apus-ailab/APUS-Reflex-4B
Text Generation • 5B • Updated • 1.51k • 4
Language-native decision models: 4B, 9B, and 35B-A3B. Standalone BF16 releases with candidate scoring, training provenance, and evaluation evidence.
Note 4B · checkpoint-5949 · standalone BF16 · native 16/32-layer decision runtime.
Note Standalone BF16 · checkpoint-3000 · training provenance and evaluation evidence included.
Note Standalone BF16 · checkpoint-5949 · training provenance and evaluation evidence included.