Dose-Response C3 (Seed 137): Original composition (~1.21% unsafe), full scale

Multi-seed training variant (seed=137) for robustness study across random seeds.

Condition C3 โ€” Original composition (~1.21% unsafe), full scale
Training set ~7.94M images
Part of a study with 4 seeds (137, 314, 789, 1331) for C0 and C3.
  • distributed/ โ€” FSDP base training shards
  • sft_distributed/ โ€” FSDP SFT shards (20K steps on Alchemist)
Downloads last month
17
Inference Providers NEW
This model isn't deployed by any Inference Provider. ๐Ÿ™‹ Ask for provider support