thek0tyara/llama-cpp-turboquant
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions 4

metal : FA support F32 K and V and head size = 32 (#16531)

Browse source 
* metal : FA support F32 K and V and head size = 32

* graph : remove obsolete comment [no ci]
This commit is contained in:
Georgi Gerganov 2025-10-13 23:07:57 +03:00 • committed by GitHub
parent e38b7c6e9e
commit e60f241eac
No known key found for this signature in database
GPG key ID: B5690EEEBB952194
4 changed files with 106 additions and 52 deletions
Download patch file Download diff file Expand all files Collapse all files

2
tests/test-backend-ops.cpp
View file

@ -6779,7 +6779,7 @@ static std::vector<std::unique_ptr<test_case>> make_test_cases_eval() {
for (int nb : { 1, 3, 32, 35, }) {
for (ggml_prec prec : {GGML_PREC_F32, GGML_PREC_DEFAULT}) {
if (hsk != 128 && prec == GGML_PREC_DEFAULT) continue;
for (ggml_type type_KV : {GGML_TYPE_F16, GGML_TYPE_BF16, GGML_TYPE_Q8_0, GGML_TYPE_Q4_0}) {
for (ggml_type type_KV : {GGML_TYPE_F32, GGML_TYPE_F16, GGML_TYPE_BF16, GGML_TYPE_Q8_0, GGML_TYPE_Q4_0}) {
test_cases.emplace_back(new test_flash_attn_ext(
hsk, hsv, nh, {nr2, nr3}, kv, nb, mask, sinks, max_bias, logit_softcap, prec, type_KV));
// run fewer test cases permuted