From d01bccde9f759b24449fdaa16306b406a50eb367 Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Tue, 18 Jul 2023 14:24:43 +0300 Subject: ci : integrate with ggml-org/ci (#2250) * ci : run ctest ggml-ci * ci : add open llama 3B-v2 tests ggml-ci * ci : disable wget progress output ggml-ci * ci : add open llama 3B-v2 tg tests for q4 and q5 quantizations ggml-ci * tests : try to fix tail free sampling test ggml-ci * ci : add K-quants ggml-ci * ci : add short perplexity tests ggml-ci * ci : add README.md * ppl : add --chunks argument to limit max number of chunks ggml-ci * ci : update README --- examples/perplexity/perplexity.cpp | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) (limited to 'examples/perplexity/perplexity.cpp') diff --git a/examples/perplexity/perplexity.cpp b/examples/perplexity/perplexity.cpp index 7e120ff1..bfad9993 100644 --- a/examples/perplexity/perplexity.cpp +++ b/examples/perplexity/perplexity.cpp @@ -32,13 +32,15 @@ void perplexity(llama_context * ctx, const gpt_params & params) { // BOS tokens will be added for each chunk before eval auto tokens = ::llama_tokenize(ctx, params.prompt, true); - int count = 0; + const int n_chunk_max = tokens.size() / params.n_ctx; - const int n_chunk = tokens.size() / params.n_ctx; + const int n_chunk = params.n_chunks < 0 ? n_chunk_max : std::min(params.n_chunks, n_chunk_max); const int n_vocab = llama_n_vocab(ctx); const int n_batch = params.n_batch; + int count = 0; double nll = 0.0; + fprintf(stderr, "%s: calculating perplexity over %d chunks, batch_size=%d\n", __func__, n_chunk, n_batch); for (int i = 0; i < n_chunk; ++i) { -- cgit v1.2.3