vrr/koboldcpp
2
0
Fork 0
mirror of https://github.com/LostRuins/koboldcpp.git synced 2025-09-10 00:54:41 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

temporarily make qwenv2l use clip on cpu for vulkan and macos

Browse source
This commit is contained in:
Concedo 2024-12-21 09:15:31 +08:00
parent fc52a38a25
commit b7d3274523
3 changed files with 17 additions and 0 deletions
Download patch file Download diff file Expand all files Collapse all files

7
gpttype_adapter.cpp
View file

@ -2316,6 +2316,13 @@ ModelLoadResult gpttype_load_model(const load_model_inputs inputs, FileFormat in
if(mmproj_filename != "" && file_format==FileFormat::GGUF_GENERIC)
{
printf("\nAttempting to apply Multimodal Projector: %s\n", mmproj_filename.c_str());
#if defined(GGML_USE_VULKAN) || defined(GGML_USE_METAL)
if(file_format_meta.model_architecture == GGUFArch::ARCH_QWEN2VL)
{
set_clip_uses_gpu(false);
printf("Clip will use CPU for this model!\n");
}
#endif
clp_ctx = clip_model_load(mmproj_filename.c_str(), /*verbosity=*/ 1);
if(clp_ctx == nullptr) {
fprintf(stderr, "%s: error: failed to load mmproj model!\n", __func__);