{ "cells": [ { "cell_type": "markdown", "id": "1a884871-7a65-4501-9063-c85ad260d0da", "metadata": {}, "source": [ "このnotebookはstockmark/stockmark-13bのモデルをkunishou/databricks-dolly-15k-jaのデータセットを用いてLoRA tuningするためのコードの例です。A100またはH100のGPUを用いることを想定しています。\n", "\n", "- モデル：https://huggingface.co/stockmark/stockmark-13b\n", "- データ：https://github.com/kunishou/databricks-dolly-15k-ja\n", "\n", "以下の例では、学習を1 epochを行います。A100 GPUで実行すると30分ほどかかります。\n", "\n", "また、ここで用いられているハイパーパラメータは最適化されたものではありませんので、必要に応じて調整してください。" ] }, { "cell_type": "markdown", "id": "93b3f4b5-2825-4ef3-a0ee-7a60155aee5d", "metadata": {}, "source": [ "# 準備" ] }, { "cell_type": "code", "execution_count": null, "id": "6a694ba9-a0fa-4f14-81cf-f35f683ba889", "metadata": {}, "outputs": [], "source": [ "import torch\n", "import datasets\n", "from transformers import AutoTokenizer, AutoModelForCausalLM, Trainer, TrainingArguments\n", "from peft import get_peft_model, LoraConfig, PeftModel, PeftConfig\n", "\n", "model_name = \"stockmark/stockmark-13b\"\n", "peft_model_name = \"stockmark-13b-adapter\"\n", "\n", "prompt_template = \"\"\"### Instruction:\n", "{instruction}\n", "\n", "### Input:\n", "{input}\n", "\n", "### Response:\n", "\"\"\"\n", "\n", "def encode(sample):\n", " prompt = prompt_template.format(instruction=sample[\"instruction\"], input=sample[\"input\"])\n", " target = sample[\"output\"]\n", " input_ids_prompt, input_ids_target = tokenizer([prompt, target], add_special_tokens=False).input_ids\n", " input_ids_prompt = [ tokenizer.bos_token_id ] + input_ids_prompt\n", " input_ids_target = input_ids_target + [ tokenizer.eos_token_id ]\n", " input_ids = input_ids_prompt + input_ids_target\n", " labels = input_ids.copy()\n", " labels[:len(input_ids_prompt)] = [-100] * len(input_ids_prompt) # ignore label tokens in a prompt for loss calculation\n", " return {\"input_ids\": input_ids, \"labels\": labels}\n", "\n", "def get_collator(tokenizer, max_length):\n", " def collator(batch):\n", " batch = [{ key: value[:max_length] for key, value in sample.items() } for sample in batch ]\n", " batch = tokenizer.pad(batch)\n", " batch[\"labels\"] = [ e + [-100] * (len(batch[\"input_ids\"][0]) - len(e)) for e in batch[\"labels\"] ]\n", " batch = { key: torch.tensor(value) for key, value in batch.items() }\n", " return batch\n", "\n", " return collator" ] }, { "cell_type": "markdown", "id": "51e6cfcf-1ac1-400e-a4bc-ea64375d0f9e", "metadata": {}, "source": [ "# データセットとモデルのロード" ] }, { "cell_type": "code", "execution_count": null, "id": "3ac80067-4e60-46c4-90da-05647cf96ccd", "metadata": {}, "outputs": [], "source": [ "# load_tokenizer\n", "tokenizer = AutoTokenizer.from_pretrained(model_name)\n", "\n", "# prepare dataset\n", "dataset_name = \"kunishou/databricks-dolly-15k-ja\"\n", "dataset = datasets.load_dataset(dataset_name)\n", "dataset = dataset.map(encode)\n", "dataset = dataset[\"train\"].train_test_split(0.1)\n", "train_dataset = dataset[\"train\"]\n", "val_dataset = dataset[\"test\"]\n", "\n", "# load model\n", "model = AutoModelForCausalLM.from_pretrained(model_name, device_map=\"auto\", torch_dtype=torch.bfloat16)\n", "\n", "peft_config = LoraConfig(\n", " task_type=\"CAUSAL_LM\",\n", " inference_mode=False,\n", " target_modules=[\"q_proj\", \"k_proj\", \"v_proj\", \"o_proj\"],\n", " r=16,\n", " lora_alpha=32,\n", " lora_dropout=0.05\n", ")\n", "\n", "model = get_peft_model(model, peft_config)\n", "model.print_trainable_parameters()" ] }, { "cell_type": "markdown", "id": "9b471da0-7fba-4127-8b07-22da4cbee6a9", "metadata": {}, "source": [ "# LoRA Tuning" ] }, { "cell_type": "code", "execution_count": null, "id": "b9bafa12-538c-4abb-b8b3-bffeb0990b46", "metadata": {}, "outputs": [], "source": [ "training_args = TrainingArguments(\n", " output_dir=\"./log_stockmark_13b\",\n", " learning_rate=2e-4,\n", " per_device_train_batch_size=2,\n", " gradient_accumulation_steps=8,\n", " per_device_eval_batch_size=16,\n", " num_train_epochs=1,\n", " logging_strategy='steps',\n", " logging_steps=10,\n", " save_strategy='epoch',\n", " evaluation_strategy='epoch',\n", " load_best_model_at_end=True,\n", " metric_for_best_model=\"eval_loss\",\n", " greater_is_better=False,\n", " save_total_limit=2\n", ")\n", "\n", "trainer = Trainer(\n", " model=model,\n", " args=training_args,\n", " train_dataset=train_dataset,\n", " eval_dataset=val_dataset,\n", " data_collator=get_collator(tokenizer, 320)\n", ")\n", "\n", "# LoRA tuning\n", "trainer.train()\n", "\n", "# save model\n", "model = trainer.model\n", "model.save_pretrained(peft_model_name)" ] }, { "cell_type": "markdown", "id": "a3f80a8e-1ac2-4bdc-8232-fe0ee18ffff5", "metadata": {}, "source": [ "# 学習したモデルのロード（Optional）\n", "異なるセッションでモデルを読み込む場合、まず最初の準備のセクションのコードを実行して、このコードを実行してください。" ] }, { "cell_type": "code", "execution_count": null, "id": "43241395-3035-4cb9-8c1c-45ffe8cd48be", "metadata": {}, "outputs": [], "source": [ "tokenizer = AutoTokenizer.from_pretrained(model_name)\n", "model = AutoModelForCausalLM.from_pretrained(model_name, device_map=\"auto\", torch_dtype=torch.bfloat16)\n", "model = PeftModel.from_pretrained(model, peft_model_name)" ] }, { "cell_type": "markdown", "id": "2ce4db1f-9bad-4c8e-9c04-d1102b299f24", "metadata": {}, "source": [ "# 推論" ] }, { "cell_type": "code", "execution_count": null, "id": "d7d6359b-e0ac-49df-a178-39bb9f79ca93", "metadata": {}, "outputs": [], "source": [ "prompt = prompt_template.format(instruction=\"自然言語処理とは？\", input=\"\")\n", "\n", "inputs = tokenizer(prompt, return_tensors=\"pt\").to(model.device)\n", "with torch.no_grad():\n", " tokens = model.generate(\n", " **inputs,\n", " max_new_tokens=128,\n", " do_sample=True,\n", " temperature=0.7\n", " )\n", "\n", "output = tokenizer.decode(tokens[0], skip_special_tokens=True)\n", "print(output)" ] } ], "metadata": { "kernelspec": { "display_name": "Python 3", "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": "3.8.10" } }, "nbformat": 4, "nbformat_minor": 5 }