mirror of
https://github.com/ggml-org/llama.cpp.git
synced 2026-09-02 02:51:08 +02:00
166fe29492
* qwen35: use post-norm hidden state for MTP * rename pre_norm to nextn * fix step35