Lumpiasty/llama.cpp
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

ggml : update ggml_prec specification (#26675)

Browse Source 
* ggml : update ggml_prec specification

[no ci]

* cont : add GGML_PREC_BF16

* cont : rework API

* cont : use new API

* cont : swap arg order

* cont : support for MUL_MAT_ID

* cont : fix accidental remove of "break;"

* cont : return bools, add doc TAG_GGML_PREC, clean-up

* cont : add search tag

* cont : ws
This commit is contained in:
Georgi Gerganov
2026-09-08 09:06:24 +03:00
committed by GitHub
parent 9dcf84e5ae
commit 5a6caa05fc
10 changed files with 134 additions and 22 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/mtmd/clip.cpp
+2 -2
View File
@@ -780,7 +780,7 @@ ggml_tensor * clip_graph::build_attn(
}
cur = ggml_flash_attn_ext(ctx0, q, k, v, kq_mask, kq_scale, 0.0f, 0.0f);
ggml_flash_attn_ext_set_prec(cur, GGML_PREC_F32);
ggml_prec_set_acc(cur, GGML_PREC_F32);
if (sinks != nullptr) {
ggml_flash_attn_ext_add_sinks(cur, sinks);
}
@@ -793,7 +793,7 @@ ggml_tensor * clip_graph::build_attn(
ggml_tensor * kq = ggml_mul_mat(ctx0, k, q);
// F32 may not needed for vision encoders?
// ggml_mul_mat_set_prec(kq, GGML_PREC_F32);
// ggml_prec_set_acc(kq, GGML_PREC_F32);
kq = ggml_soft_max_ext(ctx0, kq, kq_mask, kq_scale, 0.0f);
if (sinks != nullptr) {
tools/mtmd/models/mimovl.cpp
+1 -1
View File
@@ -2,7 +2,7 @@
ggml_tensor * clip_graph_mimovl::build_mm(ggml_tensor * w, ggml_tensor * x) const {
ggml_tensor * cur = ggml_mul_mat(ctx0, w, x);
ggml_mul_mat_set_prec(cur, GGML_PREC_F32);
ggml_prec_set_acc(cur, GGML_PREC_F32);
return cur;
}
tools/mtmd/models/qwen3tts-spkenc.cpp
+1 -1
View File
@@ -27,7 +27,7 @@ ggml_tensor * clip_graph_qwen3tts_spkenc::conv1d_same(ggml_tensor * x, ggml_tens
ggml_tensor * w2d = ggml_reshape_2d(ctx0, w, (int64_t) K * IC, OC);
ggml_tensor * y = ggml_mul_mat(ctx0, w2d, col); // [OC, T_out]
ggml_mul_mat_set_prec(y, GGML_PREC_F32);
ggml_prec_set_acc(y, GGML_PREC_F32);
ggml_tensor * b2d = ggml_reshape_2d(ctx0, b, OC, 1);
y = ggml_add(ctx0, y, b2d);
tools/tuning/fa-vec.cpp
+1 -1
View File
@@ -56,7 +56,7 @@ static ggml_tensor * fa_build_graph(ggml_context * ctx, const fa_shape & s) {
ggml_set_name(m, "m");
ggml_tensor * out = ggml_flash_attn_ext(ctx, q, k, v, m, 1.0f / sqrtf((float) s.dk), 0.0f, 0.0f);
ggml_flash_attn_ext_set_prec(out, GGML_PREC_F32);
ggml_prec_set_acc(out, GGML_PREC_F32);
ggml_set_name(out, "out");
return out;