Text-to-Image
diffusion
safety
dose-response

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

Part of a dose-response experiment studying how unsafe training data fraction affects text-to-image model safety.

Architecture PRX-1.2B
Condition C3 — Original composition (~1.21% unsafe), full scale
Training set ~7.94M images
Original dataset composition with no filtering. Contains the natural ~1.21% unsafe rate (96K unsafe in 7.94M total).

Files

  • denoiser.pt — Consolidated checkpoint (EMA weights)
  • distributed/ — FSDP distributed shards
  • config.yaml — Training configuration
Downloads last month
15
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for anonym371/dose-response-c3

Finetunes
1 model

Datasets used to train anonym371/dose-response-c3