Commit History
-
Use pairwise batch for one or two hypotheses
Verified signatureAvoid shared-prefix prefill and cache branching for small option sets. Keep shared-prefix inference for three or more hypotheses and test the routing.c959f4d -
Share Qwen3.5 prefix cache across NLI hypotheses
Verified signatureAdd batched cached continuation for hypotheses sharing one premise. Qwen3.5 recurrent linear-attention state and full-attention KV cache are branched per hypothesis after one prefix prefill. Rerank uses the new path; tiny hybrid-model tests compare against independent pair inference and check branch isolation.e93f6db
