llama : speedup tokenization (#2831)

* Speedup tokenization

On current master it takes ~3.2 seconds to tokenize
Wikitext. With this change it becomes ~525 ms.

* Fixit: it was missing the piece after the last found occurence

---------

Co-authored-by: Iwan Kawrakow <iwan.kawrakow@gmail.com>

1 files changed, 4 insertions, 0 deletions

diff --git a/examples/perplexity/perplexity.cpp b/examples/perplexity/perplexity.cpp
index b596d062..ebafa0c2 100644
--- a/examples/perplexity/perplexity.cpp
+++ b/examples/perplexity/perplexity.cpp
@@ -190,10 +190,14 @@ void perplexity(llama_context * ctx, const gpt_params & params) {
     const bool is_spm = llama_vocab_type(ctx) == LLAMA_VOCAB_TYPE_SPM;
     const bool add_bos = is_spm;
 
+    auto tim1 = std::chrono::high_resolution_clock::now();
     fprintf(stderr, "%s: tokenizing the input ..\n", __func__);
 
     auto tokens = ::llama_tokenize(ctx, params.prompt, add_bos);
 
+    auto tim2 = std::chrono::high_resolution_clock::now();
+    fprintf(stderr, "%s: tokenization took %g ms\n",__func__,1e-3*std::chrono::duration_cast<std::chrono::microseconds>(tim2-tim1).count());
+
     const int n_chunk_max = tokens.size() / params.n_ctx;
 
     const int n_chunk = params.n_chunks < 0 ? n_chunk_max : std::min(params.n_chunks, n_chunk_max);