vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2026-05-08 18:30:50 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 3

emergency fix for q25vl

Browse source
This commit is contained in:
Concedo 2025-04-27 16:46:33 +08:00
parent 1b0481f4b1
commit f8b7ddeac0
1 changed files with 5 additions and 4 deletions
Download patch file Download diff file Expand all files Collapse all files

9
examples/llava/clip.cpp
View file

@ -1743,7 +1743,8 @@ struct clip_model_loader {
if (ctx_clip.proj_type == PROJECTOR_TYPE_MINICPMV
|| ctx_clip.proj_type == PROJECTOR_TYPE_GLM_EDGE
|| ctx_clip.proj_type == PROJECTOR_TYPE_QWEN2VL) {
|| ctx_clip.proj_type == PROJECTOR_TYPE_QWEN2VL
|| ctx_clip.proj_type == PROJECTOR_TYPE_QWEN25VL) {
n_layer += 1;
}
@ -2856,7 +2857,7 @@ bool clip_image_preprocess(struct clip_ctx * ctx, const clip_image_u8 * img, str
}
return true;
}
else if (ctx->proj_type == PROJECTOR_TYPE_QWEN2VL) {
else if (ctx->proj_type == PROJECTOR_TYPE_QWEN2VL || ctx->proj_type == PROJECTOR_TYPE_QWEN25VL) {
clip_image_u8 resized;
auto patch_size = clip_get_patch_size(ctx) * 2;
int nx = ceil((float)img->nx / patch_size) * patch_size;
@ -3255,7 +3256,7 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
else {
// non-minicpmv models
if (ctx->proj_type == PROJECTOR_TYPE_QWEN2VL) {
if (ctx->proj_type == PROJECTOR_TYPE_QWEN2VL || ctx->proj_type == PROJECTOR_TYPE_QWEN25VL) {
// pw * ph = number of tokens output by ViT after apply patch merger
// ipw * ipw = number of vision token been processed inside ViT
const int merge_ratio = 2;
@ -3395,7 +3396,7 @@ bool clip_image_batch_encode(clip_ctx * ctx, const int n_threads, const clip_ima
}
}
if (use_window_attn && ctx->proj_type == PROJECTOR_TYPE_QWEN25VL) {
if (use_window_attn && (ctx->proj_type == PROJECTOR_TYPE_QWEN2VL || ctx->proj_type == PROJECTOR_TYPE_QWEN25VL)) {
struct ggml_tensor * window_idx = ggml_graph_get_tensor(gf, "window_idx");
struct ggml_tensor * inv_window_idx = ggml_graph_get_tensor(gf, "inv_window_idx");
struct ggml_tensor * window_mask = ggml_graph_get_tensor(gf, "window_mask");