From 4edde21b3d68e853ac296aafa08cd90810657ba0 Mon Sep 17 00:00:00 2001 From: Andrei Betlen Date: Wed, 21 Feb 2024 11:05:58 -0500 Subject: [PATCH] feat: Update llama.cpp --- llama_cpp/llama_cpp.py | 2 ++ vendor/llama.cpp | 2 +- 2 files changed, 3 insertions(+), 1 deletion(-) diff --git a/llama_cpp/llama_cpp.py b/llama_cpp/llama_cpp.py index e946adb..3ebe82b 100644 --- a/llama_cpp/llama_cpp.py +++ b/llama_cpp/llama_cpp.py @@ -191,6 +191,7 @@ LLAMA_TOKEN_TYPE_BYTE = 6 # LLAMA_FTYPE_MOSTLY_Q3_K_XS = 22, // except 1d tensors # LLAMA_FTYPE_MOSTLY_IQ3_XXS = 23, // except 1d tensors # LLAMA_FTYPE_MOSTLY_IQ1_S = 24, // except 1d tensors +# LLAMA_FTYPE_MOSTLY_IQ4_NL = 25, // except 1d tensors # LLAMA_FTYPE_GUESSED = 1024, // not specified in the model file # }; @@ -217,6 +218,7 @@ LLAMA_FTYPE_MOSTLY_Q2_K_S = 21 LLAMA_FTYPE_MOSTLY_Q3_K_XS = 22 LLAMA_FTYPE_MOSTLY_IQ3_XXS = 23 LLAMA_FTYPE_MOSTLY_IQ1_S = 24 +LLAMA_FTYPE_MOSTLY_IQ4_NL = 25 LLAMA_FTYPE_GUESSED = 1024 # enum llama_rope_scaling_type { diff --git a/vendor/llama.cpp b/vendor/llama.cpp index c0a8c6d..89febfe 160000 --- a/vendor/llama.cpp +++ b/vendor/llama.cpp @@ -1 +1 @@ -Subproject commit c0a8c6db371cb3e4379900867b948879f5842201 +Subproject commit 89febfed9322c8849520dc63c93ee4f5fd72556e