ggml-org/llama.cpp

server: avoid full prompt eval when 'prompt >= ctx'

已關閉

#6,855 建立於 2024年4月23日

 (3 則留言) (0 個反應) (0 位負責人)C++ (21,737 個分叉)batch import
enhancementgood first issue

倉庫指標

星標
 (124,043 顆星)
PR 合併指標
 (平均合併 6天 8小時) (30 天內合併 389 個 PR)

描述

When using the server for multi-turn chat, soon or later the prompt is going to surpass the context size, the current approach truncate the prompt by half of the context size excluding n_keep:

https://github.com/ggerganov/llama.cpp/blob/192090bae47960f0d38d4967abe398a5d190057e/examples/server/server.cpp#L1969-L1983

By doing that, common_part is going to match only n_keep tokens (when cache_prompt: true):

https://github.com/ggerganov/llama.cpp/blob/192090bae47960f0d38d4967abe398a5d190057e/examples/server/server.cpp#L2011-L2016

Technically, this is not a full prompt eval, n_keep is not revaluated, but it would be better to avoid this if possible, specially because prompt eval is slow on CPU.

貢獻者指南