Public/koboldcpp
1
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-09-19 09:15:18 +02:00
Code Issues Packages Projects Releases Wiki Activity

Merge commit '8a70973557ff2605dada63f0dd2c08aa1cfcceb2' into concedo_experimental

Browse Source 
# Conflicts:
#	docs/backend/CANN.md
#	docs/backend/SYCL.md
#	examples/model-conversion/scripts/utils/tensor-info.py
#	ggml/src/ggml-opencl/ggml-opencl.cpp
#	ggml/src/ggml-opencl/kernels/expm1.cl
#	ggml/src/ggml-opencl/kernels/mean.cl
#	ggml/src/ggml-opencl/kernels/softplus.cl
#	ggml/src/ggml-opencl/kernels/sum_rows.cl
#	ggml/src/ggml-webgpu/ggml-webgpu-shader-lib.hpp
#	ggml/src/ggml-webgpu/ggml-webgpu.cpp
#	ggml/src/ggml-webgpu/wgsl-shaders/common_decls.tmpl
#	ggml/src/ggml-webgpu/wgsl-shaders/embed_wgsl.py
#	ggml/src/ggml-webgpu/wgsl-shaders/get_rows.wgsl
#	ggml/src/ggml-webgpu/wgsl-shaders/mul_mat.wgsl
#	ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_decls.tmpl
#	ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_reg_tile.wgsl
#	ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_subgroup_matrix.wgsl
#	ggml/src/ggml-webgpu/wgsl-shaders/mul_mat_vec.wgsl
#	ggml/src/ggml-webgpu/wgsl-shaders/scale.wgsl
#	tools/server/webui/src/lib/components/app/chat/ChatScreen/ChatScreen.svelte
This commit is contained in:
Concedo
2026-02-20 14:36:49 +08:00
parent 4115f1c54d 8a70973557
commit 9eb9e4eb83
81 changed files with 5979 additions and 4504 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tools/mtmd/clip.cpp
+11 -3
View File
@@ -391,9 +391,17 @@ ggml_tensor * clip_graph::build_vit(
/* nb2 */ cur->nb[1],
/* offset */ ggml_row_size(cur->type, 2 * n_embd));
// TODO: q/k norm requires row size == n_embd, while here it's d_head
// we can add support in the future if needed
GGML_ASSERT(layer.q_norm == nullptr && layer.k_norm == nullptr);
if (layer.q_norm) {
GGML_ASSERT(layer.q_norm->ne[0] == Qcur->ne[0]);
Qcur = build_norm(Qcur, layer.q_norm, NULL, norm_t, eps, il);
cb(Qcur, "Qcur_norm", il);
}
if (layer.k_norm) {
GGML_ASSERT(layer.k_norm->ne[0] == Kcur->ne[0]);
Kcur = build_norm(Kcur, layer.k_norm, NULL, norm_t, eps, il);
cb(Kcur, "Kcur_norm", il);
}
} else {
// separate q, k, v