Downloads · 30 days
12
38% of all-time downloads
unconst/Affine-5czsc2fc98-r568-r252-odpo-hirank-longctx-midextra-merged
Affine-5czsc2fc98-r568-r252-odpo-hirank-longctx-midextra-merged is a machine learning model from unconst. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as apache-2.0.
- Base / parent king: unconst/Affine-5czsc2fc98-r252-merged @ b42d6245d77fe30885ea8a90387771e1bc465e0f (SN120 reign-33 crown). - Method: Offline DPO on teacher-anchored Reason pairs (chosen = higher lpC(yC|z)−lpC(yC|∅…
Downloads · 30 days
12
38% of all-time downloads
All-time downloads
32
Public
Parameters
35.1B
70.2 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors70.2 GB · 100%
From the Hugging Face model README
unconst/Affine-5czsc2fc98-r252-merged @ b42d6245d77fe30885ea8a90387771e1bc465e0f (SN120 reign-33 crown).lpC(y_C|z)−lpC(y_C|∅) thought). Optimized for Reason v3 (teacher-side only; no lpA / L1lift).dpo_duel_reason.jsonl), LongCtx filter; experiment
mining/experiments/r568-r252-offline-dpo-hialpha-hirank-lobeta-longctx-midextrasteps/.5e-6, LoRA r=64 / α=128, β=0.02, max_len=16384, max_steps=1800 (stopped at 312 after data exhaustion / train.done), epoch=1.mine-r226-marsplan-fullft-1 (brave-raven-a9, 8×B200) GPUs 6,7.r252 reign-33 pending after HF push (SKIP_LOCAL_TKC on brave; remote warm TKC). Decision rule: paired margin > max(2·SE, δ=0.002) and median thought ≥80 and B pass ≥0.30.This card is the training write-up required before any Stage-5 submit.