From 1c4a244291e0e6f141a7041d20846da56025223d Mon Sep 17 00:00:00 2001 From: Aaron Miller Date: Thu, 13 Jul 2023 18:24:01 -0400 Subject: [PATCH] bump mem allocation a bit --- gpt4all-backend/bert.cpp | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/gpt4all-backend/bert.cpp b/gpt4all-backend/bert.cpp index 7a83289b..29532c48 100644 --- a/gpt4all-backend/bert.cpp +++ b/gpt4all-backend/bert.cpp @@ -871,7 +871,7 @@ struct bert_ctx * bert_load_from_file(const char *fname) // TODO: Max tokens should be a param? int32_t N = new_bert->model.hparams.n_max_tokens; - new_bert->mem_per_input = 1.9 * (new_bert->mem_per_token * N); // add 10% to account for ggml object overhead + new_bert->mem_per_input = 2.2 * (new_bert->mem_per_token * N); // add 10% to account for ggml object overhead } #if defined(DEBUG_BERT)