NYU-MLDA/OpenABC

The dataset comprises 29 open-source hardware IPs synthesized through 1,500 random synthesis flows using yosys-abc, capturing initial, intermediate, and final AIGs with labels including graph statistics, atomic transformation sequences, and post-mapping area/delay metrics. Data is pre-converted to PyTorch-Geometric format for direct ML consumption, eliminating costly preprocessing while supporting QoR prediction, area/delay estimation, and self-supervised learning tasks. The complete dataset (1.4TB across 14 chunks) and ML-ready variant (19GB) are hosted separately, with the generation pipeline automatable via provided Python utilities for synthesis scripting, AIG-to-GraphML conversion, and PyG serialization.

144 stars. No commits in the last 6 months.