mirror of
https://github.com/RYDE-WORK/llama.cpp.git
synced 2026-01-19 21:23:26 +08:00
Key changes: * BERT conversion: fix abuse of LlamaHfVocab, do not set BOS or EOS * Nomic Embed conversion: pad vocab instead of slicing embedding tensor * llama_tokenize: handle added special tokens like HF does