From 53b8eaa316e80aafe09388e76360c0064146af16 Mon Sep 17 00:00:00 2001 From: Nexesenex <124105151+Nexesenex@users.noreply.github.com> Date: Sat, 24 Aug 2024 21:57:07 +0200 Subject: [PATCH] Remove deprecated rules for token embeddings --- src/llama.cpp | 3 --- 1 file changed, 3 deletions(-) diff --git a/src/llama.cpp b/src/llama.cpp index fbae392f2..e2ea5e071 100644 --- a/src/llama.cpp +++ b/src/llama.cpp @@ -16514,9 +16514,6 @@ static ggml_type llama_tensor_get_type(quantize_state_internal & qs, ggml_type n if (qs.model.hparams.n_vocab >= 127999 || qs.model.hparams.n_head() <= 20) new_type = GGML_TYPE_IQ3_S; new_type = GGML_TYPE_IQ4_XS; } - else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_S || ftype == LLAMA_FTYPE_MOSTLY_IQ2_M) new_type = GGML_TYPE_IQ3_XXS; - else if (ftype == LLAMA_FTYPE_MOSTLY_IQ2_XL) new_type = GGML_TYPE_IQ3_XXS; - else if (ftype == LLAMA_FTYPE_MOSTLY_IQ3_XXS) new_type = GGML_TYPE_IQ3_S; else if (new_type == GGML_TYPE_Q4_0_4_4 || new_type == GGML_TYPE_Q4_0_4_8 || new_type == GGML_TYPE_Q4_0_8_8) { new_type = GGML_TYPE_Q4_0; }