tqcq/llama.cpp
0
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-06-27 12:05:03 +00:00
Code Issues Packages Projects Releases Wiki Activity

Extend llama_kv_cache_seq_rm to allow matching any sequence (#3843)

Browse Source 
* Extend llama_kv_cache_seq_rm to allow matichng any sequence

* Replace llama_kv_cache_tokens_rm with llama_kv_cache_clear

Use llama_kv_cache_clear for cache clearing

Change calls to llama_kv_cache_tokens_rm that want to delete by position to use llama_kv_cache_seq_rm functionality
This commit is contained in:
Kerfuffle
2023-10-29 11:31:40 -06:00
committed by GitHub
parent 2046eb4345
commit 6e08281e58
8 changed files with 30 additions and 32 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
examples/perplexity/perplexity.cpp
View File

@ -210,7 +210,7 @@ static results_perplexity perplexity_v2(llama_context * ctx, const gpt_params &
const auto t_start = std::chrono::high_resolution_clock::now();
// clear the KV cache
llama_kv_cache_tokens_rm(ctx, -1, -1);
llama_kv_cache_clear(ctx);
for (int j = 0; j < num_batches; ++j) {
const int batch_start = start + j * n_batch;
@ -339,7 +339,7 @@ static results_perplexity perplexity(llama_context * ctx, const gpt_params & par
const auto t_start = std::chrono::high_resolution_clock::now();
// clear the KV cache
llama_kv_cache_tokens_rm(ctx, -1, -1);
llama_kv_cache_clear(ctx);
for (int j = 0; j < num_batches; ++j) {
const int batch_start = start + j * n_batch;
@ -573,7 +573,7 @@ static void hellaswag_score(llama_context * ctx, const gpt_params & params) {
}
// clear the KV cache
llama_kv_cache_tokens_rm(ctx, -1, -1);
llama_kv_cache_clear(ctx);
auto logits = hellaswag_evaluate_tokens(ctx, query_embd, 0, params.n_batch, n_vocab);
if (logits.empty()) {