FED: Fast and Efficient Dataset Deduplication Framework with GPU Acceleration
This work addresses the bottleneck of slow data deduplication for researchers and practitioners training large language models, offering a significant but incremental improvement over existing GPU methods.
The paper tackled the problem of inefficient dataset deduplication for large language models by proposing FED, a GPU-accelerated framework that optimizes MinHash LSH, resulting in speed-ups of up to 107.2 times over CPU-based tools and 6.3 times over GPU-based tools while maintaining high deduplication quality with Jaccard similarity over 0.96.
Dataset deduplication plays a crucial role in enhancing data quality, ultimately improving the training performance and efficiency of large language models. A commonly used method for data deduplication is the MinHash LSH algorithm. Recently, NVIDIA introduced a GPU-based MinHash LSH deduplication method, but it remains suboptimal, leaving room for further improvement in processing efficiency. This paper proposes a GPU-accelerated deduplication framework, FED, that optimizes MinHash LSH for GPU clusters and leverages computationally efficient, partially reusable non-cryptographic hash functions. FED significantly outperforms the CPU-based deduplication tool in SlimPajama (using 64 logical CPU cores) by up to 107.2 times and the GPU-based tool in NVIDIA NeMo Curator by up to 6.3 times when processing 30 million documents on a node with four GPUs. Notably, our method dramatically accelerates the previously time-consuming MinHash signature generation phase, achieving speed-ups of up to 260 compared to the CPU baseline. Despite these gains in efficiency, FED maintains high deduplication quality, with the duplicate document sets reaching a Jaccard similarity of over 0.96 compared to those identified by the standard MinHash algorithm. In large-scale experiments, the deduplication of 1.2 trillion tokens is completed in just 6 hours in a four-node, 16-GPU environment. The related code is publicly available on GitHub (\href{https://github.com/mcrl/FED}{https://github.com/mcrl/FED}).