Package Comparison
All packages benchmarked on the same dataset and capacity. SeqPacker provides 11 algorithms — competitors offer one each.
Key takeaway: All packages achieve similar packing efficiency (the math is the same). The difference is raw speed — SeqPacker's Rust core is 5–50× faster than any Python implementation.
Average Speedup vs Fastest Competitor
Average speedup across all 7 datasets at the selected capacity. This measures how consistently each SeqPacker algorithm outperforms the fastest Python package per dataset.
Efficiency Parity
Packing efficiency is determined by the algorithm, not the implementation language. All packages using the same strategy achieve identical results.
Python Packages: Head-to-Head
Zooming in on just the competitor packages vs SeqPacker's recommended default (OBFD). The speed gap is enormous.
Bottom line: If you're packing sequences for LLM training, SeqPacker gives you the same quality as Python packages at a fraction of the latency. Drop-in replacement via pip install seqpacker.