tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 03:55:20 +00:00
Code Issues Packages Projects Releases Wiki Activity

mtmd : fix the calculation of n_tokens for smolvlm (#13381)

Browse Source 
Co-authored-by: Taichi Nishimura <Taichi.A.Nishimura@sony.com>
This commit is contained in:
welix
2025-05-08 22:03:53 +09:00
committed by GitHub
parent 6562e5a4d6
commit 0ccc121354
1 changed files with 1 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tools/mtmd/clip.cpp
View File

@ -3010,7 +3010,7 @@ int clip_n_output_tokens(const struct clip_ctx * ctx, struct clip_image_f32 * im
int n_per_side_2d_pool = n_per_side / params.proj_scale_factor;
n_patches = n_per_side_2d_pool * n_per_side_2d_pool;
} else if (ctx->proj_type == PROJECTOR_TYPE_IDEFICS3) {
n_patches /= params.proj_scale_factor;
n_patches /= (params.proj_scale_factor * params.proj_scale_factor);
} else if (ctx->proj_type == PROJECTOR_TYPE_PIXTRAL) {
int n_merge = params.spatial_merge_size;
int n_patches_x = img->nx / params.patch_size / (n_merge > 0 ? n_merge : 1);