unconst/Affine-5czsc2fc98-r596-r252-odpo-hirank-midbeta-softctx-megaextra-merged

TEXT GENERATIONConcurrent Unit Cost:3Model Size:35.1BQuant:FP8Context Size:32kTool Calling:SupportedPublished:Aug 17, 2026Architecture:Transformer Featherless Exclusive Cold

unconst/Affine-5czsc2fc98-r596-r252-odpo-hirank-midbeta-softctx-megaextra-merged is a 35.1 billion parameter Affine SN120 challenger model, developed by unconst, specifically trained using offline DPO to optimize for higher teacher-side Reason scores. This model is a specialized checkpoint derived from unconst/Affine-5czsc2fc98-r252-merged, focusing on improving reasoning capabilities. It is intended for specialized Affine miner submissions and evaluation server Reason duels, rather than general chat applications.

Loading preview...

Overview

This model, unconst/Affine-5czsc2fc98-r596-r252-odpo-hirank-midbeta-softctx-megaextra-merged, is a 35.1 billion parameter Affine SN120 challenger. It was developed by unconst and trained using an offline DPO (Direct Preference Optimization) method, specifically targeting improved "Reason v3" scores based on teacher-anchored preferences. The training utilized a "SoftCtx × HiRank" pair set with a MidBeta β=0.1 configuration, building upon the unconst/Affine-5czsc2fc98-r252-merged base model.

Key Training Details

  • Base Model: unconst/Affine-5czsc2fc98-r252-merged
  • Methodology: Offline DPO, optimizing for higher teacher-side Reason scores on mined pairs.
  • Data: SoftCtx × HiRank pair set with MidBeta β=0.1.
  • Hyperparameters: LoRA r=64, α=128; β=0.1; lr=5e-6; max_len=12288.
  • Performance: Achieved a margin of +0.006196 against the live king reign34 in local n80 evaluations, with a z-score of 2.63 over 75 samples.

Intended Use

This model is highly specialized and is primarily intended for:

  • Affine miner submissions (SN120).
  • Evaluation server Reason duels.

It is not designed or optimized as a general-purpose chat model.

Finetunes
1 models