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

kompute : fix basic f16 get_rows, 28 -> 26 failures

Browse source
This commit is contained in:
Jared Van Bortel 2024-01-25 15:22:11 -05:00
parent 11b305082b
commit de9fba0d39
1 changed files with 1 additions and 1 deletions
Download patch file Download diff file Expand all files Collapse all files

2
kompute-shaders/op_getrows_f16.comp
View file

@ -27,5 +27,5 @@ void main() {
const uint i = gl_WorkGroupID.x;
const int r = inB[i + pcs.inBOff];
dequantize_row_f16(r*pcs.nb01/2/*bytes for float16*/ + pcs.inAOff, i*pcs.nb1 + pcs.outOff, pcs.ne00);
dequantize_row_f16(r*pcs.nb01/2/*bytes for float16*/ + pcs.inAOff, i*pcs.nb1/4 + pcs.outOff, pcs.ne00);
}