Kentucky-Open-Science/KOS-V5-Instruct
Update README.md
Add Retrieval & embeddings section + link to KOS-V5-Retriever adapter
Card: RESTORE full model card (logo, specs, pre/post-training, medical foundation, IFEval-in-context, quickstart, deployment) that a prior push had gutted; all numbers = leg-T step 40 on consistent official harnesses (IFEval 72.19 pristine, BFCL 85/84/80.5, peer 84.71)
Card: peer Qwen3-4B-2507 IFEval = 84.71 (exact, pristine-harness canonical value)
Card: restore full peer comparison (Qwen3-4B-2507) that was wrongly stripped; update KOS column to leg-T (BFCL now above peer, IFEval/knowledge honest vs peer)
Card: knowledge benchmarks re-measured on this checkpoint (MMLU 0.4512, PubMedQA 0.706, MedQA 0.3802, MedMCQA 0.3648) — jobs 202025/202026
KOS-V5-Instruct 2026-07-21: recover tool calling via official-BFCL GRPO (leg T, step 40)
Card: Catbird logo + intro, official BFCL results
Card: Catbird logo + intro, official BFCL results
KOS-V5-Instruct 2026-07-20: v2mix SFT + official-IFEval GRPO (leg A, step 164)
Card: Catbird logo + intro, official BFCL results
Card: Catbird logo + intro, official BFCL results
Card: Catbird logo + intro, official BFCL results
Card: Catbird logo + intro, official BFCL results
Add Catbird logo (matches KOS-V5-Base)
Card: tool/function calling recorded as NOT EVALUATED
KOS-V5-Instruct (Catbird) — IF-GRPO step 888
initial commit