From 44f906e8537fcec965e312d621c80556d6aa9bec Mon Sep 17 00:00:00 2001 From: Georgi Gerganov Date: Tue, 6 Jun 2023 20:16:57 +0300 Subject: metal : add f16 support --- llama.cpp | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) (limited to 'llama.cpp') diff --git a/llama.cpp b/llama.cpp index 70341d04..73f68600 100644 --- a/llama.cpp +++ b/llama.cpp @@ -961,7 +961,6 @@ static void llama_model_load_internal( model.hparams = ml->file_loaders.at(0)->hparams; llama_file_version file_version = ml->file_loaders.at(0)->file_version; auto & hparams = model.hparams; - uint32_t n_ff = ((2*(4*hparams.n_embd)/3 + hparams.n_mult - 1)/hparams.n_mult)*hparams.n_mult; { switch (hparams.n_layer) { @@ -975,6 +974,8 @@ static void llama_model_load_internal( hparams.n_ctx = n_ctx; } + const uint32_t n_ff = ((2*(4*hparams.n_embd)/3 + hparams.n_mult - 1)/hparams.n_mult)*hparams.n_mult; + { fprintf(stderr, "%s: format = %s\n", __func__, llama_file_version_name(file_version)); fprintf(stderr, "%s: n_vocab = %u\n", __func__, hparams.n_vocab); -- cgit v1.2.3