Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-04 03:51:17 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge commit '34af94cd9ab277632e27caeec2d41de2fd091b31' into concedo_experimental

Browse Source 
# Conflicts:
#	.github/workflows/docker.yml
#	.github/workflows/make-release.yml
#	.github/workflows/release.yml
#	.pi/gg/SYSTEM.md
#	CMakeLists.txt
#	build-xcframework.sh
#	docs/development/HOWTO-add-model.md
#	docs/ops.md
#	docs/ops/SYCL.csv
#	docs/speculative.md
#	examples/sycl/update-ops-doc.sh
#	ggml/CMakeLists.txt
#	ggml/src/ggml-sycl/cpy.cpp
#	ggml/src/ggml-sycl/ggml-sycl.cpp
#	scripts/make-release-checks.sh
#	scripts/sync-ggml.last
#	tests/test-chat-auto-parser.cpp
#	tests/test-chat.cpp
#	tests/test-jinja.cpp
#	tests/test-llama-archs.cpp
#	tests/testing.h
#	tools/llama-bench/llama-bench.cpp
#	tools/server/README.md
This commit is contained in:
Concedo
2026-08-20 17:08:46 +08:00
parent 1fc6e11ab8 34af94cd9a
commit f0cd0225aa
135 changed files with 3116 additions and 312 deletions
Download Patch File Download Diff File Expand all files Collapse all files
src/llama-quant.cpp
+6 -1
View File
@@ -476,7 +476,12 @@ static ggml_type llama_tensor_get_type_impl(quantize_state_impl & qs, ggml_type
} else if (ftype == LLAMA_FTYPE_MOSTLY_MXFP4_MOE) {
// MoE tensors -> MXFP4
// other tensors -> Q8_0
if (tensor->ne[2] > 1) {
// MLA projection tensors are also 3D, so match expert tensor roles explicitly.
const bool is_bailingmoe3_expert = arch == LLM_ARCH_BAILINGMOE3 &&
(category == tensor_category::FFN_UP ||
category == tensor_category::FFN_GATE ||
category == tensor_category::FFN_DOWN);
if (tensor->ne[2] > 1 && (arch != LLM_ARCH_BAILINGMOE3 || is_bailingmoe3_expert)) {
new_type = GGML_TYPE_MXFP4;
} else {
new_type = GGML_TYPE_Q8_0;