AnalysisDevelopers
Jun 19, 10:24 AM
llama.cpp PR adds Eagle3 support for Qwen3.5/3.6
A pull request in llama.cpp adds support for Eagle3 speculative decoding for Qwen3.5 and Qwen3.6 models. The PR aims to compare Eagle3 with the existing MTP method.
A pull request in llama.cpp adds support for Eagle3 speculative decoding for Qwen3.5 and Qwen3.6 models. The PR aims to compare Eagle3 with the existing MTP method.