File size: 11,741 Bytes
53e0dae | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 | # Pretrain TinyLlama
This tutorial will walk you through pretraining [TinyLlama](https://github.com/jzhang38/TinyLlama/).
> [!TIP]
> To get started with zero setup, clone the [TinyLlama studio on Lightning AI](https://lightning.ai/lightning-ai/studios/llm-pretrain-tinyllama-1-1b).
## What's TinyLlama?
[TinyLlama](https://github.com/jzhang38/TinyLlama/) is architecturally the same as Meta AI's LLama 2, but only has 1.1B parameters and is instead trained on multiple epochs on a mix of [SlimPajama](https://huggingface.co/datasets/cerebras/SlimPajama-627B) and [Starcoder](https://huggingface.co/datasets/bigcode/starcoderdata) datasets.
Here is a quick fact sheet:
| Name | Description |
|-------------------------------|--------------------------------------------------------------------------------------------------------------------------------------------------------------|
| Parameters | 1.1B |
| Model Size | Layers: 22, Heads: 32, Query Groups: 4, Embedding Size: 2048, Intermediate Size: 5632 |
| Sequence Length | 2048 |
| Learning Rate | 4e-4 |
| Learning Rate Schedule | Cosine with 2000 warmup steps |
| Training Data | [SlimPajama](https://huggingface.co/datasets/cerebras/slimpajama-627b) (893 GB), [Starcoder](https://huggingface.co/datasets/bigcode/starcoderdata) (290 GB) |
| Combined Dataset Size | Around 950B tokens |
| Total Tokens During Training | 3 trillion (3 epochs) |
| Time to complete training | ~ 4 weeks with 64 A100 GPUs |
| Model FLOPs Utilization (MFU) | 52% |
(this table was sourced from the author's [README](https://github.com/jzhang38/TinyLlama/))
## Download datasets
You can download the data using git lfs:
```bash
# Make sure you have git-lfs installed (https://git-lfs.com):
sudo apt install git-lfs
```
```bash
git clone https://huggingface.co/datasets/cerebras/slimpajama-627b data/slimpajama-raw
git clone https://huggingface.co/datasets/bigcode/starcoderdata data/starcoderdata-raw
```
Around 1.2 TB of disk space is required to store both datasets.
## Prepare the datasets for training
In order to start pretraining litgpt on it, you need to read, tokenize, and write the data in binary chunks. This will leverage the `litdata` optimization pipeline and streaming dataset.
First, install additional dependencies for preprocessing:
```bash
pip install '.[all]'
```
You will need to have the tokenizer config available:
```bash
litgpt download meta-llama/Llama-2-7b-hf \
--access_token your_hf_token \
--tokenizer_only true
```
Then, run the preprocessing script for each dataset and split.
You will require **1.1 TB** of disk space for Starcoder and **2.5** TB of space for the SlimPajama dataset.
**Starcoder:**
```bash
python litgpt/data/prepare_starcoder.py \
--input_dir data/starcoderdata-raw \
--output_dir data/starcoder \
--tokenizer_path checkpoints/meta-llama/Llama-2-7b-hf
```
**SlimPajama:**
```bash
python litgpt/data/prepare_slimpajama.py \
--input_dir data/slimpajama-raw/validation \
--output_dir data/slimpajama/val \
--tokenizer_path checkpoints/meta-llama/Llama-2-7b-hf
python litgpt/data/prepare_slimpajama.py \
--input_dir data/slimpajama-raw/test \
--output_dir data/slimpajama/test \
--tokenizer_path checkpoints/meta-llama/Llama-2-7b-hf
python litgpt/data/prepare_slimpajama.py \
--input_dir data/slimpajama-raw/train \
--output_dir data/slimpajama/train \
--tokenizer_path checkpoints/meta-llama/Llama-2-7b-hf
```
If you want to run on a small slice of the datasets first, pass the flag `--fast_dev_run=true` to the commands above.
In the above we are assuming that you will be using the same tokenizer as used in LlaMA/TinyLlama, but any trained [SentencePiece](https://github.com/google/sentencepiece) tokenizer with a 32000 vocabulary size will do here.
## Pretraining
Running the pretraining script with its default settings requires at least 8 A100 GPUs.
```bash
litgpt pretrain --config config_hub/pretrain/tinyllama.yaml
```
> [!TIP]
> Use the `litgpt pretrain --data.help TinyLlama` command to list additional dataset options.
The script will save checkpoints periodically to the folder `out/`.
By default, the `pretrain` script will pretrain the model with FSDP in
`bfloat16` mixed precision and gradient accumulation.
Note that `pretrain` is not actually a model-specific training script, so feel free [try other configurations](../config_hub)
or change the model type and size by passing a different string to the model name argument, for example:
```shell
litgpt pretrain Gemma-2b
```
The currently supported model names can be listed by executing `litgpt pretrain` without any additional arguments.
Keep in mind that training with a single machine will take weeks. To speed up the process, you'll need access to a cluster.
Once you're in a cluster, you can follow [these instructions](https://lightning.ai/docs/fabric/stable/fundamentals/launch.html#launch-on-a-cluster)
to launch the script across machines:
- [Lightning AI](https://lightning.ai/docs/fabric/stable/guide/multi_node/cloud.html)
- [SLURM cluster](https://lightning.ai/docs/fabric/stable/guide/multi_node/slurm.html)
- [Barebones cluster](https://lightning.ai/docs/fabric/stable/guide/multi_node/barebones.html)
- [MPI](https://lightning.ai/docs/fabric/stable/guide/multi_node/other.html)
The script exposes several hyperparameters you can tweak through the command line.
For instance, `--train.micro_batch_size` should be adjusted so the process will use the available
GPU memory. For more tips to avoid out-of-memory issues, please also see the more detailed
[Dealing with out-of-memory (OOM) errors](oom.md) guide.
Last, logging is kept minimal in the script, but for long-running experiments we recommend switching to a proper experiment tracker.
As an example, we included WandB (set `--logger_name=wandb`) to show how you can integrate any experiment tracking framework.
For reference, [here are the loss curves for our reproduction](https://api.wandb.ai/links/awaelchli/y7pzdpwy).
## Resume training
The checkpoints saved during pretraining contain all the information to resume if needed.
Simply rerun the script with the `--resume` argument added:
```bash
litgpt pretrain tiny-llama\
--config config_hub/pretrain/tinyllama.yaml \
--resume out/pretrain/tiny-llama/step-00060500
```
**Important:** Each checkpoint is a directory. Point to the directory, not the 'lit_model.pth' file inside of it.
> [!TIP]
> Use the `litgpt pretrain --data.help TinyLlama` command to list additional dataset options.
## Export checkpoints
After training is completed, you can convert the checkpoint to a format that can be loaded for evaluation, inference, finetuning etc.
```bash
litgpt convert_pretrained_checkpoint out/pretrain/tiny-llama/step-00060500 \
--output_dir checkpoints/tiny-llama/final
```
After conversion, the output folder will contain these files:
```
checkpoints/tiny-llama/final
├── model_config.yaml
├── lit_model.pth
├── tokenizer_config.json
├── tokenizer.json
└── tokenizer.model
```
You can then use this checkpoint folder to run [evaluation](evaluation.md), [inference](inference.md), [finetuning](finetune_lora.md) or [process the checkpoint further](convert_lit_models.md).
## Project templates
The following [Lightning Studio](https://lightning.ai/lightning-ai/studios) templates provide LitGPT pretraining projects in reproducible environments with multi-GPU and multi-node support:
| | |
|---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------|
| <p align="left">[Prepare the TinyLlama 1T token dataset](https://lightning.ai/lightning-ai/studios/prepare-the-tinyllama-1t-token-dataset) <br> [<img src="https://pl-public-data.s3.amazonaws.com/assets_litgpt/readme/3.webp" width="300"></p>](https://lightning.ai/lightning-ai/studios/prepare-the-tinyllama-1t-token-dataset) | [Pretrain LLMs - TinyLlama 1.1B](https://lightning.ai/lightning-ai/studios/pretrain-llms-tinyllama-1-1b) <br> <p align="left">[<img src="https://pl-public-data.s3.amazonaws.com/assets_litgpt/readme/4.webp" width="300"></p>](https://lightning.ai/lightning-ai/studios/pretrain-llms-tinyllama-1-1b) |
| [Continued Pretraining with TinyLlama 1.1B](https://lightning.ai/lightning-ai/studios/continued-pretraining-with-tinyllama-1-1b) <br> <p align="left">[<img src="https://pl-public-data.s3.amazonaws.com/assets_litgpt/readme/1.webp" width="300"></p>](https://lightning.ai/lightning-ai/studios/continued-pretraining-with-tinyllama-1-1b) | |
| |
|