happyz
/
llama.cpp
mirror of https://github.com/ggerganov/llama.cpp.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

fix: use tensor stride for fused QKV support in vaetki

This commit is contained in:
suhyun-hwang 2026-01-11 18:56:50 +09:00
parent c947c74a4c
commit 8bbeab0616
2 changed files with 4 additions and 17 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

14
convert_hf_to_gguf.py
View File

@ -7969,20 +7969,6 @@ class VaetkiVisionModel(MmprojModel):
if name.startswith("model.visual."): if name.startswith("model.visual."):
name = name.replace("model.visual.", "visual.") name = name.replace("model.visual.", "visual.")
# Split fused QKV tensors (build_vit fused QKV doesn't work for VAETKI)
if ".qkv." in name:
if data_torch.ndim == 2:
c3, _ = data_torch.shape
else:
c3 = data_torch.shape[0]
assert c3 % 3 == 0
c = c3 // 3
return [
(self.map_tensor_name(name.replace("qkv", "q")), data_torch[:c]),
(self.map_tensor_name(name.replace("qkv", "k")), data_torch[c:c * 2]),
(self.map_tensor_name(name.replace("qkv", "v")), data_torch[c * 2:]),
]
return [(self.map_tensor_name(name), data_torch)] return [(self.map_tensor_name(name), data_torch)]

7
tools/mtmd/models/vaetki.cpp
View File

@ -33,13 +33,14 @@ ggml_cgraph * clip_graph_vaetki::build() {
auto add_pos = [&](ggml_tensor * cur, const clip_layer &) -> ggml_tensor * { auto add_pos = [&](ggml_tensor * cur, const clip_layer &) -> ggml_tensor * {
// split CLS and patch tokens // split CLS and patch tokens
// use cur->nb[2] to support both fused QKV (nb[2]=3*n_embd) and separate Q/K/V (nb[2]=n_embd)
ggml_tensor * cur_cls = ggml_view_3d(ctx0, cur, d_head, n_head, 1, ggml_tensor * cur_cls = ggml_view_3d(ctx0, cur, d_head, n_head, 1,
ggml_row_size(cur->type, d_head), ggml_row_size(cur->type, d_head),
ggml_row_size(cur->type, d_head * n_head), 0); cur->nb[2], 0);
ggml_tensor * cur_patch = ggml_view_3d(ctx0, cur, d_head, n_head, n_pos_patches, ggml_tensor * cur_patch = ggml_view_3d(ctx0, cur, d_head, n_head, n_pos_patches,
ggml_row_size(cur->type, d_head), ggml_row_size(cur->type, d_head),
ggml_row_size(cur->type, d_head * n_head), cur->nb[2],
ggml_row_size(cur->type, d_head * n_head)); cur->nb[2]);
// apply RoPE to CLS token using class_pos_emb // apply RoPE to CLS token using class_pos_emb
if (cls_cos && cls_sin) { if (cls_cos && cls_sin) {