Files
ollama/x
Jesse Gross 2e9d68dc38 mlxrunner: unify the MTP decode paths
Greedy is a special case of sampled decoding — at temperature 0 the sampler
yields a point mass, so rejection-sampling acceptance reduces to argmax-match
— so collapse the separate greedy, sampled, and serial paths into one. MTP
now honors any temperature, penalty, and top-k/p/min-p setting; logprobs
remain the only gated feature.
2026-06-22 15:25:45 -07:00
..
2026-06-01 10:44:21 -07:00