hugpadhugpad

ApocalypseParty / G4-31B-configCA-GRPO-r4b100

Other31B410Updated Jul 29, 2026
Hugging Face

Downloads

41

Likes

0

Last updated

Jul 29, 2026

License

Not available

G4-31B-configCA with a GRPO LoRA (diversity training, run 4b step 100) merged in. Reduces story-attractor collapse on underspecified creative prompts while keeping coherence. Same chat template and usage as the base model.

Tags

  • gemma4

Technical specifications

Parameters31B
LibraryNot available
LanguagesNot available
LicenseNot available