vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-10 17:14:36 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge commit '335eb04a91' into concedo_experimental

Browse source 
# Conflicts:
#	.github/workflows/build.yml
#	CONTRIBUTING.md
#	Makefile
#	docs/build.md
#	examples/llama.swiftui/llama.swiftui/UI/ContentView.swift
#	examples/run/run.cpp
#	ggml/CMakeLists.txt
#	ggml/src/ggml-cpu/CMakeLists.txt
#	ggml/src/ggml-cuda/CMakeLists.txt
#	ggml/src/ggml-musa/CMakeLists.txt
This commit is contained in:
Concedo 2025-02-24 11:55:14 +08:00
commit 159c47f0e6
20 changed files with 984 additions and 29 deletions
Download patch file Download diff file Expand all files Collapse all files

6
examples/llava/clip.cpp
View file

@ -2811,9 +2811,13 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
if (!ctx->has_glm_projector) {
struct ggml_tensor * patches = ggml_graph_get_tensor(gf, "patches");
// The patches vector is used to get rows to index into the embeds with;
// we should skip dim 0 only if we have CLS to avoid going out of bounds
// when retrieving the rows.
int patch_offset = ctx->has_class_embedding ? 1 : 0;
int* patches_data = (int*)malloc(ggml_nbytes(patches));
for (int i = 0; i < num_patches; i++) {
patches_data[i] = i + 1;
patches_data[i] = i + patch_offset;
}
ggml_backend_tensor_set(patches, patches_data, 0, ggml_nbytes(patches));
free(patches_data);