From 1f4977c1d14a554ee662905878d14d2b405286eb Mon Sep 17 00:00:00 2001 From: Akarshan Date: Sat, 18 Oct 2025 13:36:03 +0530 Subject: [PATCH] fix mmap settings and adjust flash attention --- extensions/llamacpp-extension/src/index.ts | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/extensions/llamacpp-extension/src/index.ts b/extensions/llamacpp-extension/src/index.ts index 9bb3e6559..e9bddb78f 100644 --- a/extensions/llamacpp-extension/src/index.ts +++ b/extensions/llamacpp-extension/src/index.ts @@ -1649,12 +1649,12 @@ export default class llamacpp_extension extends AIEngine { if (cfg.main_gpu !== undefined && cfg.main_gpu != 0) args.push('--main-gpu', String(cfg.main_gpu)) // Note: Older llama.cpp versions are no longer supported - if (cfg.flash_attn !== undefined || cfg.flash_attn !== '') args.push('--flash-attn', String(cfg.flash_attn)) //default: auto = ON when supported + if (cfg.flash_attn !== undefined || !cfg.flash_attn || cfg.flash_attn !== '') args.push('--flash-attn', String(cfg.flash_attn)) //default: auto = ON when supported // Boolean flags if (cfg.ctx_shift) args.push('--context-shift') if (cfg.cont_batching) args.push('--cont-batching') - args.push('--no-mmap') + if (cfg.no_mmap) args.push('--no-mmap') if (cfg.mlock) args.push('--mlock') if (cfg.no_kv_offload) args.push('--no-kv-offload') if (isEmbedding) {