Hugging Face's logo
Activity Feed

AI & ML interests

None defined yet.

Recent Activity
cabbage972  updated a dataset 3 days ago
permissive-post-train/evaluation-records
cabbage972  updated a collection 4 days ago
Phase 2: reasoning SFT
cabbage972  updated a collection 4 days ago
Phase 2: reasoning SFT
View all activity

permissive-post-train 's collections 4

Mixture ablation, decontamination, 0.4B and masking
The 100B mixture ablation, MixtureVitae before and after decontamination, the 0.4B comparison, and SFT with prompt tokens masked.
Front-loading study
Four web datasets pretrained with and without MixtureVitae's instruction and reasoning subset, then extended to 16K and given Tulu3 SFT and DPO.
Phase 2: reasoning SFT
Reasoning SFT on 30K samples of OpenThoughts3 (OT30K) or another reasoning dataset, and the checkpoints each run starts from.
Phase 1: Tulu3 SFT and DPO
Eight 1.7B-scale bases, each also extended to 16K, then given Tulu3 SFT and DPO.
Mixture ablation, decontamination, 0.4B and masking
The 100B mixture ablation, MixtureVitae before and after decontamination, the 0.4B comparison, and SFT with prompt tokens masked.
Phase 2: reasoning SFT
Reasoning SFT on 30K samples of OpenThoughts3 (OT30K) or another reasoning dataset, and the checkpoints each run starts from.
Front-loading study
Four web datasets pretrained with and without MixtureVitae's instruction and reasoning subset, then extended to 16K and given Tulu3 SFT and DPO.
Phase 1: Tulu3 SFT and DPO
Eight 1.7B-scale bases, each also extended to 16K, then given Tulu3 SFT and DPO.