры LoRA:\n", "* `r` (ранг матрицы) - самый важный параметр, определяет размер дополнительной матрицы (адаптера), которую мы будем обучать в процессе файнтюнинга (рекомендуется выбирать кратно степеням двойки - 8, 16, 32 и.т.д). Так же от этого парамаетра зависит время которое будет затрачено на файнтюн, чем больше r, тем больше время.\n", "* `target_modules` - целевые модули, веса которых будут меняться при обучении \n" ] }, { "cell_type": "markdown", "metadata": { "id": "vITh0KVJ10qX" }, "source": [ "#
🤹‍♀️ Подготовка датасета\n", "\n", "Для файнтюна Ламы, требуется перевести каждую запись из нашего датасета в формат `Alpaca prompt`.\n", "\n", "И обязательно в конце промпта добавляем токен конца генерации (`EOS-token`), чтобы избежать бесконечной генерации." ] }, { "cell_type": "code", "execution_count": 3, "metadata": { "id": "LjY75GoYUCB8" }, "outputs": [], "source": [ "alpaca_prompt = \"\"\"Below is an instruction that describes a task, paired with an input that provides further context. \n", "Write a response that appropriately completes the request.\n", "\n", "### Instruction:\n", "{}\n", "\n", "### Input:\n", "{}\n", "\n", "### Response:\n", "{}\"\"\"\n", "\n", "EOS_TOKEN = tokenizer.eos_token # Must add EOS_TOKEN\n", "\n", "# Функция преобразования полей датасета в alpaca prompt\n", "def formatting_prompts_func(examples):\n", " instructions = examples[\"Instruction\"]\n", " inputs = examples[\"Input\"]\n", " outputs = examples[\"Response\"]\n", " texts = []\n", " for instruction, input, output in zip(instructions, inputs, outputs):\n", " # Must add EOS_TOKEN, otherwise your generation will go on forever!\n", " text = alpaca_prompt.format(instruction, input, output) + EOS_TOKEN\n", " texts.append(text)\n", " return { \"text\" : texts, }\n" ] }, { "cell_type": "code", "execution_count": 4, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "1Q_jm9q_TYGT", "outputId": "8108be31-7102-4cf1-85b8-2c1d60c126fd" }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "{'Input': 'Запуск канала о Data Science',\n", " 'Instruction': 'Write a post on the following topic',\n", " 'Response': 'Всем привет!\\n'\n", " '\\n'\n", " 'Решил запустить свой канал. Буду рассказывать здесь про свой '\n", " 'опыт в Data Science и лайфхаки',\n", " 'text': 'Below is an instruction that describes a task, paired with an input '\n", " 'that provides further context. Write a response that appropriately '\n", " 'completes the request.\\n'\n", " '\\n'\n", " '### Instruction:\\n'\n", " 'Write a post on the following topic\\n'\n", " '\\n'\n", " '### Input:\\n'\n", " 'Запуск канала о Data Science\\n'\n", " '\\n'\n", " '### Response:\\n'\n", " 'Всем привет!\\n'\n", " '\\n'\n", " 'Решил запустить свой канал. Буду рассказывать здесь про свой опыт в '\n", " 'Data Science и лайфхаки<|end_of_text|>'}\n" ] } ], "source": [ "from pprint import pprint\n", "from datasets import load_dataset\n", "\n", "# Скачиваем подготовленный датасет с HuggingFace \n", "dataset = load_dataset(\"Ivanich/datafeeling_posts\", split = \"train\")\n", "\n", "# Преобразуем в alpaca_prompt с помощью нашей функции и метода map.\n", "dataset = dataset.map(formatting_prompts_func, batched = True,)\n", "\n", "pprint(dataset[0])" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "
\n", " \n", "Видим, что готовый промпт добавился в колонку `text`." ] }, { "cell_type": "markdown", "metadata": { "id": "mDu8fiZZ63Sb" }, "source": [ "## Перед тем как ворваться в файнтюнинг\n", "Посмотрим, как модель справляется с задачей до файнтюна?" ] }, { "cell_type": "code", "execution_count": 5, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "9gqTLCuk5pmp", "outputId": "ba37d0b0-0a5c-4e53-c5d5-52cfd3b461f5" }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "['<|begin_of_text|>Below is an instruction that describes a task, paired with '\n", " 'an input that provides further context. Write a response that appropriately '\n", " 'completes the request.\\n'\n", " '\\n'\n", " '### Instruction:\\n'\n", " 'Write post about the following topic: \\n'\n", " '\\n'\n", " '### Input:\\n'\n", " 'Как запустить gen ai стартап\\n'\n", " '\\n'\n", " '### Response:\\n'\n", " '1. Запустить генетический алгоритм, который генерирует идеи стартапов.\\n'\n", " '2. Выбрать идеи, которые кажутся вам наиболее перспективными.\\n'\n", " '3. Выполнить анализ рынка и конкурентов для этих идей.\\n'\n", " '4. Выполнить маркетинговый анализ и анализ потребительского поведения.\\n'\n", " '5. Выполнить анализ бизнес-моделей и стратегий.\\n'\n", " '6. Выполнить анализ конкурентоспособности и рисков.\\n'\n", " '7. Выполнить анализ финансового состояния и возможностей.\\n'\n", " '8. Вып']\n" ] } ], "source": [ "FastLanguageModel.for_inference(model)\n", "inputs = tokenizer(\n", "[\n", " alpaca_prompt.format(\n", " \"Write post about the following topic: \", # instruction\n", " \"Как запустить gen ai стартап\", # input\n", " \"\", # output - leave this blank for generation!\n", " )\n", "], return_tensors = \"pt\").to(\"cuda\")\n", "\n", "outputs = model.generate(**inputs, max_new_tokens = 128, use_cache = True)\n", "pprint(tokenizer.batch_decode(outputs))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "
\n", " \n", "Видим, что модель выдаёт какой-то план с пунктами, и это не очень похоже на пост для канала." ] }, { "cell_type": "markdown", "metadata": { "id": "idAEIeSQ3xdS" }, "source": [ "#
🚀 Запуск файнтюна / Дообучение модели \n", "Теперь давайте воспользуемся `SFTTrainer` - supervised fine-tuning (обучение без учителя) от `Huggingface TRL` . Так нам не придется писать свою reward модель, которой необходимо оценивать ответ Llama!
\n", "Дополнительную документацию можно найти здесь: [Документация TRL SFT](https://huggingface.co/docs/trl/sft_trainer).
\n", "Также `unsloth` поддерживает [`DPOTrainer`](https://huggingface.co/docs/trl/dpo_trainer) от TRL!\n", "\n", "Мы делаем 60 шагов, чтобы ускорить процесс, но вы можете установить `num_train_epochs=1` для полного запуска и отключить `max_steps=None`. \n" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "##
Крутая настройка для самых продвинутых | Wandb\n", "\n", "При желании можно подключить библиотеку [Weights&Biases](https://wandb.ai/site) - бесплатный трекер экспериментов. И файнтюннить по взрослому." ] }, { "cell_type": "code", "execution_count": 7, "metadata": {}, "outputs": [], "source": [ "# import wandb\n", "\n", "# wb_token = 'WANDB_API_KEY' # ключ c сайта Weights & Biases https://wandb.ai/site\n", "# wandb.login(key=wb_token)\n", "# далее раскомментируйте параметр record_to" ] }, { "cell_type": "code", "execution_count": 11, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "95_Nn-89DhsL", "outputId": "476f105b-b374-4861-eec5-88b0f3c859a6" }, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "max_steps is given, it will override any value given in num_train_epochs\n" ] } ], "source": [ "from trl import SFTTrainer\n", "from transformers import TrainingArguments\n", "from unsloth import is_bfloat16_supported\n", "\n", "trainer = SFTTrainer(\n", " model = model,\n", " tokenizer = tokenizer,\n", " train_dataset = dataset,\n", " dataset_text_field = \"text\",\n", " max_seq_length = max_seq_length,\n", " dataset_num_proc = 2,\n", " packing = False, # Can make training 5x faster for short sequences.\n", " args = TrainingArguments(\n", " per_device_train_batch_size = 4,\n", " gradient_accumulation_steps = 2,\n", " warmup_steps = 10,\n", " #num_train_epochs = 1, # Set this for 1 full training run.\n", " max_steps = 60,\n", " learning_rate = 2e-4,\n", " fp16 = not is_bfloat16_supported(),\n", " bf16 = is_bfloat16_supported(),\n", " logging_steps = 1,\n", " optim = \"adamw_8bit\",\n", " weight_decay = 0.01,\n", " lr_scheduler_type = \"linear\",\n", " seed = 3407,\n", " output_dir = \"outputs\",\n", " # report_to=\"wandb\", # Если используете Weights & Biases\n", " ),\n", ")" ] }, { "cell_type": "code", "execution_count": 14, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "2ejIt2xSNKKp", "outputId": "c4b07c62-3201-4a90-f024-560a86785d23" }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "GPU = Tesla T4. Max memory = 14.748 GB.\n", "12.742 GB of memory reserved.\n" ] } ], "source": [ "#@title Show current memory stats\n", "gpu_stats = torch.cuda.get_device_properties(0)\n", "start_gpu_memory = round(torch.cuda.max_memory_reserved() / 1024 / 1024 / 1024, 3)\n", "max_memory = round(gpu_stats.total_memory / 1024 / 1024 / 1024, 3)\n", "\n", "print(f\"GPU = {gpu_stats.name}. Max memory = {max_memory} GB.\")\n", "print(f\"{start_gpu_memory} GB of memory reserved.\")" ] }, { "cell_type": "code", "execution_count": 12, "metadata": { "colab": { "base_uri": "https://localhost:8080/", "height": 1000 }, "id": "yqxqAZ7KJ4oL", "outputId": "bbe9d498-00da-414d-99ad-14388acf1014" }, "outputs": [ { "name": "stderr", "output_type": "stream", "text": [ "==((====))== Unsloth - 2x faster free finetuning | Num GPUs = 1\n", " \\\\ /| Num examples = 587 | Num Epochs = 4\n", "O^O/ \\_/ \\ Batch size per device = 8 | Gradient Accumulation steps = 4\n", "\\ / Total batch size = 32 | Total steps = 60\n", " \"-____-\" Number of trainable parameters = 41,943,040\n" ] }, { "data": { "text/html": [ "\n", "
\n", " \n", " \n", " [60/60 43:35, Epoch 3/4]\n", "
\n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", " \n", "
StepTraining Loss
12.236400
22.247700
32.328900
42.165200
52.294000
62.186900
72.269100
82.214700
92.289900
102.298600
112.244600
122.201800
132.153100
142.188300
152.166900
162.182000
172.188200
182.098000
192.084500
202.096200
212.201400
222.091800
232.053300
242.117500
252.018800
262.011700
271.953300
281.964100
292.048200
301.907600
312.030600
322.014700
331.982400
341.982400
351.898200
362.025500
371.779900
382.007600
391.968800
401.961200
411.911900
421.888400
431.826900
441.803100
451.816400
461.819600
471.691800
481.928200
491.881500
501.897900
511.940600
521.860200
531.859400
541.853800
551.901000
561.895100
571.795600
581.882400
591.795100
601.654000

" ], "text/plain": [ "" ] }, "metadata": {}, "output_type": "display_data" } ], "source": [ "# Запускаем тренировку!\n", "trainer_stats = trainer.train()" ] }, { "cell_type": "code", "execution_count": 15, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "pCqnaKmlO1U9", "outputId": "7e351717-6f83-4794-dff1-633468bc3b3f" }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "2669.216 seconds used for training.\n", "44.49 minutes used for training.\n", "Peak reserved memory = 12.742 GB.\n", "Peak reserved memory for training = 0.0 GB.\n", "Peak reserved memory % of max memory = 86.398 %.\n", "Peak reserved memory for training % of max memory = 0.0 %.\n" ] } ], "source": [ "#@title Show final memory and time stats\n", "used_memory = round(torch.cuda.max_memory_reserved() / 1024 / 1024 / 1024, 3)\n", "used_memory_for_lora = round(used_memory - start_gpu_memory, 3)\n", "used_percentage = round(used_memory /max_memory*100, 3)\n", "lora_percentage = round(used_memory_for_lora/max_memory*100, 3)\n", "print(f\"{trainer_stats.metrics['train_runtime']} seconds used for training.\")\n", "print(f\"{round(trainer_stats.metrics['train_runtime']/60, 2)} minutes used for training.\")\n", "print(f\"Peak reserved memory = {used_memory} GB.\")\n", "print(f\"Peak reserved memory for training = {used_memory_for_lora} GB.\")\n", "print(f\"Peak reserved memory % of max memory = {used_percentage} %.\")\n", "print(f\"Peak reserved memory for training % of max memory = {lora_percentage} %.\")" ] }, { "cell_type": "markdown", "metadata": { "id": "ekOmTR1hSNcr" }, "source": [ "#

Инференс 💬\n", "Давайте запустим модель! Можете изменить инструкцию и ввод, Response оставим пустым!\n", "\n", "Попробовать вдвое ускорить инференс можно в Colab для **Llama-3.1 8b Instruct** [здесь](https://colab.research.google.com/drive/1T-YBVfnphoVc8E2E854qF3jdia2Ll2W2?usp=sharing)" ] }, { "cell_type": "code", "execution_count": 17, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "kR3gIAX-SM2q", "outputId": "ec9e915d-ddd7-4249-d135-1676b26f6c69", "scrolled": true }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "['<|begin_of_text|>Below is an instruction that describes a task, paired with '\n", " 'an input that provides further context. Write a response that appropriately '\n", " 'completes the request.\\n'\n", " '\\n'\n", " '### Instruction:\\n'\n", " 'Write post about the following topic: \\n'\n", " '\\n'\n", " '### Input:\\n'\n", " 'как запустить gen ai стартап\\n'\n", " '\\n'\n", " '### Response:\\n'\n", " '🎤 Сегодня у меня был интересный опыт. \\n'\n", " '\\n'\n", " '🔝 Встретился с 2-мя ребятами, которые хотят запустить свой gen ai стартап. \\n'\n", " '\\n'\n", " '🏆 В целом, все было классно. У них крутой опыт, хороший тимбилдинг и '\n", " 'интересная задумка. \\n'\n", " '\\n'\n", " '🤔 Однако, я все же посоветовал им не запускать стартап, а просто скинуть '\n", " 'свою идею в нашу ML лотерею. \\n'\n", " '\\n'\n", " '🤔 Понятно, что это не совсем']\n" ] } ], "source": [ "# alpaca_prompt = Copied from above\n", "FastLanguageModel.for_inference(model) # Enable native 2x faster inference\n", "inputs = tokenizer(\n", "[\n", " alpaca_prompt.format(\n", " \"Write post about the following topic: \", # instruction\n", " \"как запустить gen ai стартап\", # input\n", " \"\", # output - leave this blank for generation!\n", " )\n", "], return_tensors = \"pt\").to(\"cuda\")\n", "\n", "outputs = model.generate(**inputs, max_new_tokens = 128, use_cache = True)\n", "pprint(tokenizer.batch_decode(outputs))" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "
\n", " \n", "✅ После файнтюна модель по тому же запросу пишет пост, и явно угадывается стиль постов канала [Datafeeling](https://t.me/datafeeling)." ] }, { "cell_type": "markdown", "metadata": { "id": "CrSvZObor0lY" }, "source": [ " **Запуск генерации в режиме стриминга:** Вы также можете использовать `TextStreamer` для непрерывного инференса, чтобы вы могли видеть генерацию токен за токеном!" ] }, { "cell_type": "code", "execution_count": 18, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "e2pEuRb1r2Vg", "outputId": "b289b319-ddff-430b-ff0e-d29a403ae59e" }, "outputs": [ { "name": "stdout", "output_type": "stream", "text": [ "<|begin_of_text|>Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.\n", "\n", "### Instruction:\n", "Write post about the following topic: \n", "\n", "### Input:\n", "Как стать kaggle master\n", "\n", "### Response:\n", "🏆 Как стать kaggle master\n", "\n", "🤔 На Kaggle сейчас много мэтчей, в которых можно скинуть лидерборд. И это прекрасно, но вот только не все так просто. \n", "\n", "🤫 Если ты хочешь добиться успеха, то придется много потрудиться. В противном случае ты просто не добьешься успеха. И это нормально, не судите. \n", "\n", "🤔 Поэтому, если ты хочешь добиться успеха, то придется много потрудиться. И это нормально, не судите. \n", "\n", "�\n" ] } ], "source": [ "from transformers import TextStreamer\n", "\n", "FastLanguageModel.for_inference(model) # Enable native 2x faster inference\n", "inputs = tokenizer(\n", "[\n", " alpaca_prompt.format(\n", " \"Write post about the following topic: \", # instruction\n", " \"Как стать kaggle master\", # input\n", " \"\", # output - leave this blank for generation!\n", " )\n", "], return_tensors = \"pt\").to(\"cuda\")\n", "\n", "\n", "text_streamer = TextStreamer(tokenizer)\n", "_ = model.generate(**inputs, streamer = text_streamer, max_new_tokens = 128)" ] }, { "cell_type": "markdown", "metadata": { "id": "uMuVrWbjAzhc" }, "source": [ "###
Сохранение, загрузка файнтюненных моделей\n", "Чтобы сохранить окончательную модель в качестве адаптеров `LoRA`, используйте:\n", "* либо `push_to_hub` от **Huggingface** для онлайн-сохранения \n", "* либо `save_pretrained` для локального сохранения\n", "\n", "**Важно:** сохраняется ТОЛЬКО адаптер `LoRA`, а не полная модель. Как сохранить в 16-битном формате или `GGUF` смотрите в [этом ноутбуке](https://colab.research.google.com/drive/1Ys44kVvmeZtnICzWz0xgpRnrIOjZAuxp?usp=sharing)!" ] }, { "cell_type": "code", "execution_count": 4, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "3Pt-g2YquOa8", "outputId": "5ff05c43-7bb8-4c34-b3c9-35102ba254f5" }, "outputs": [], "source": [ "from getpass import getpass\n", "\n", "hf_token = getpass(prompt=\"Введите ваш HuggingFaceHub API ключ\")\n", "hf_username = 'Ivanich' # Вводим свой username на HugginFace" ] }, { "cell_type": "code", "execution_count": 23, "metadata": { "colab": { "base_uri": "https://localhost:8080/", "height": 130, "referenced_widgets": [ "497302a6acc84773a84ff23cd36da214", "b62cf033432f40e0a9ee259515167b51", "22eed99a28c74a80a42f95cc2529913b", "1f07aa6d2b744092ac1b2f13352b3909", "7a7468b6571c4b818f706b13e37345b5", "d2117400d4744ac498d1582a3336f905", "36e5a40703714c76ba41451af838a1fe", "0374d3f515e640ea80e11cb8cffdbf7d", "e08504a1906e41bf8cfe2e7d231e0f2f", "1417bf853c6241038d6d01fae7183091", "9d78a73117ff4b59b6f6499b7f978d14", "297389dcbf3040f19fb6773f776f3f2e", "1a8353acfc6e4a77b5e754ec3a9d164e", "a6784a8092dd4652bb19a142b05986b1", "74d5c798f06346e9bd1cf96bf72669ff", "d937f93cbe12423d9198d740769eaaad", "3d7de1da6dc5424bb29a8510b1cfc549", "d9f595f07ea14377ad4f2af81b3dc236", "3267638f39f34833955764925bc59ad9", "7a23197e84e449a6b77b893f4bcc81fb", "6747454af34a467690d783bae764fbfa", "d6ef70581b8841d4b7455f1074373478", "177ec807de33414281f03b5790021dc8", "a1242cc96bf9450abbbc00593bc7c8a8", "906abe2734de4a7bbceb71fefb265ae8", "de7e5ca6727c4c0d99522d35981a4ad2", "7ca294a7b4aa40668f8ea154f744a4b4", "d27389522e9c437695f0e93a0ecf1f09", "16b0f13d883d48809d4703704f43204c", "2c90127799134934b6eb68325838a8a8", "c0a6e820a2eb41dc9c6244f614272ef5", "a4a6ed211ca44d509cb82311c27caba3", "c7a2e8fad27e4fa2b9c80a970002d6bd" ] }, "id": "upcOlWe7A1vc", "outputId": "64f23760-f101-4ef6-e7f8-384eec50db9d" }, "outputs": [ { "data": { "application/vnd.jupyter.widget-view+json": { "model_id": "497302a6acc84773a84ff23cd36da214", "version_major": 2, "version_minor": 0 }, "text/plain": [ "README.md: 0%| | 0.00/588 [00:00\n", " \n", "Если перейти по ссылке, которую выдает HF после загрузки на хаб - увидим, что сохранилась не вся модель, а только веса адаптера LoRA (168 Mb). " ] }, { "cell_type": "markdown", "metadata": { "id": "AEEcJ4qfC7Lp" }, "source": [ "Now if you want to load the LoRA adapters we just saved for inference, set `False` to `True`:" ] }, { "cell_type": "code", "execution_count": null, "metadata": { "colab": { "base_uri": "https://localhost:8080/" }, "id": "MKX_XKs_BNZR", "outputId": "7431de93-bc80-48b7-9491-1f3ce12615fd" }, "outputs": [ { "data": { "text/plain": [ "['<|begin_of_text|>Below is an instruction that describes a task, paired with an input that provides further context. Write a response that appropriately completes the request.\\n\\n### Instruction:\\nWrite post about the following topic\\n\\n### Input:\\nРаспределение долей в стартапе\\n\\n### Response:\\nЯ думаю, что это очень важный вопрос: как распределять доли в стартапе? И, что еще важнее, как распределять доли в стартапе, когда вы еще не знаете, каким будет стартап? Очень много людей в стартапах, которые делают это по-старому, то есть, распределяют доли по статусу: founder, cofounder, employee, contractor, investor, etc. Это не так плохо, но, с другой стороны, я не знаю ни одного стартапа, который бы был успеш']" ] }, "execution_count": 12, "metadata": {}, "output_type": "execute_result" } ], "source": [ "if False:\n", " from unsloth import FastLanguageModel\n", " model, tokenizer = FastLanguageModel.from_pretrained(\n", " model_name = \"datafeeling_model\", # YOUR MODEL YOU USED FOR TRAINING\n", " max_seq_length = max_seq_length,\n", " dtype = dtype,\n", " load_in_4bit = load_in_4bit,\n", " )\n", " FastLanguageModel.for_inference(model) # Enable native 2x faster inference\n", "\n", "# alpaca_prompt = You MUST copy from above!\n", "\n", "inputs = tokenizer(\n", "[\n", " alpaca_prompt.format(\n", " \"Write post about the following topic\", # instruction\n", " \"Распределение долей в стартапе\", # input\n", " \"\", # output - leave this blank for generation!\n", " )\n", "], return_tensors = \"pt\").to(\"cuda\")\n", "\n", "outputs = model.generate(**inputs, max_new_tokens = 128, use_cache = True)\n", "tokenizer.batch_decode(outputs)" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "✅ Отлично, видим, что результат получился похожим на правду" ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "#
А что ещё из полезного?\n", "\n", "
\n", " \n", "* [RAFT](https://arxiv.org/abs/2403.10131) - RAG + Fine Tuning - файнтюнят модель, чтобы отбирать документы для RAG\n", "* [Finetuning ChatGPT](https://platform.openai.com/docs/guides/fine-tuning) - OpenAI и Anthropic объявили, что можно будет файнтюнить их модели на своих данных, а потом использовать эти версии моделей по API (платно).\n", "* [Датасет](https://huggingface.co/datasets/Ivanich/datafeeling_posts) с постами канала Datafeeling, собранный в этом [ноутбуке](https://github.com/a-milenkin/LLM_practical_course/blob/main/notebooks/M5_2_Dataset_prepare.ipynb).\n", "* Пример файнтюнинга на готовом датасете с HuggingFace c медицинскими данными. Оформили в виде [статьи на Хабр](https://habr.com/ru/articles/832984/)." ] }, { "cell_type": "markdown", "metadata": {}, "source": [ "#
🧸 Выводы и заключения ✅\n", "\n", "**Плюсы FT:**\n", "* Более консистентные результаты генрации (например, можно передать стилистику автора)\n", "* Уменьшение галлюцинаций (следование новым доменным знаниями)\n", "* Можно натренировать под специфический юзкейс\n", "* Не забывает данные (хранятся в весах модели)\n", "* Можно передать больше данных, чем в RAG (т.к. нет ограничений контекстного окна)\n", "* Можно использовать небольшие модели - меньше затрат на дальнейшее использование\n", "* Приватность - данные не передаются открыто в промпте, а \"зашиты\" в весах (меньше возможностей для промпт-хакинга)\n", "* Выше скорость инференса по сравнению с RAG (т.к. не нужно делать семантический поиск по базе и короче промпт, быстрее начинается генерация)\n", "\n", "**Минусы FT:**\n", "* Нужно собрать гораздо больше данных (самая трудоемкая и важная часть)\n", "* Повышенный порог входа - нужны более глубокие технические знания\n", "* Нужно больше ресурсов и времени на старте\n", "* Необходимость переучивать при выходе новых версий модели\n", "* Иногда все равно придется использовать RAG для доступа к новой часто изменяющейся информации" ] }, { "cell_type": "code", "execution_count": null, "metadata": {}, "outputs": [], "source": [] } ], "metadata": { "accelerator": "GPU", "colab": { "gpuType": "T4", "provenance": [] }, "kernelspec": { "display_name": "cv", "language": "python", "name": "python3" }, "language_info": { "codemirror_mode": { "name": "ipython", "version": 3 }, "file_extension": ".py", "mimetype": "text/x-python", "name": "python", "nbconvert_exporter": "python", "pygments_lexer": "ipython3", "version": "3.10.12" }, "widgets": { "application/vnd.jupyter.widget-state+json": { "0374d3f515e640ea80e11cb8cffdbf7d": { "model_module": "@jupyter-widgets/base", "model_module_version": "1.2.0", "model_name": "LayoutModel", "state": { "_model_module": "@jupyter-widgets/base", "_model_module_version": "1.2.0", "_model_name": "LayoutModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "LayoutView", "align_content": null, "align_items": null, "align_self": null, "border": null, "bottom": null, "display": null, "flex": null, "flex_flow": null, "grid_area": null, "grid_auto_columns": null, "grid_auto_flow": null, "grid_auto_rows": null, "grid_column": null, "grid_gap": null, "grid_row": null, "grid_template_areas": null, "grid_template_columns": null, "grid_template_rows": null, "height": null, "justify_content": null, "justify_items": null, "left": null, "margin": null, "max_height": null, "max_width": null, "min_height": null, "min_width": null, "object_fit": null, "object_position": null, "order": null, "overflow": null, "overflow_x": null, "overflow_y": null, "padding": null, "right": null, "top": null, "visibility": null, "width": null } }, "1417bf853c6241038d6d01fae7183091": { "model_module": "@jupyter-widgets/base", "model_module_version": "1.2.0", "model_name": "LayoutModel", "state": { "_model_module": "@jupyter-widgets/base", "_model_module_version": "1.2.0", "_model_name": "LayoutModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "LayoutView", "align_content": null, "align_items": null, "align_self": null, "border": null, "bottom": null, "display": null, "flex": null, "flex_flow": null, "grid_area": null, "grid_auto_columns": null, "grid_auto_flow": null, "grid_auto_rows": null, "grid_column": null, "grid_gap": null, "grid_row": null, "grid_template_areas": null, "grid_template_columns": null, "grid_template_rows": null, "height": null, "justify_content": null, "justify_items": null, "left": null, "margin": null, "max_height": null, "max_width": null, "min_height": null, "min_width": null, "object_fit": null, "object_position": null, "order": null, "overflow": null, "overflow_x": null, "overflow_y": null, "padding": null, "right": null, "top": null, "visibility": null, "width": null } }, "16b0f13d883d48809d4703704f43204c": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "DescriptionStyleModel", "state": { "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "DescriptionStyleModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "StyleView", "description_width": "" } }, "177ec807de33414281f03b5790021dc8": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "HBoxModel", "state": { "_dom_classes": [], "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "HBoxModel", "_view_count": null, "_view_module": "@jupyter-widgets/controls", "_view_module_version": "1.5.0", "_view_name": "HBoxView", "box_style": "", "children": [ "IPY_MODEL_a1242cc96bf9450abbbc00593bc7c8a8", "IPY_MODEL_906abe2734de4a7bbceb71fefb265ae8", "IPY_MODEL_de7e5ca6727c4c0d99522d35981a4ad2" ], "layout": "IPY_MODEL_7ca294a7b4aa40668f8ea154f744a4b4" } }, "1a8353acfc6e4a77b5e754ec3a9d164e": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "HTMLModel", "state": { "_dom_classes": [], "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "HTMLModel", "_view_count": null, "_view_module": "@jupyter-widgets/controls", "_view_module_version": "1.5.0", "_view_name": "HTMLView", "description": "", "description_tooltip": null, "layout": "IPY_MODEL_3d7de1da6dc5424bb29a8510b1cfc549", "placeholder": "​", "style": "IPY_MODEL_d9f595f07ea14377ad4f2af81b3dc236", "value": "100%" } }, "1f07aa6d2b744092ac1b2f13352b3909": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "HTMLModel", "state": { "_dom_classes": [], "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "HTMLModel", "_view_count": null, "_view_module": "@jupyter-widgets/controls", "_view_module_version": "1.5.0", "_view_name": "HTMLView", "description": "", "description_tooltip": null, "layout": "IPY_MODEL_1417bf853c6241038d6d01fae7183091", "placeholder": "​", "style": "IPY_MODEL_9d78a73117ff4b59b6f6499b7f978d14", "value": " 588/588 [00:00<00:00, 33.5kB/s]" } }, "22eed99a28c74a80a42f95cc2529913b": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "FloatProgressModel", "state": { "_dom_classes": [], "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "FloatProgressModel", "_view_count": null, "_view_module": "@jupyter-widgets/controls", "_view_module_version": "1.5.0", "_view_name": "ProgressView", "bar_style": "success", "description": "", "description_tooltip": null, "layout": "IPY_MODEL_0374d3f515e640ea80e11cb8cffdbf7d", "max": 588, "min": 0, "orientation": "horizontal", "style": "IPY_MODEL_e08504a1906e41bf8cfe2e7d231e0f2f", "value": 588 } }, "297389dcbf3040f19fb6773f776f3f2e": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "HBoxModel", "state": { "_dom_classes": [], "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "HBoxModel", "_view_count": null, "_view_module": "@jupyter-widgets/controls", "_view_module_version": "1.5.0", "_view_name": "HBoxView", "box_style": "", "children": [ "IPY_MODEL_1a8353acfc6e4a77b5e754ec3a9d164e", "IPY_MODEL_a6784a8092dd4652bb19a142b05986b1", "IPY_MODEL_74d5c798f06346e9bd1cf96bf72669ff" ], "layout": "IPY_MODEL_d937f93cbe12423d9198d740769eaaad" } }, "2c90127799134934b6eb68325838a8a8": { "model_module": "@jupyter-widgets/base", "model_module_version": "1.2.0", "model_name": "LayoutModel", "state": { "_model_module": "@jupyter-widgets/base", "_model_module_version": "1.2.0", "_model_name": "LayoutModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "LayoutView", "align_content": null, "align_items": null, "align_self": null, "border": null, "bottom": null, "display": null, "flex": null, "flex_flow": null, "grid_area": null, "grid_auto_columns": null, "grid_auto_flow": null, "grid_auto_rows": null, "grid_column": null, "grid_gap": null, "grid_row": null, "grid_template_areas": null, "grid_template_columns": null, "grid_template_rows": null, "height": null, "justify_content": null, "justify_items": null, "left": null, "margin": null, "max_height": null, "max_width": null, "min_height": null, "min_width": null, "object_fit": null, "object_position": null, "order": null, "overflow": null, "overflow_x": null, "overflow_y": null, "padding": null, "right": null, "top": null, "visibility": null, "width": null } }, "3267638f39f34833955764925bc59ad9": { "model_module": "@jupyter-widgets/base", "model_module_version": "1.2.0", "model_name": "LayoutModel", "state": { "_model_module": "@jupyter-widgets/base", "_model_module_version": "1.2.0", "_model_name": "LayoutModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "LayoutView", "align_content": null, "align_items": null, "align_self": null, "border": null, "bottom": null, "display": null, "flex": null, "flex_flow": null, "grid_area": null, "grid_auto_columns": null, "grid_auto_flow": null, "grid_auto_rows": null, "grid_column": null, "grid_gap": null, "grid_row": null, "grid_template_areas": null, "grid_template_columns": null, "grid_template_rows": null, "height": null, "justify_content": null, "justify_items": null, "left": null, "margin": null, "max_height": null, "max_width": null, "min_height": null, "min_width": null, "object_fit": null, "object_position": null, "order": null, "overflow": null, "overflow_x": null, "overflow_y": null, "padding": null, "right": null, "top": null, "visibility": null, "width": null } }, "36e5a40703714c76ba41451af838a1fe": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "DescriptionStyleModel", "state": { "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "DescriptionStyleModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "StyleView", "description_width": "" } }, "3d7de1da6dc5424bb29a8510b1cfc549": { "model_module": "@jupyter-widgets/base", "model_module_version": "1.2.0", "model_name": "LayoutModel", "state": { "_model_module": "@jupyter-widgets/base", "_model_module_version": "1.2.0", "_model_name": "LayoutModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "LayoutView", "align_content": null, "align_items": null, "align_self": null, "border": null, "bottom": null, "display": null, "flex": null, "flex_flow": null, "grid_area": null, "grid_auto_columns": null, "grid_auto_flow": null, "grid_auto_rows": null, "grid_column": null, "grid_gap": null, "grid_row": null, "grid_template_areas": null, "grid_template_columns": null, "grid_template_rows": null, "height": null, "justify_content": null, "justify_items": null, "left": null, "margin": null, "max_height": null, "max_width": null, "min_height": null, "min_width": null, "object_fit": null, "object_position": null, "order": null, "overflow": null, "overflow_x": null, "overflow_y": null, "padding": null, "right": null, "top": null, "visibility": null, "width": null } }, "497302a6acc84773a84ff23cd36da214": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "HBoxModel", "state": { "_dom_classes": [], "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "HBoxModel", "_view_count": null, "_view_module": "@jupyter-widgets/controls", "_view_module_version": "1.5.0", "_view_name": "HBoxView", "box_style": "", "children": [ "IPY_MODEL_b62cf033432f40e0a9ee259515167b51", "IPY_MODEL_22eed99a28c74a80a42f95cc2529913b", "IPY_MODEL_1f07aa6d2b744092ac1b2f13352b3909" ], "layout": "IPY_MODEL_7a7468b6571c4b818f706b13e37345b5" } }, "6747454af34a467690d783bae764fbfa": { "model_module": "@jupyter-widgets/base", "model_module_version": "1.2.0", "model_name": "LayoutModel", "state": { "_model_module": "@jupyter-widgets/base", "_model_module_version": "1.2.0", "_model_name": "LayoutModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "LayoutView", "align_content": null, "align_items": null, "align_self": null, "border": null, "bottom": null, "display": null, "flex": null, "flex_flow": null, "grid_area": null, "grid_auto_columns": null, "grid_auto_flow": null, "grid_auto_rows": null, "grid_column": null, "grid_gap": null, "grid_row": null, "grid_template_areas": null, "grid_template_columns": null, "grid_template_rows": null, "height": null, "justify_content": null, "justify_items": null, "left": null, "margin": null, "max_height": null, "max_width": null, "min_height": null, "min_width": null, "object_fit": null, "object_position": null, "order": null, "overflow": null, "overflow_x": null, "overflow_y": null, "padding": null, "right": null, "top": null, "visibility": null, "width": null } }, "74d5c798f06346e9bd1cf96bf72669ff": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "HTMLModel", "state": { "_dom_classes": [], "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "HTMLModel", "_view_count": null, "_view_module": "@jupyter-widgets/controls", "_view_module_version": "1.5.0", "_view_name": "HTMLView", "description": "", "description_tooltip": null, "layout": "IPY_MODEL_6747454af34a467690d783bae764fbfa", "placeholder": "​", "style": "IPY_MODEL_d6ef70581b8841d4b7455f1074373478", "value": " 1/1 [00:02<00:00,  2.84s/it]" } }, "7a23197e84e449a6b77b893f4bcc81fb": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "ProgressStyleModel", "state": { "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "ProgressStyleModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "StyleView", "bar_color": null, "description_width": "" } }, "7a7468b6571c4b818f706b13e37345b5": { "model_module": "@jupyter-widgets/base", "model_module_version": "1.2.0", "model_name": "LayoutModel", "state": { "_model_module": "@jupyter-widgets/base", "_model_module_version": "1.2.0", "_model_name": "LayoutModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "LayoutView", "align_content": null, "align_items": null, "align_self": null, "border": null, "bottom": null, "display": null, "flex": null, "flex_flow": null, "grid_area": null, "grid_auto_columns": null, "grid_auto_flow": null, "grid_auto_rows": null, "grid_column": null, "grid_gap": null, "grid_row": null, "grid_template_areas": null, "grid_template_columns": null, "grid_template_rows": null, "height": null, "justify_content": null, "justify_items": null, "left": null, "margin": null, "max_height": null, "max_width": null, "min_height": null, "min_width": null, "object_fit": null, "object_position": null, "order": null, "overflow": null, "overflow_x": null, "overflow_y": null, "padding": null, "right": null, "top": null, "visibility": null, "width": null } }, "7ca294a7b4aa40668f8ea154f744a4b4": { "model_module": "@jupyter-widgets/base", "model_module_version": "1.2.0", "model_name": "LayoutModel", "state": { "_model_module": "@jupyter-widgets/base", "_model_module_version": "1.2.0", "_model_name": "LayoutModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "LayoutView", "align_content": null, "align_items": null, "align_self": null, "border": null, "bottom": null, "display": null, "flex": null, "flex_flow": null, "grid_area": null, "grid_auto_columns": null, "grid_auto_flow": null, "grid_auto_rows": null, "grid_column": null, "grid_gap": null, "grid_row": null, "grid_template_areas": null, "grid_template_columns": null, "grid_template_rows": null, "height": null, "justify_content": null, "justify_items": null, "left": null, "margin": null, "max_height": null, "max_width": null, "min_height": null, "min_width": null, "object_fit": null, "object_position": null, "order": null, "overflow": null, "overflow_x": null, "overflow_y": null, "padding": null, "right": null, "top": null, "visibility": null, "width": null } }, "906abe2734de4a7bbceb71fefb265ae8": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "FloatProgressModel", "state": { "_dom_classes": [], "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "FloatProgressModel", "_view_count": null, "_view_module": "@jupyter-widgets/controls", "_view_module_version": "1.5.0", "_view_name": "ProgressView", "bar_style": "success", "description": "", "description_tooltip": null, "layout": "IPY_MODEL_2c90127799134934b6eb68325838a8a8", "max": 167832240, "min": 0, "orientation": "horizontal", "style": "IPY_MODEL_c0a6e820a2eb41dc9c6244f614272ef5", "value": 167832240 } }, "9d78a73117ff4b59b6f6499b7f978d14": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "DescriptionStyleModel", "state": { "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "DescriptionStyleModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "StyleView", "description_width": "" } }, "a1242cc96bf9450abbbc00593bc7c8a8": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "HTMLModel", "state": { "_dom_classes": [], "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "HTMLModel", "_view_count": null, "_view_module": "@jupyter-widgets/controls", "_view_module_version": "1.5.0", "_view_name": "HTMLView", "description": "", "description_tooltip": null, "layout": "IPY_MODEL_d27389522e9c437695f0e93a0ecf1f09", "placeholder": "​", "style": "IPY_MODEL_16b0f13d883d48809d4703704f43204c", "value": "adapter_model.safetensors: " } }, "a4a6ed211ca44d509cb82311c27caba3": { "model_module": "@jupyter-widgets/base", "model_module_version": "1.2.0", "model_name": "LayoutModel", "state": { "_model_module": "@jupyter-widgets/base", "_model_module_version": "1.2.0", "_model_name": "LayoutModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "LayoutView", "align_content": null, "align_items": null, "align_self": null, "border": null, "bottom": null, "display": null, "flex": null, "flex_flow": null, "grid_area": null, "grid_auto_columns": null, "grid_auto_flow": null, "grid_auto_rows": null, "grid_column": null, "grid_gap": null, "grid_row": null, "grid_template_areas": null, "grid_template_columns": null, "grid_template_rows": null, "height": null, "justify_content": null, "justify_items": null, "left": null, "margin": null, "max_height": null, "max_width": null, "min_height": null, "min_width": null, "object_fit": null, "object_position": null, "order": null, "overflow": null, "overflow_x": null, "overflow_y": null, "padding": null, "right": null, "top": null, "visibility": null, "width": null } }, "a6784a8092dd4652bb19a142b05986b1": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "FloatProgressModel", "state": { "_dom_classes": [], "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "FloatProgressModel", "_view_count": null, "_view_module": "@jupyter-widgets/controls", "_view_module_version": "1.5.0", "_view_name": "ProgressView", "bar_style": "success", "description": "", "description_tooltip": null, "layout": "IPY_MODEL_3267638f39f34833955764925bc59ad9", "max": 1, "min": 0, "orientation": "horizontal", "style": "IPY_MODEL_7a23197e84e449a6b77b893f4bcc81fb", "value": 1 } }, "b62cf033432f40e0a9ee259515167b51": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "HTMLModel", "state": { "_dom_classes": [], "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "HTMLModel", "_view_count": null, "_view_module": "@jupyter-widgets/controls", "_view_module_version": "1.5.0", "_view_name": "HTMLView", "description": "", "description_tooltip": null, "layout": "IPY_MODEL_d2117400d4744ac498d1582a3336f905", "placeholder": "​", "style": "IPY_MODEL_36e5a40703714c76ba41451af838a1fe", "value": "README.md: 100%" } }, "c0a6e820a2eb41dc9c6244f614272ef5": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "ProgressStyleModel", "state": { "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "ProgressStyleModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "StyleView", "bar_color": null, "description_width": "" } }, "c7a2e8fad27e4fa2b9c80a970002d6bd": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "DescriptionStyleModel", "state": { "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "DescriptionStyleModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "StyleView", "description_width": "" } }, "d2117400d4744ac498d1582a3336f905": { "model_module": "@jupyter-widgets/base", "model_module_version": "1.2.0", "model_name": "LayoutModel", "state": { "_model_module": "@jupyter-widgets/base", "_model_module_version": "1.2.0", "_model_name": "LayoutModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "LayoutView", "align_content": null, "align_items": null, "align_self": null, "border": null, "bottom": null, "display": null, "flex": null, "flex_flow": null, "grid_area": null, "grid_auto_columns": null, "grid_auto_flow": null, "grid_auto_rows": null, "grid_column": null, "grid_gap": null, "grid_row": null, "grid_template_areas": null, "grid_template_columns": null, "grid_template_rows": null, "height": null, "justify_content": null, "justify_items": null, "left": null, "margin": null, "max_height": null, "max_width": null, "min_height": null, "min_width": null, "object_fit": null, "object_position": null, "order": null, "overflow": null, "overflow_x": null, "overflow_y": null, "padding": null, "right": null, "top": null, "visibility": null, "width": null } }, "d27389522e9c437695f0e93a0ecf1f09": { "model_module": "@jupyter-widgets/base", "model_module_version": "1.2.0", "model_name": "LayoutModel", "state": { "_model_module": "@jupyter-widgets/base", "_model_module_version": "1.2.0", "_model_name": "LayoutModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "LayoutView", "align_content": null, "align_items": null, "align_self": null, "border": null, "bottom": null, "display": null, "flex": null, "flex_flow": null, "grid_area": null, "grid_auto_columns": null, "grid_auto_flow": null, "grid_auto_rows": null, "grid_column": null, "grid_gap": null, "grid_row": null, "grid_template_areas": null, "grid_template_columns": null, "grid_template_rows": null, "height": null, "justify_content": null, "justify_items": null, "left": null, "margin": null, "max_height": null, "max_width": null, "min_height": null, "min_width": null, "object_fit": null, "object_position": null, "order": null, "overflow": null, "overflow_x": null, "overflow_y": null, "padding": null, "right": null, "top": null, "visibility": null, "width": null } }, "d6ef70581b8841d4b7455f1074373478": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "DescriptionStyleModel", "state": { "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "DescriptionStyleModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "StyleView", "description_width": "" } }, "d937f93cbe12423d9198d740769eaaad": { "model_module": "@jupyter-widgets/base", "model_module_version": "1.2.0", "model_name": "LayoutModel", "state": { "_model_module": "@jupyter-widgets/base", "_model_module_version": "1.2.0", "_model_name": "LayoutModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "LayoutView", "align_content": null, "align_items": null, "align_self": null, "border": null, "bottom": null, "display": null, "flex": null, "flex_flow": null, "grid_area": null, "grid_auto_columns": null, "grid_auto_flow": null, "grid_auto_rows": null, "grid_column": null, "grid_gap": null, "grid_row": null, "grid_template_areas": null, "grid_template_columns": null, "grid_template_rows": null, "height": null, "justify_content": null, "justify_items": null, "left": null, "margin": null, "max_height": null, "max_width": null, "min_height": null, "min_width": null, "object_fit": null, "object_position": null, "order": null, "overflow": null, "overflow_x": null, "overflow_y": null, "padding": null, "right": null, "top": null, "visibility": null, "width": null } }, "d9f595f07ea14377ad4f2af81b3dc236": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "DescriptionStyleModel", "state": { "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "DescriptionStyleModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "StyleView", "description_width": "" } }, "de7e5ca6727c4c0d99522d35981a4ad2": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "HTMLModel", "state": { "_dom_classes": [], "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "HTMLModel", "_view_count": null, "_view_module": "@jupyter-widgets/controls", "_view_module_version": "1.5.0", "_view_name": "HTMLView", "description": "", "description_tooltip": null, "layout": "IPY_MODEL_a4a6ed211ca44d509cb82311c27caba3", "placeholder": "​", "style": "IPY_MODEL_c7a2e8fad27e4fa2b9c80a970002d6bd", "value": " 176M/? [00:02<00:00, 82.4MB/s]" } }, "e08504a1906e41bf8cfe2e7d231e0f2f": { "model_module": "@jupyter-widgets/controls", "model_module_version": "1.5.0", "model_name": "ProgressStyleModel", "state": { "_model_module": "@jupyter-widgets/controls", "_model_module_version": "1.5.0", "_model_name": "ProgressStyleModel", "_view_count": null, "_view_module": "@jupyter-widgets/base", "_view_module_version": "1.2.0", "_view_name": "StyleView", "bar_color": null, "description_width": "" } } } } }, "nbformat": 4, "nbformat_minor": 4 }