fsid_non_clustered_0.05_threshold_olmo-7b-unlearnt

SimNPO-unlearned checkpoint of allenai/Olmo-3-1025-7B.

Method: SimNPO (Simple Noise-contrastive Policy Optimization)
Forget set threshold: 0.05 memorization fraction
Training: FSDP (2x A100-80GB), gradient checkpointing, 1 epoch

See unlearning-cleanslate/fsid_olmo-7b for evaluation artifacts, scores, and validation results.

Downloads last month
5
Safetensors
Model size
4B params
Tensor type
F32
·
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Model tree for unlearning-cleanslate/fsid_non_clustered_0.05_threshold_olmo-7b-unlearnt

Finetuned
(66)
this model