posttrainllm/pace-intent-router-v8

Text Classification • Updated Jul 18 • 13

Note Released weights; 57.1% sealed V1 accuracy at 3.8ms. Rejected as the production winner.

posttrainllm/qwen3-4b-file-ops-distilled

Text Generation • 4B • Updated Jul 3 • 45

Note Routed specialist: file-operations hard gate improved from 58% to 100%; breadth regression is disclosed.

posttrainllm/qwen3-4b-rest-fused

Text Generation • 4B • Updated Jul 3 • 12 • 1

Note Research specialist: file-operations depth 100%, breadth 65% versus stock 59.6%; rerun a product-specific gate before wiring.

posttrainllm/qwen3-4b-multibackend-distilled

Text Generation • 4B • Updated Jul 3 • 12

Note Rejected negative-transfer artifact: depth 100%, breadth 31%; retained as a failed comparison, not a promoted specialist.

posttrainllm/vibethinker-3b-mlx

Text Generation • 3B • Updated Jul 3 • 22 • 1

Note Report-only MLX conversion of WeiboAI/VibeThinker-3B; a conversion artifact, not a product-specific specialist.

posttrainllm/vibethinker-3b-agentic-distilled

Text Generation • 3B • Updated Jul 3 • 65

Note Inconclusive weights archive with current evaluation missing; not a shipped specialist.