metadata
license: apache-2.0
datasets:
- Quest-AI/quest-corruption-truncated4grpo-2k-dataset-v1
base_model:
- Quest-AI/quest-corruption-14b-s110-r3
Alternative version of my prior 14b GRPO attempt with a higher learning rate, on a limited selection of data (2k entries instead of 6k).
Wandb with hparams and code in the files section can be found here:
https://wandb.ai/kalomaze/verifiers-examples/runs/mayn2ctv?nw=nwuserkalomaze