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

rename n_ctx to kv_size

Browse source
This commit is contained in:
Pierrick HYMBERT 2024-02-18 20:59:26 +01:00 • committed by Georgi Gerganov
parent ef96e8b1f7
commit 606873401c
No known key found for this signature in database
GPG key ID: 449E073F9DC10735
48 changed files with 403 additions and 393 deletions
Download patch file Download diff file Expand all files Collapse all files

8
examples/llava/llava.cpp
View file

@ -103,15 +103,15 @@ static bool clip_llava_handle_patches(clip_ctx * ctx_clip, std::vector<float *>
const size_t num_images = num_patches_width * num_patches_height + 1;
// TODO: size calculation is not calculated - it's only tens of MB
size_t ctx_size = 0;
size_t kv_size = 0;
{
ctx_size += clip_embd_nbytes(ctx_clip) * num_images * 8; // image_features
ctx_size += 1024*1024 * ggml_type_size(GGML_TYPE_F32);
kv_size += clip_embd_nbytes(ctx_clip) * num_images * 8; // image_features
kv_size += 1024*1024 * ggml_type_size(GGML_TYPE_F32);
}
struct ggml_init_params params {
/*.mem_size =*/ ctx_size,
/*.mem_size =*/ kv_size,
/*.mem_buffer =*/ NULL,
/*.no_alloc =*/ false, // NOTE: this should be false when using the legacy API
};