CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-10-31 08:51:55 +00:00
Code Issues Packages Projects Releases Wiki Activity

clip : fix visual encoders with no CLS (#11982)

Browse Source 
Signed-off-by: Alex-Brooks <Alex.Brooks@ibm.com>
This commit is contained in:
Alex Brooks
2025-02-20 23:11:03 -07:00
committed by GitHub
parent c392e5094d
commit ee02ad02c5
1 changed files with 5 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
examples/llava/clip.cpp
View File

@@ -2712,9 +2712,13 @@ 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);