matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-15 18:27:08 +00:00
Code Issues Packages Projects Releases Wiki Activity

llama: update vendored code to commit 40c6d79f (#7875)

Browse Source
This commit is contained in:
Jeffrey Morgan
2024-12-10 19:21:34 -08:00
committed by GitHub
parent a37f4a86a7
commit 527cc97899
289 changed files with 58552 additions and 41806 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
llama/ggml-cuda/quantize.cu vendored
View File

@@ -1,5 +1,5 @@
/**
* llama.cpp - commit 3f1ae2e32cde00c39b96be6d01c2997c29bae555 - do not edit this file
* llama.cpp - commit 40c6d79fb52f995f47507fedfeaae2ac05d9b35c - do not edit this file
*
* MIT License
*
@@ -95,8 +95,8 @@ static __global__ void quantize_mmq_q8_1(
// Exchange max. abs. value between vals_per_scale/4 threads.
#pragma unroll
for (int mask = vals_per_scale/8; mask > 0; mask >>= 1) {
amax = fmaxf(amax, __shfl_xor_sync(0xFFFFFFFF, amax, mask, WARP_SIZE));
for (int offset = vals_per_scale/8; offset > 0; offset >>= 1) {
amax = fmaxf(amax, __shfl_xor_sync(0xFFFFFFFF, amax, offset, WARP_SIZE));
}
float sum;
@@ -105,8 +105,8 @@ static __global__ void quantize_mmq_q8_1(
// Exchange calculate sum across vals_per_sum/4 threads.
#pragma unroll
for (int mask = vals_per_sum/8; mask > 0; mask >>= 1) {
sum += __shfl_xor_sync(0xFFFFFFFF, sum, mask, WARP_SIZE);
for (int offset = vals_per_sum/8; offset > 0; offset >>= 1) {
sum += __shfl_xor_sync(0xFFFFFFFF, sum, offset, WARP_SIZE);
}
}