vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama : disable FA for AMD

Browse source
This commit is contained in:
Georgi Gerganov 2024-04-24 16:48:10 +03:00
parent 8937ec5307
commit ce281b904c
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
3 changed files with 12 additions and 2 deletions
Download patch file Download diff file Expand all files Collapse all files

3
ggml-cuda/fattn.cu
View file

@ -2,7 +2,10 @@
#include "fattn.cuh"
#include <cstdint>
#if FP16_MMA_AVAILABLE
#include <mma.h>
#endif
#define FATTN_KQ_STRIDE 256
#define HALF_MAX_HALF __float2half(65504.0f/2) // Use neg. of this instead of -INFINITY to initialize KQ max vals to avoid NaN upon subtraction.