refactor: update llama, ref #26

-- Penalty sampler API initalization was changed

Author: pminev

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
         )
     );
 

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,