From 2df380170602d49040a99209a07bbb0bcebaa4f3 Mon Sep 17 00:00:00 2001 From: Marcus Dunn Date: Mon, 23 Oct 2023 09:29:11 -0700 Subject: [PATCH] changed token functions to use new model variants --- examples/perplexity/perplexity.cpp | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/examples/perplexity/perplexity.cpp b/examples/perplexity/perplexity.cpp index 7d0038bd4..3c2542e8c 100644 --- a/examples/perplexity/perplexity.cpp +++ b/examples/perplexity/perplexity.cpp @@ -227,7 +227,7 @@ static results_perplexity perplexity_v2(llama_context * ctx, const gpt_params & // add BOS token for the first batch of each chunk if (add_bos && j == 0) { - tokens[batch_start] = llama_token_bos(ctx); + tokens[batch_start] = llama_token_bos(llama_get_model(ctx)); } const auto batch_logits = llama_get_logits(ctx); @@ -350,7 +350,7 @@ static results_perplexity perplexity(llama_context * ctx, const gpt_params & par // add BOS token for the first batch of each chunk if (add_bos && j == 0) { - tokens[batch_start] = llama_token_bos(ctx); + tokens[batch_start] = llama_token_bos(llama_get_model(ctx)); } if (llama_decode(ctx, llama_batch_get_one(tokens.data() + batch_start, batch_size, j * n_batch, 0))) {