saidutta69/RaceBench-v1.1
RaceBench v1.1 A curated SFT dataset blending agentic tool-use traces with high-quality distillation data — purpose-built for making small models (0.5B-3B) capable enough to replace API-based frontier models in edge deployments. Quality over quantity. Every row passed a quality threshold of >=60/100. v1.1 fixes the v1.0 agent dilution bug and upgrades to premium traces. Dataset Composition Blended from two source datasets: saidutta69/fable-5-premium —… See the full description on the dataset page: https://huggingface.co/datasets/saidutta69/RaceBench-v1.1.
Use self-hosted banner image
add: quality_distribution.png (copied from v1, v1.1 overall mean 60.8)
fix: restore branded banner, quality distribution, Why RaceBench, citation - match v1 format
v1.1 README: Option B quality-expanded (premium >=0.8)
v1.1 Option B: 376,736 rows (353,456 distill + 23,280 premium >=0.8 x4) - fixes agent dilution, upgrades to premium
initial commit