Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

add unsloth ipynb #194

Merged
merged 3 commits into from
Sep 10, 2024
Merged
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
357 changes: 357 additions & 0 deletions integrations/model-training/unsloth/notebooks/Comet_and_unsloth.ipynb
Original file line number Diff line number Diff line change
@@ -0,0 +1,357 @@
{
"cells": [
{
"cell_type": "markdown",
"metadata": {
"id": "6RF2EQaKDoxr"
},
"source": [
"<a href=\"https://www.comet.com/site/?utm_medium=colab&utm_source=comet-examples&utm_campaign=unsloth\" >\n",
" <img src=\"https://cdn.comet.ml/img/notebook_logo.png\">\n",
"</a>"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "KDjO5WbeDtp0"
},
"source": [
"# Comet and unsloth\n",
"\n",
"[Comet](https://www.comet.com/site/?utm_medium=colab&utm_source=comet-examples&utm_campaign=unsloth) is an MLOps platform designed to help data scientists and teams build better models faster! Comet provides tooling to track, explain, manage, and monitor your models in a single place! It works with Jupyter notebooks and scripts and-- most importantly--it's 100% free to get started!\n",
"\n",
"[unsloth](https://github.com/unslothai/unsloth) dramatically improves the speed and efficiency of LLM fine-tuning for models including Llama, Phi-3, Gemma, Mistral, and more. For a full listed of 100+ supported unsloth models, [see here](https://huggingface.co/unsloth).\n",
"\n",
"Instrument your torchtune training runs with Comet to start managing experiments with efficiency, reproducibility, and collaboration in mind.\n",
"\n",
"Find more information about [our integration with torchtune here](https://www.comet.com/docs/v2/integrations/third-party-tools/unsloth?utm_medium=colab&utm_source=comet-examples&utm_campaign=unsloth) or [learn about our other integrations here](https://www.comet.com/docs/v2/integrations?utm_medium=colab&utm_source=comet-examples&utm_campaign=unsloth)."
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "f0fhpJZgsYII"
},
"source": [
"## ⚙ Install and import dependencies"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "5zTSjY9r4cGc"
},
"outputs": [],
"source": [
"%pip install comet_ml \"unsloth[colab-new] @ git+https://github.com/unslothai/unsloth.git\" \"torch>=2.4.0\" xformers trl peft accelerate bitsandbytes triton"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "p8GTqJOW88fI"
},
"outputs": [],
"source": [
"import comet_ml\n",
"\n",
"comet_ml.login()\n",
"exp = comet_ml.Experiment(project_name=\"comet-example-unsloth\")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "vdOozTnqq8pL"
},
"outputs": [],
"source": [
"from huggingface_hub import notebook_login\n",
"\n",
"notebook_login()"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "5bfxeJqKwpKj"
},
"source": [
"## ⚙ Download model"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "bZei5Dvw7_kM"
},
"outputs": [],
"source": [
"from unsloth import FastLanguageModel\n",
"import torch\n",
"\n",
"max_seq_length = 2048\n",
"dtype = (\n",
" None # None for auto detection. Float16 for Tesla T4, V100, Bfloat16 for Ampere+\n",
")\n",
"load_in_4bit = True"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "NG--Sbr1n5Cv"
},
"source": [
"Find the full list of [100+ supported unsloth models here](https://huggingface.co/unsloth). For a full list of supported 4-bit models see [here](https://huggingface.co/collections/unsloth/load-4bit-models-4x-faster-659042e3a41c3cbad582e734)"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "L09m09Xs8Y2e"
},
"outputs": [],
"source": [
"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,\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "tbcH3PNSwRvy"
},
"source": [
"## ⚙ Add LoRA adapters"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "BnXS0Trn8md9"
},
"outputs": [],
"source": [
"model = FastLanguageModel.get_peft_model(\n",
" model,\n",
" r=16, # Suggested 8, 16, 32, 64, 128\n",
" target_modules=[\n",
" \"q_proj\",\n",
" \"k_proj\",\n",
" \"v_proj\",\n",
" \"o_proj\",\n",
" \"gate_proj\",\n",
" \"up_proj\",\n",
" \"down_proj\",\n",
" ],\n",
" lora_alpha=16,\n",
" lora_dropout=0, # Supports any, but = 0 is optimized\n",
" bias=\"none\", # Supports any, but = \"none\" is optimized\n",
" use_gradient_checkpointing=\"unsloth\", # True or \"unsloth\" for very long context\n",
" random_state=3407,\n",
" use_rslora=False, # rank stabilized LoRA\n",
" loftq_config=None, # LoftQ\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "AoqQQ5n4wcfx"
},
"source": [
"## ⚙ Data preparation"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "z9XFfoIz8sTI"
},
"outputs": [],
"source": [
"alpaca_prompt = \"\"\"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",
"{}\n",
"\n",
"### Input:\n",
"{}\n",
"\n",
"### Response:\n",
"{}\"\"\"\n",
"\n",
"EOS_TOKEN = tokenizer.eos_token # add EOS_TOKEN\n",
"\n",
"\n",
"def formatting_prompts_func(examples):\n",
" instructions = examples[\"instruction\"]\n",
" inputs = examples[\"input\"]\n",
" outputs = examples[\"output\"]\n",
" texts = []\n",
" for instruction, input, output in zip(instructions, inputs, outputs):\n",
" # 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 {\n",
" \"text\": texts,\n",
" }\n",
"\n",
"\n",
"pass\n",
"\n",
"from datasets import load_dataset\n",
"\n",
"dataset = load_dataset(\"yahma/alpaca-cleaned\", split=\"train\")\n",
"dataset = dataset.map(\n",
" formatting_prompts_func,\n",
" batched=True,\n",
")"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "wljLKG7LwiHE"
},
"source": [
"## ⚙ Training"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "Ra62g_5f8vYo"
},
"outputs": [],
"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=2,\n",
" gradient_accumulation_steps=4,\n",
" warmup_steps=5,\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",
" ),\n",
")"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "e5EHhd5zANxX"
},
"outputs": [],
"source": [
"trainer_stats = trainer.train()"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "8II72I92Bm9s"
},
"outputs": [],
"source": [
"exp.end()"
]
},
{
"cell_type": "markdown",
"metadata": {
"id": "0IdjUa_0pzPQ"
},
"source": [
"## ⚙ Inference"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {
"id": "76su-yv9Aad0"
},
"outputs": [],
"source": [
"# alpaca_prompt = Copied from above\n",
"FastLanguageModel.for_inference(model)\n",
"inputs = tokenizer(\n",
" [\n",
" alpaca_prompt.format(\n",
" \"Continue the fibonnaci sequence.\", # instruction\n",
" \"1, 1, 2, 3, 5, 8\", # input\n",
" \"\", # output - leave this blank for generation\n",
" )\n",
" ],\n",
" return_tensors=\"pt\",\n",
").to(\"cuda\")\n",
"\n",
"outputs = model.generate(**inputs, max_new_tokens=64, use_cache=True)\n",
"tokenizer.batch_decode(outputs)"
]
},
{
"cell_type": "code",
"source": [],
"metadata": {
"id": "h6alQJYpwlAM"
},
"execution_count": null,
"outputs": []
}
],
"metadata": {
"accelerator": "GPU",
"colab": {
"gpuType": "L4",
"machine_shape": "hm",
"provenance": [],
"toc_visible": true
},
"kernelspec": {
"display_name": "Python 3",
"name": "python3"
},
"language_info": {
"name": "python"
}
},
"nbformat": 4,
"nbformat_minor": 0
}
Loading