vrr/prima.cpp
2
0
Fork 0
mirror of https://github.com/Lizonghang/prima.cpp.git synced 2025-09-11 06:54:50 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions

Merge branch 'tao' into dev

Browse source
This commit is contained in:
leeetao  2025-02-24 16:48:43 +00:00
commit 224d14eb4c
4 changed files with 198 additions and 74 deletions
Download patch file Download diff file Expand all files Collapse all files

6
common/common.cpp
View file

@ -873,13 +873,17 @@ static bool assign_layers_to_device(
// model-specific constants
const int n_embd_k_gqa = llama_model_n_embd_k_gqa(model);
const int n_embd_v_gqa = llama_model_n_embd_v_gqa(model);
if (n_embd_k_gqa <= 0 || n_embd_v_gqa <= 0) {
LOG_ERR("Invalid model parameters,n_embd_k_gqa and n_embd_v_gqa cannot be less than 0");
return false;
}
const int n_kv = cparams.n_ctx;
const int64_t b = dev_info_set[0].model_bytes.nb_layer;
const int64_t bo = dev_info_set[0].model_bytes.nb_output;
const int64_t b_prime = b + 2 * (n_embd_k_gqa + n_embd_v_gqa) * n_kv;
#if defined(USE_HIGHS)
#if defined(USE_HIGHS)
const device_info &master = dev_info_set[0];
const int n_vocab = llama_n_vocab(model);
const int64_t bi = dev_info_set[0].model_bytes.nb_input;