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/embedding/embedding.cpp
View file

@ -88,11 +88,11 @@ int main(int argc, char ** argv) {
}
const int n_ctx_train = llama_n_ctx_train(model);
const int n_ctx = llama_n_ctx(ctx);
const int kv_size = llama_kv_size(ctx);
if (n_ctx > n_ctx_train) {
if (kv_size > n_ctx_train) {
fprintf(stderr, "%s: warning: model was trained on only %d context tokens (%d specified)\n",
__func__, n_ctx_train, n_ctx);
__func__, n_ctx_train, kv_size);
}
// print system information
@ -106,7 +106,7 @@ int main(int argc, char ** argv) {
// max batch size
const uint64_t n_batch = params.n_batch;
GGML_ASSERT(params.n_batch == params.n_ctx);
GGML_ASSERT(params.n_batch == params.kv_size);
// tokenize the prompts and trim
std::vector<std::vector<int32_t>> inputs;