From c29a4b53de20d24211363c81b743350ccfbe0659 Mon Sep 17 00:00:00 2001
From: Plamen Minev <pacominev@gmail.com>
Date: Wed, 18 Dec 2024 12:40:53 +0200
Subject: [PATCH] refactor: update llama, ref #26

-- Penalty sampler API initalization was changed
---
 code/ac/llama/Sampler.cpp | 7 +------
 code/ac/llama/Sampler.hpp | 3 ---
 llama.cpp                 | 2 +-
 3 files changed, 2 insertions(+), 10 deletions(-)

diff --git a/code/ac/llama/Sampler.cpp b/code/ac/llama/Sampler.cpp
index 146acc6..77a4094 100644
--- a/code/ac/llama/Sampler.cpp
+++ b/code/ac/llama/Sampler.cpp
@@ -37,15 +37,10 @@ Sampler::Sampler(Model& model, const Params& params)
 
     llama_sampler_chain_add(chain,
         llama_sampler_init_penalties(
-            llama_n_vocab(lmodel),
-            llama_token_eos(lmodel),
-            llama_token_nl(lmodel),
             params.repetitionPenalty.numTokens,
             params.repetitionPenalty.repeat,
             params.repetitionPenalty.freq,
-            params.repetitionPenalty.present,
-            params.penalizeNewline,
-            params.ignoreEos
+            params.repetitionPenalty.present
         )
     );
 
diff --git a/code/ac/llama/Sampler.hpp b/code/ac/llama/Sampler.hpp
index 2b34b83..b869239 100644
--- a/code/ac/llama/Sampler.hpp
+++ b/code/ac/llama/Sampler.hpp
@@ -62,9 +62,6 @@ class AC_LLAMA_EXPORT Sampler {
             float threshold = 0.10f;   // > 0.5 disables XTC
         } xtc;
 
-        bool penalizeNewline = false; // consider newlines as a repeatable token
-        bool ignoreEos = false;
-
         std::vector<SamplingType> samplerSequence = {
             SamplingType::Top_K,
             SamplingType::Typical_P,
diff --git a/llama.cpp b/llama.cpp
index fbc979b..0e70ba6 160000
--- a/llama.cpp
+++ b/llama.cpp
@@ -1 +1 @@
-Subproject commit fbc979bf6e986a7b07f61b1a4bf8d0ac70e1bbdd
+Subproject commit 0e70ba686e6c717a0aa41d88284e2a392c2bd0cd