GPT

A Generative Pre-trained Transformer (GPT) trained on the Openwebtext dataset. The default implementation uses r50k_base BPE tokenization with a network architecture similar to OpenAI's GPT series but can easily be customized and scaled up or down to meet your needs and compute budget with some parameter adjustments. In addition, you may incorporate your own training data alongside Openwebtext for additional pre-training samples or for fine-tuning for a specific task after pre-training. It also supports PyTorch's built-in Distributed Data Parallel (DDP) protocol with ZeRO Redundancy sharding for efficient training over multiple CUDA-enabled GPU clusters.

Download the Repository

Clone the project locally using git:

git clone https://github.com/andrewdalpino/GPT

Requirements

• Python 3.10 or later
• A CUDA-enabled GPU with 12G of VRAM or more

Install Project Dependencies

Project dependencies are specified in the requirements.txt file. You can install them with pip using the following command from the project root. I recommend using a virtual environment such as venv to keep package dependencies on your system tidy.

python -m venv ./.venv

source ./.venv/bin/activate

pip install -r requirements.txt

Quick Start

If you'd just like to start training right away, the default settings should work on most single-GPU systems with 12G of VRAM or more.

python pre-train.py

Note that it will take a while to download and pre-process the dataset the first time that the training script is run.

If you have a larger system you can increase the training load by increasing the capacity of the network and batch_size at runtime.

python pre-train.py --embedding_dimensions=1024 --num_hidden_layers=24 --batch_size=8

To distribute the training workload over a cluster of GPUs or multiple cluster nodes, use PyTorch's torchrun extension to launch a distributed data parallel session.

torchrun --standalone --nnodes=1 --nproc-per-node=8 pre-train.py --batch_size=16 --gradient_accumulation_steps=32

Note that when training in data-parallel mode it's important that the gradient_accumulation_steps divides evenly into the world size for maximum performance. For example, if we have an 8 GPU cluster, we could perform 32 gradient accumulation steps in exactly 4 passes over the network.

After training, you can generate text from the model by running the generate.py script from the commandline with a prompt.

python generate.py

Pre-training Arguments

Argument	Default	Type	Description
--batch_size	1	int	The number of samples to pass through the network at a time.
--gradient_accumulation_steps	128	int	The number of batches to pass through the network before updating the weights.
--samples_per_epoch	4096	int	The number of training samples to pass through the network every epoch.
--learning_rate	5e-4	float	The global step size taken after every gradient accumulation step.
--max_gradient_norm	1.0	float	Clip gradients above this threshold before stepping.
--num_epochs	2145	int	The number of epochs to train for.
--eval_interval	10	int	Evaluate the model after this many epochs on the testing set.
--block_size	1024	int	The number of tokens within the context window for every sample.
--embedding_dimensions	1024	int	The dimensionality of the token embeddings.
--num_attention_heads	16	int	The number of attention heads within every block.
--num_hidden_layers	24	int	The number of attention/MLP blocks within the hidden layer of the network.
--dropout	0.1	float	The proportion of signals to send to zero during training as regularization.
--activation_checkpointing	False	bool	Should we use activation checkpointing?
--checkpoint_interval	20	int	Save the model parameters to disk every this many epochs.
--checkpoint_path	"./out/checkpoint.pt"	string	The path to the checkpoint file on disk.
--dataset_path	"./dataset"	string	The path to the dataset files on disk.
--num_dataset_processes	8	int	The number of processes (CPUs) to use to process the dataset.
--resume	False	bool	Should we resume training from the last checkpoint?
--device	"cuda"	string	The device to run the computation on.
--seed	None	int	The seed for the random number generator.

Instruction-tuning Arguments

Argument	Default	Type	Description
--base_model_path	"./out/checkpoint.pt"	string	The path to the pre-trained model.
--batch_size	1	int	The number of samples to pass through the network at a time.
--gradient_accumulation_steps	128	int	The number of batches to pass through the network before updating the weights.
--learning_rate	5e-4	float	The global step size taken after every gradient accumulation step.
--mask_input	False	bool	Should we mask the input part of the sample i.e. only train on the output?
--rank	8	int	The rank of the LoRA decomposition matrices.
--alpha	1.0	float	The strength of the LoRA signal.
--dropout	0.05	float	The proportion of signals to send to zero during training as regularization.
--num_epochs	4	int	The number of epochs to train for.
--eval_interval	1	int	Evaluate the model after this many epochs on the testing set.
--checkpoint_interval	1	int	Save the model parameters to disk every this many epochs.
--checkpoint_path	"./out/lora_instruction.pt"	string	The path to the checkpoint file on disk.
--resume	False	bool	Should we resume training from the last checkpoint?
--device	"cuda"	string	The device to run the computation on.
--seed	None	int	The seed for the random number generator.

Generation Arguments

Argument	Default	Type	Description
--checkpoint_path	"./out/checkpoint.pt"	string	The path to the checkpoint file on disk.
--lora_path	None	string	The path to the LoRA checkpoint.
--max_tokens	500	int	The maximum number of tokens that the model should generate per sample.
--temperature	1.0	float	The amount of regularization applied to the candidate token probabilities.
--top_k	500	int	Only sample from this many candidate tokens with the highest probabilities.
--top_p	0.9	float	Of the top_k tokens, drop all but the top_p portion of the cumulative probability distribution.
--device	"cuda"	string	The device to run the computation on.
--seed	None	int	The seed for the random number generator.

Beam Search Arguments

Argument	Default	Type	Description
--checkpoint_path	"./out/checkpoint.pt"	string	The path to the checkpoint file on disk.
--lora_path	None	string	The path to the LoRA checkpoint.
--max_tokens	200	int	The maximum number of tokens that the model should generate per sample.
--num_candidates	3	int	The number of candidate sequences to output.
--beam_width	16	int	The number of candidate sequences to keep track of during search.
--device	"cuda"	string	The device to run the computation on.
--seed	None	int	The seed for the random number generator.

References:

• A. Radford, et al. Language Models are Unsupervised Multitask Learners, OpenAI, 2019.
• T. Brown, et al. Language Models are Few-Shot Learners. OpenAI, 2020.
• A. Kazemnejad, et al. The Impact of Positional Encoding on Length Generalization in Transformers, 37th Conference on Neural Information Processing Systems (NeurIPS 2023).
• S. Rajbhandari, et al. ZeRO: Memory Optimizations Toward Training Trillion Parameter Models, 2020.
• J. R. Hermans, et al. Accumulated Gradient Normalization, JMLR: Workshop and Conference Proceedings, 2017.
• T. Chen, et al. Training Deep Nets with Sublinear Memory Cost. MIT, 2019.
• B. Zhang, et al. Root Mean Square Layer Normalization. 33rd Conference on Neural Information Processing Systems, NeurIPS 2019.

License

The code is licensed MIT and the tutorial is licensed CC BY-NC 4.0.