Modelpublic

caffeine-fighter/aikor-writing-assessment-slot1

sourceHugging Faceapache-2.0updated 1mo agoView on Hugging Face

0likes36downloads

Model Card

AIKOR writing-assessment — first HF submission candidate

Self-contained candidate for the first Hugging Face submission. The public repository root on main is caffeine-fighter/aikor-writing-assessment-slot1; no adapter path or revision is assumed.

—Training base: lokeshe09/Qwen3.6-27B-bnb-4bit at c9ba383a73c73d8125736f87bc2f5c6064ebb2e2
—Official upstream: Qwen/Qwen3.6-27B at 6a9e13bd6fc8f0983b9b99948120bc37f49c13e9
—Fusion: one rank-concat LoRA delta 0.8 * Delta_A + 0.2 * Delta_B, merged into the exact training-base bytes
—Context contract: 32768 tokens
—Required API: /health, /v1/models, /v1/chat/completions

The weight-delta fusion is not identical to averaging nonlinear A/B predictions. Official validation was report-only and was not used to tune the fusion. The repository is published for the first submission candidate; the official competition submission itself has not been performed.