Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-03 19:42:14 +02:00
Code Issues Packages Projects Releases Wiki Activity

adjusted layer estimation

Browse Source
This commit is contained in:
Concedo
2024-07-24 22:51:02 +08:00
parent cca2fa9a6c
commit d1f7832d21
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
koboldcpp.py
+1 -1
View File
@@ -612,7 +612,7 @@ def autoset_gpu_layers(filepath,ctxsize,gpumem): #shitty algo to determine how m
headcount = ggufmeta[1]
headkvlen = (ggufmeta[2] if ggufmeta[2] > 0 else 128)
ratio = mem/(fsize*csmul*1.5)
computemem = layers*4*headkvlen*cs*4*1.35 # For now the first 4 is the hardcoded result for a blasbatchsize of 512. Ideally we automatically calculate blasbatchsize / 4 but I couldn't easily grab the value yet - Henk
computemem = layers*4*headkvlen*cs*4*1.4 # For now the first 4 is the hardcoded result for a blasbatchsize of 512. Ideally we automatically calculate blasbatchsize / 4 but I couldn't easily grab the value yet - Henk
contextmem = layers*headcount*headkvlen*cs*4
reservedmem = 1.5*1024*1024*1024 # Users often don't have their GPU's VRAM worth of memory, we assume 500MB to avoid driver swapping + 500MB for the OS + 500MB for background apps / browser - Henk
if headcount > 0: