BTLM-3B-8K: 7B Parameter Performance in a 3B Parameter Model

TL;DR

BTLM-3B-8K is a new open-source 3 billion parameter language model that outperforms similar models, offers excellent long-context capabilities, and is optimized for low-resource environments, making powerful NLP more accessible.

Contribution

The paper introduces BTLM-3B-8K, a state-of-the-art 3B parameter language model with innovative training techniques and architecture optimizations that enable high performance and long-context understanding.

Findings

Outperforms existing 3B models by 2-5.5% on downstream tasks.

Competitive with some 7B parameter models.

Requires only 3GB memory with 4-bit precision, enabling deployment on edge devices.

Abstract

We introduce the Bittensor Language Model, called "BTLM-3B-8K", a new state-of-the-art 3 billion parameter open-source language model. BTLM-3B-8K was trained on 627B tokens from the SlimPajama dataset with a mixture of 2,048 and 8,192 context lengths. BTLM-3B-8K outperforms all existing 3B parameter models by 2-5.5% across downstream tasks. BTLM-3B-8K is even competitive with some 7B parameter models. Additionally, BTLM-3B-8K provides excellent long context performance, outperforming MPT-7B-8K and XGen-7B-8K on tasks up to 8,192 context length. We trained the model on a cleaned and deduplicated SlimPajama dataset; aggressively tuned the \textmu P hyperparameters and schedule; used ALiBi position embeddings; and adopted the SwiGLU nonlinearity. On Hugging Face, the most popular models have 7B parameters, indicating that users prefer the quality-size ratio of 7B models. Compacting the 7B parameter model to one with 3B parameters, with little performance impact, is an important milestone. BTLM-3B-8K needs only 3GB of memory with 4-bit precision and takes 2.5x less inference compute than 7B models, helping to open up access to a powerful language model on mobile and edge devices. BTLM-3B-8K is available under an Apache 2.0 license on Hugging Face: https://huggingface.co/cerebras/btlm-3b-8k-base.