update llm sft notebook

examples/train/notebook/llm/infer.ipynb

@@ -0,0 +1,117 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## Inference\n",
+    "We have trained a well-trained checkpoint through the `self-cognition-sft.ipynb` tutorial, and here we use `PtEngine` to do the inference on it."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 8,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# import some libraries\n",
+    "import os\n",
+    "os.environ['CUDA_VISIBLE_DEVICES'] = '0'\n",
+    "\n",
+    "from swift.llm import InferEngine, InferRequest, PtEngine, RequestConfig, get_template\n",
+    "from swift.tuners import Swift"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 9,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Hyperparameters for inference\n",
+    "last_model_checkpoint = '/mnt/nas2/huangjintao.hjt/work/llmscope/examples/train/notebook/output/Qwen2.5-3B-Instruct/checkpoint-93'\n",
+    "\n",
+    "# model\n",
+    "model = 'Qwen/Qwen2.5-3B-Instruct'  # model_id or model_path\n",
+    "system = 'You are a helpful assistant.'\n",
+    "infer_backend = 'pt'\n",
+    "\n",
+    "# generation_config\n",
+    "max_new_tokens = 512\n",
+    "temperature = 0\n",
+    "stream = True"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Get model and template, and load LoRA weights.\n",
+    "engine = PtEngine(model)\n",
+    "engine.model = Swift.from_pretrained(engine.model, last_model_checkpoint, inference_mode=True)\n",
+    "template = get_template(engine.model.model_meta.template, engine.tokenizer, default_system=system)\n",
+    "# The default mode of the template is 'pt', so there is no need to make any changes.\n",
+    "# template.set_mode('pt')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "query_list = [\n",
+    "    'who are you?',\n",
+    "    \"What should I do if I can't sleep at night?\",\n",
+    "    '你是谁训练的？',\n",
+    "]\n",
+    "\n",
+    "def infer_stream(engine: InferEngine, infer_request: InferRequest):\n",
+    "    request_config = RequestConfig(max_tokens=max_new_tokens, temperature=temperature, stream=True)\n",
+    "    gen = engine.infer([infer_request], request_config)\n",
+    "    query = infer_request.messages[0]['content']\n",
+    "    print(f'query: {query}\\nresponse: ', end='')\n",
+    "    for resp_list in gen:\n",
+    "        print(resp_list[0].choices[0].delta.content, end='', flush=True)\n",
+    "    print()\n",
+    "\n",
+    "def infer(engine: InferEngine, infer_request: InferRequest):\n",
+    "    request_config = RequestConfig(max_tokens=max_new_tokens, temperature=temperature)\n",
+    "    resp_list = engine.infer([infer_request], request_config)\n",
+    "    query = infer_request.messages[0]['content']\n",
+    "    response = resp_list[0].choices[0].message.content\n",
+    "    print(f'query: {query}')\n",
+    "    print(f'response: {response}')\n",
+    "\n",
+    "\n",
+    "infer_func = infer_stream if stream else infer\n",
+    "for query in query_list:\n",
+    "    infer_func(engine, InferRequest(messages=[{'role': 'user', 'content': query}]))\n",
+    "    print('-' * 50)\n"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "test_py310",
+   "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": "3.10.15"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}

examples/train/notebook/llm/self-cognition-sft.ipynb

@@ -0,0 +1,222 @@
+{
+ "cells": [
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "## 10-minute self-cognition fine-tuning\n",
+    "\n",
+    "Here is a demonstration of using python to perform self-cognition fine-tuning of Qwen2-7B-Instruct. Through this tutorial, you can quickly understand some details of swift sft, which will be of great help in customizing ms-swift for you~\n",
+    "\n",
+    "Are you ready? Let's begin the journey..."
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {
+    "vscode": {
+     "languageId": "shellscript"
+    }
+   },
+   "outputs": [],
+   "source": [
+    "# install ms-swift\n",
+    "!git clone https://github.com/modelscope/ms-swift.git\n",
+    "%cd ms-swift\n",
+    "!pip install -e '.[llm]'"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# import some libraries\n",
+    "import os\n",
+    "os.environ['CUDA_VISIBLE_DEVICES'] = '0'\n",
+    "\n",
+    "from swift.llm import get_model_tokenizer, load_dataset, get_template, EncodePreprocessor\n",
+    "from swift.utils import get_logger, find_all_linears, get_model_parameter_info, plot_images, seed_everything\n",
+    "from swift.tuners import Swift, LoraConfig\n",
+    "\n",
+    "from swift.trainers import Seq2SeqTrainer, Seq2SeqTrainingArguments\n",
+    "\n",
+    "from functools import partial\n",
+    "\n",
+    "\n",
+    "logger = get_logger()\n",
+    "seed_everything(42)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": 2,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Hyperparameters for training\n",
+    "# model\n",
+    "model = 'Qwen/Qwen2.5-3B-Instruct'  # model_id or model_path\n",
+    "system = 'You are a helpful assistant.'\n",
+    "output_dir = 'output/Qwen2.5-3B-Instruct'\n",
+    "\n",
+    "# dataset\n",
+    "dataset = ['AI-ModelScope/alpaca-gpt4-data-zh#500', 'AI-ModelScope/alpaca-gpt4-data-en#500',\n",
+    "           'swift/self-cognition#500']  # dataset_id or dataset_path\n",
+    "max_length = 2048\n",
+    "split_dataset_ratio = 0.01  # Split validation set\n",
+    "num_proc = 4  # The number of processes for data loading.\n",
+    "# The following two parameters are used to override the placeholders in the self-cognition dataset.\n",
+    "model_name = ['小黄', 'Xiao Huang']  # The Chinese name and English name of the model\n",
+    "model_author = ['魔搭', 'ModelScope']  # The Chinese name and English name of the model author\n",
+    "\n",
+    "# lora\n",
+    "lora_rank = 8\n",
+    "lora_alpha = 32\n",
+    "\n",
+    "# training_args\n",
+    "training_args = Seq2SeqTrainingArguments(\n",
+    "    output_dir=output_dir,\n",
+    "    learning_rate=1e-4,\n",
+    "    per_device_train_batch_size=1,\n",
+    "    per_device_eval_batch_size=1,\n",
+    "    gradient_checkpointing=True,\n",
+    "    weight_decay=0.1,\n",
+    "    lr_scheduler_type='cosine',\n",
+    "    warmup_ratio=0.05,\n",
+    "    report_to=['tensorboard'],\n",
+    "    logging_first_step=True,\n",
+    "    save_strategy='steps',\n",
+    "    save_steps=50,\n",
+    "    eval_strategy='steps',\n",
+    "    eval_steps=50,\n",
+    "    gradient_accumulation_steps=16,\n",
+    "    num_train_epochs=1,\n",
+    "    metric_for_best_model='loss',\n",
+    "    save_total_limit=2,\n",
+    "    logging_steps=5,\n",
+    "    dataloader_num_workers=1,\n",
+    ")\n",
+    "\n",
+    "\n",
+    "output_dir = os.path.abspath(os.path.expanduser(output_dir))\n",
+    "logger.info(f'output_dir: {output_dir}')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Obtain the model and template, and add a trainable Lora layer on the model.\n",
+    "model, tokenizer = get_model_tokenizer(model)\n",
+    "logger.info(f'model_info: {model.model_info}')\n",
+    "template = get_template(model.model_meta.template, tokenizer, default_system=system, max_length=max_length)\n",
+    "template.set_mode('train')\n",
+    "\n",
+    "target_modules = find_all_linears(model)\n",
+    "lora_config = LoraConfig(task_type='CAUSAL_LM', r=lora_rank, lora_alpha=lora_alpha,\n",
+    "                         target_modules=target_modules)\n",
+    "model = Swift.prepare_model(model, lora_config)\n",
+    "logger.info(f'lora_config: {lora_config}')\n",
+    "\n",
+    "# Print model structure and trainable parameters.\n",
+    "logger.info(f'model: {model}')\n",
+    "model_parameter_info = get_model_parameter_info(model)\n",
+    "logger.info(f'model_parameter_info: {model_parameter_info}')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Download and load the dataset, split it into a training set and a validation set,\n",
+    "# and encode the text data into tokens.\n",
+    "train_dataset, val_dataset = load_dataset(dataset, split_dataset_ratio=split_dataset_ratio, num_proc=num_proc,\n",
+    "        model_name=model_name, model_author=model_author)\n",
+    "\n",
+    "logger.info(f'train_dataset: {train_dataset}')\n",
+    "logger.info(f'val_dataset: {val_dataset}')\n",
+    "logger.info(f'train_dataset[0]: {train_dataset[0]}')\n",
+    "\n",
+    "train_dataset = EncodePreprocessor(template=template)(train_dataset, num_proc=num_proc)\n",
+    "val_dataset = EncodePreprocessor(template=template)(val_dataset, num_proc=num_proc)\n",
+    "logger.info(f'encoded_train_dataset[0]: {train_dataset[0]}')\n",
+    "\n",
+    "# Print a sample\n",
+    "template.print_inputs(train_dataset[0])"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Get the trainer and start the training.\n",
+    "model.enable_input_require_grads()  # Compatible with gradient checkpointing\n",
+    "trainer = Seq2SeqTrainer(\n",
+    "    model=model,\n",
+    "    args=training_args,\n",
+    "    data_collator=template.data_collator,\n",
+    "    train_dataset=train_dataset,\n",
+    "    eval_dataset=val_dataset,\n",
+    "    template=template,\n",
+    ")\n",
+    "trainer.model_accepts_loss_kwargs = True  # Compatible with transformers >= 4.46\n",
+    "trainer.train()\n",
+    "\n",
+    "last_model_checkpoint = trainer.state.last_model_checkpoint\n",
+    "logger.info(f'last_model_checkpoint: {last_model_checkpoint}')"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "# Visualize the training loss.\n",
+    "# You can also use the TensorBoard visualization interface during training by entering\n",
+    "# `tensorboard --logdir '{output_dir}/runs'` at the command line.\n",
+    "images_dir = os.path.join(output_dir, 'images')\n",
+    "logger.info(f'images_dir: {images_dir}')\n",
+    "plot_images(images_dir, training_args.logging_dir, ['train/loss'], 0.9)  # save images\n",
+    "\n",
+    "# Read and display the image.\n",
+    "# The light yellow line represents the actual loss value,\n",
+    "# while the yellow line represents the loss value smoothed with a smoothing factor of 0.9.\n",
+    "from IPython.display import display\n",
+    "from PIL import Image\n",
+    "image = Image.open(os.path.join(images_dir, 'train_loss.png'))\n",
+    "display(image)"
+   ]
+  }
+ ],
+ "metadata": {
+  "kernelspec": {
+   "display_name": "py310",
+   "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": "3.10.15"
+  }
+ },
+ "nbformat": 4,
+ "nbformat_minor": 2
+}