highperfocused/ollama
1
0
Fork 0
mirror of https://github.com/ollama/ollama.git synced 2025-08-03 05:13:09 +02:00
Code Issues Packages Projects Releases Wiki Activity

llama: use dynamic backend loading for mllama and clip (#8835)

Browse Source
This commit is contained in:
Jeffrey Morgan
2025-02-05 09:46:56 -08:00
committed by GitHub
parent c852b8e021
commit cd3fbf1c49
3 changed files with 36 additions and 82 deletions
Download Patch File Download Diff File Expand all files Collapse all files

31
llama/mllama.cpp vendored
View File

@@ -558,30 +558,15 @@ struct mllama_ctx *mllama_model_load(const char *fname, const int verbosity = 1)
mllama_ctx *new_mllama = new mllama_ctx{};
#ifdef GGML_USE_CUDA
new_mllama->backend = ggml_backend_cuda_init(0);
LOG("vision using CUDA backend");
#endif
#ifdef GGML_USE_METAL
new_mllama->backend = ggml_backend_metal_init();
LOG("vision using Metal backend");
#endif
#ifdef GGML_USE_CANN
new_mllama->backend = ggml_backend_cann_init(0);
LOG("vision using CANN backend");
#endif
#ifdef GGML_USE_VULKAN
new_mllama->backend = ggml_backend_vk_init(0);
LOG("vision using Vulkan backend");
#endif
if (!new_mllama->backend) {
new_mllama->backend = ggml_backend_cpu_init();
LOG("vision using CPU backend");
ggml_backend_t backend = ggml_backend_init_best();
if (backend == nullptr) {
LOG("%s: failed to initialize backend\n", __func__);
mllama_free(new_mllama);
gguf_free(ctx);
return nullptr;
}
LOG("%s: using %s backend\n", __func__, ggml_backend_name(backend));
new_mllama->backend = backend;
// load tensors
{