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.

Sign up for GitHub

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 torchtune colab #520

Merged
merged 1 commit into from
May 3, 2024
Merged

Add torchtune colab #520

Changes from all commits
Commits
Show all changes
1 commit
Select commit
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
257 changes: 257 additions & 0 deletions colabs/torchtune/torchtune_and_wandb.ipynb
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -0,0 +1,257 @@
{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<a href=\"https://colab.research.google.com/github/wandb/examples/blob/master/colabs/torchtune/torchtune_and_wandb.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>\n",
"<!--- @wandbcode{torchtune-colab} -->"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"<img src=\"http://wandb.me/logo-im-png\" width=\"400\" alt=\"Weights & Biases\" />\n",
"<!--- @wandbcode{torchtune-colab} -->"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Getting Started with torchtune and Weigths & Biases\n",
"\n",
"In this notebook you will learn how to use [torchtune](https://github.com/pytorch/torchtune) with [Weights & Biases](https://wandb.ai) to monitor your training runs."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"> You need to select a machine a GPU, go to Runtime > Change runtime type > select a GPU (L40, A100 ideally)"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Setup the libraries"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"!git clone https://github.com/pytorch/torchtune"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"cd torchtune/"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"!python -m pip install -qqq \".[dev]\""
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"import wandb\n",
"wandb.login()"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## Download a Model\n",
"We will download a model from the Hugging Face Hub.\n",
"> you will need to provide an access token or call `huggingface-cli login`"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Download a model checkpoint using the provided `tune download` CLI\n",
"\n"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"!tune download mistralai/Mistral-7B-v0.1 \\\n",
" --output-dir /tmp/Mistral-7B-v0.1/ \\\n",
" --hf-token <HF_TOKEN>"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Let's create a torchtune config that enables W&B, to do so, we can grab the original Mistral 7B LoRA recipe and change the following lines to use W&B as our `metric_logger`:\n",
"```yaml\n",
"# Logging\n",
"metric_logger:\n",
" _component_: torchtune.utils.metric_logging.WandBLogger # <---You only need this to enable W&B\n",
" project: mistral_lora # <--- The W&B project to save our logs to\n",
"log_every_n_steps: 1\n",
"\n",
"```"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": []
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Let's save a modified version of the recipe using `%%writefile`:"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"%%writefile mistral_wandb_lora.yaml\n",
"tokenizer:\n",
" _component_: torchtune.models.mistral.mistral_tokenizer\n",
" path: /tmp/Mistral-7B-v0.1/tokenizer.model\n",
"\n",
"# Dataset\n",
"dataset:\n",
" _component_: torchtune.datasets.alpaca_dataset\n",
" train_on_input: True\n",
"seed: null\n",
"shuffle: True\n",
"\n",
"# Model Arguments\n",
"model:\n",
" _component_: torchtune.models.mistral.lora_mistral_7b\n",
" lora_attn_modules: ['q_proj', 'k_proj', 'v_proj']\n",
" apply_lora_to_mlp: True\n",
" apply_lora_to_output: True\n",
" lora_rank: 64\n",
" lora_alpha: 16\n",
"\n",
"checkpointer:\n",
" _component_: torchtune.utils.FullModelHFCheckpointer\n",
" checkpoint_dir: /tmp/Mistral-7B-v0.1\n",
" checkpoint_files: [\n",
" pytorch_model-00001-of-00002.bin,\n",
" pytorch_model-00002-of-00002.bin\n",
" ]\n",
" recipe_checkpoint: null\n",
" output_dir: /tmp/Mistral-7B-v0.1\n",
" model_type: MISTRAL\n",
"resume_from_checkpoint: False\n",
"\n",
"optimizer:\n",
" _component_: torch.optim.AdamW\n",
" lr: 2e-5\n",
"\n",
"lr_scheduler:\n",
" _component_: torchtune.modules.get_cosine_schedule_with_warmup\n",
" num_warmup_steps: 100\n",
"\n",
"loss:\n",
" _component_: torch.nn.CrossEntropyLoss\n",
"\n",
"# Fine-tuning arguments\n",
"batch_size: 2\n",
"epochs: 1\n",
"max_steps_per_epoch: 100\n",
"gradient_accumulation_steps: 2\n",
"compile: False\n",
"\n",
"# Training env\n",
"device: cuda\n",
"\n",
"# Memory management\n",
"enable_activation_checkpointing: True\n",
"\n",
"# Reduced precision\n",
"dtype: bf16\n",
"############################### Enable W&B #####################################\n",
"################################################################################\n",
"# Logging\n",
"metric_logger:\n",
" _component_: torchtune.utils.metric_logging.WandBLogger # <---You only need this to enable W&B\n",
" project: mistral_lora # <--- The W&B project to save our logs to\n",
"log_every_n_steps: 1\n",
"################################################################################\n",
"################################################################################\n",
"output_dir: /tmp/Mistral-7B-v0.1\n",
"log_peak_memory_stats: False\n",
"\n",
"# Profiler (disabled)\n",
"profiler:\n",
" _component_: torchtune.utils.profiler\n",
" enabled: False"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"Let's run the recipe with this modified config with W&B enabled"
]
},
{
"cell_type": "code",
"execution_count": null,
"metadata": {},
"outputs": [],
"source": [
"!tune run lora_finetune_single_device --config mistral_wandb_lora.yaml"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"That's it! Now you can click on the URL and continue monitoring your training on the Weights & Biases UI"
]
}
],
"metadata": {
"accelerator": "GPU",
"colab": {
"include_colab_link": true,
"provenance": [],
"toc_visible": true
},
"kernelspec": {
"display_name": "Python 3",
"name": "python3"
}
},
"nbformat": 4,
"nbformat_minor": 0
}
Loading