llm: enable flash attention by default

2024-06-08 22:55:22 -07:00
1 changed files with 1 additions and 0 deletions
--- a/envconfig/config.go
+++ b/envconfig/config.go
@ -90,6 +90,7 @@ func init() {
 	NumParallel = 1
 	MaxRunners = 1
 	MaxQueuedRequests = 512
+	FlashAttention = true

 	LoadConfig()
 }