vbatts/llama.cpp
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

llama_control_vector_load: let gguf_init_from_file allocate the ggml_context

Browse source
This commit is contained in:
Anon 2024-03-26 01:28:18 +00:00 • committed by trollkotze
parent 6e1fbf87b0
commit 181879f942
1 changed files with 2 additions and 13 deletions
Download patch file Download diff file Expand all files Collapse all files

15
common/common.cpp
View file

@ -2650,12 +2650,7 @@ static llama_control_vector_data llama_control_vector_load_one(const llama_contr
// calculate size of ctx needed for tensors, ensure tensors are f32, and find max layer // calculate size of ctx needed for tensors, ensure tensors are f32, and find max layer
{ {
struct ggml_init_params meta_params = { ggml_context * meta_ctx = nullptr;
/* .mem_size = */ ggml_tensor_overhead() * 128 + ggml_graph_overhead(),
/* .mem_buffer = */ nullptr,
/* .no_alloc = */ true,
};
ggml_context * meta_ctx = ggml_init(meta_params);
struct gguf_init_params meta_gguf_params = { struct gguf_init_params meta_gguf_params = {
/* .no_alloc = */ true, /* .no_alloc = */ true,
/* .ctx = */ &meta_ctx, /* .ctx = */ &meta_ctx,
@ -2720,13 +2715,7 @@ static llama_control_vector_data llama_control_vector_load_one(const llama_contr
} }
// load and scale tensors into final control vector context // load and scale tensors into final control vector context
struct ggml_init_params ggml_params = { struct ggml_context * ctx = nullptr;
/* .mem_size = */ ggml_tensor_overhead() * n_tensors + n_bytes,
/* .mem_buffer = */ nullptr,
/* .no_alloc = */ false,
};
struct ggml_context * ctx = ggml_init(ggml_params);
struct gguf_init_params params = { struct gguf_init_params params = {
/*.no_alloc = */ false, /*.no_alloc = */ false,
/*.ctx = */ &ctx, /*.ctx = */ &ctx,