CS348Project/llama.cpp
5
0
Fork 0
mirror of https://github.com/ggml-org/llama.cpp.git synced 2025-11-01 09:01:57 +00:00
Code Issues Packages Projects Releases Wiki Activity

parallel : process system prompt once + configurable paramters + llama API

Browse Source
This commit is contained in:
Georgi Gerganov
2023-09-19 17:00:42 +03:00
parent 82e20e9ba0
commit 4b5f3cd6bf
9 changed files with 187 additions and 93 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
examples/perplexity/perplexity.cpp
View File

@@ -207,7 +207,7 @@ 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_keep_seq(ctx, -1);
llama_kv_cache_tokens_rm(ctx, -1, -1);
for (int j = 0; j < num_batches; ++j) {
const int batch_start = start + j * n_batch;
@@ -335,7 +335,7 @@ 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_keep_seq(ctx, -1);
llama_kv_cache_tokens_rm(ctx, -1, -1);
for (int j = 0; j < num_batches; ++j) {
const int batch_start = start + j * n_batch;
@@ -568,7 +568,7 @@ static void hellaswag_score(llama_context * ctx, const gpt_params & params) {
}
// clear the KV cache
llama_kv_cache_keep_seq(ctx, -1);
llama_kv_cache_tokens_rm(ctx, -1, -1);
auto logits = hellaswag_evaluate_tokens(ctx, query_embd, 0, params.n_batch, n_vocab, params.n_threads);
if (logits.empty()) {