matt/ollama37
1
0
Fork 0
mirror of https://github.com/dogkeeper886/ollama37.git synced 2025-12-19 20:27:01 +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

12
llama/ggml-cuda/fattn.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
*
@@ -39,9 +39,9 @@ static void ggml_cuda_flash_attn_ext_wmma_f16(ggml_backend_cuda_context & ctx, g
const ggml_tensor * KQV = dst;
const ggml_tensor * Q = dst->src[0];
const int32_t precision = KQV->op_params[3];
const enum ggml_prec prec = ggml_flash_attn_ext_get_prec(KQV);
if (precision != GGML_PREC_DEFAULT) {
if (prec != GGML_PREC_DEFAULT) {
if (Q->ne[1] <= 32 || Q->ne[0] > 128) {
constexpr int cols_per_block = 16;
switch (Q->ne[0]) {
@@ -327,11 +327,11 @@ void ggml_cuda_flash_attn_ext(ggml_backend_cuda_context & ctx, ggml_tensor * dst
ggml_cuda_set_device(ctx.device);
const int cc = ggml_cuda_info().devices[ggml_cuda_get_device()].cc;
const int32_t precision = KQV->op_params[3];
const enum ggml_prec prec = ggml_flash_attn_ext_get_prec(KQV);
// On AMD the tile kernels perform poorly, use the vec kernel instead:
if (cc >= CC_OFFSET_AMD) {
if (precision == GGML_PREC_DEFAULT && fast_fp16_available(cc)) {
if (prec == GGML_PREC_DEFAULT && fast_fp16_available(cc)) {
ggml_cuda_flash_attn_ext_vec_f16(ctx, dst);
} else {
ggml_cuda_flash_attn_ext_vec_f32(ctx, dst);
@@ -358,7 +358,7 @@ void ggml_cuda_flash_attn_ext(ggml_backend_cuda_context & ctx, ggml_tensor * dst
}
if (Q->ne[1] == 1 && Q->ne[0] % (2*WARP_SIZE) == 0) {
if (precision == GGML_PREC_DEFAULT) {
if (prec == GGML_PREC_DEFAULT) {
ggml_cuda_flash_attn_ext_vec_f16(ctx, dst);
return;
} else if(Q->ne[0] <= 128) {