Updated binaries.

- llama.cpp: b8c1476e44
 - Build action: https://github.com/SciSharp/LLamaSharp/actions/runs/8886754252/job/24400788991

LLama.Examples/Examples/BatchedExecutorGuidance.cs

@@ -79,7 +79,7 @@ public class BatchedExecutorGuidance
                     guidance.Prompt(g);
 
                     // Early exit if we reach the natural end of the guided sentence
-                    if (g == model.Tokens.EOS)
+                    if (model.Tokens.IsEndOfGeneration(g))
                         break;
 
                     // Update progress bar

LLama/LLamaStatelessExecutor.cs

@@ -123,8 +123,8 @@ namespace LLama
                     );
                 }
 
-                // Check if this is the EOS token
-                if (id == _weights.Tokens.EOS)
+                // Check if this token should end generation
+                if (_weights.Tokens.IsEndOfGeneration(id))
                     break;
 
                 // Decode this token into text

LLama/Native/LLamaModelMetadataOverride.cs

@@ -43,6 +43,8 @@ public unsafe struct LLamaModelMetadataOverride
     /// </summary>
     [FieldOffset(136)]
     public long BoolValue;
+
+    //todo: char val_str[128];
 }
 
 /// <summary>
@@ -65,4 +67,9 @@ public enum LLamaModelKvOverrideType
     /// Overriding a bool value
     /// </summary>
     Bool = 2,
+
+    /// <summary>
+    /// Overriding a string value
+    /// </summary>
+    String = 3,
 }

LLama/Native/LLamaModelParams.cs

@@ -81,6 +81,16 @@ namespace LLama.Native
         }
         private sbyte _use_mlock;
 
+        /// <summary>
+        /// validate model tensor data
+        /// </summary>
+        public bool check_tensors
+        {
+            readonly get => Convert.ToBoolean(_check_tensors);
+            set => _check_tensors = Convert.ToSByte(value);
+        }
+        private sbyte _check_tensors;
+
         /// <summary>
         /// Create a LLamaModelParams with default values
         /// </summary>

LLama/Native/LLamaModelQuantizeParams.cs

@@ -70,6 +70,16 @@ namespace LLama.Native
         }
         private sbyte _pure;
 
+        /// <summary>
+        /// quantize to the same number of shards
+        /// </summary>
+        public bool keep_split
+        {
+            get => Convert.ToBoolean(_keep_split);
+            set => _keep_split = Convert.ToSByte(value);
+        }
+        private sbyte _keep_split;
+
         /// <summary>
         /// pointer to importance matrix data
         /// </summary>

LLama/Native/LLamaVocabPreType.cs

@@ -0,0 +1,17 @@
+namespace LLama.Native;
+
+/// <summary>
+/// 
+/// </summary>
+/// <remarks>llama_vocab_pre_type</remarks>
+internal enum LLamaVocabPreType
+{
+    LLAMA_VOCAB_PRE_TYPE_DEFAULT = 0,
+    LLAMA_VOCAB_PRE_TYPE_LLAMA3 = 1,
+    LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_LLM = 2,
+    LLAMA_VOCAB_PRE_TYPE_DEEPSEEK_CODER = 3,
+    LLAMA_VOCAB_PRE_TYPE_FALCON = 4,
+    LLAMA_VOCAB_PRE_TYPE_MPT = 5,
+    LLAMA_VOCAB_PRE_TYPE_STARCODER = 6,
+    LLAMA_VOCAB_PRE_TYPE_GPT2 = 7,
+}

LLama/Native/NativeApi.Sampling.cs

@@ -176,7 +176,7 @@ namespace LLama.Native
         public static extern LLamaToken llama_sample_token_greedy(SafeLLamaContextHandle ctx, ref LLamaTokenDataArrayNative candidates);
 
         /// <summary>
-        /// Randomly selects a token from the candidates based on their probabilities.
+        /// Randomly selects a token from the candidates based on their probabilities using the RNG of ctx.
         /// </summary>
         /// <param name="ctx"></param>
         /// <param name="candidates">Pointer to LLamaTokenDataArray</param>

LLama/Native/NativeApi.cs

@@ -133,6 +133,14 @@ namespace LLama.Native
         [DllImport(libraryName, CallingConvention = CallingConvention.Cdecl)]
         public static extern uint llama_n_seq_max(SafeLLamaContextHandle ctx);
 
+        /// <summary>
+        /// Get the pooling type for this context
+        /// </summary>
+        /// <param name="ctx"></param>
+        /// <returns></returns>
+        [DllImport(libraryName, CallingConvention = CallingConvention.Cdecl)]
+        public static extern LLamaPoolingType llama_pooling_type(SafeLLamaContextHandle ctx);
+
         /// <summary>
         /// Get the embeddings for the a specific sequence.
         /// Equivalent to: llama_get_embeddings(ctx) + ctx->output_ids[i]*n_embd
@@ -218,19 +226,20 @@ namespace LLama.Native
         /// <param name="model"></param>
         /// <param name="llamaToken"></param>
         /// <param name="buffer">buffer to write string into</param>
+        /// <param name="special">If true, special tokens are rendered in the output</param>
         /// <returns>The length written, or if the buffer is too small a negative that indicates the length required</returns>
-        public static int llama_token_to_piece(SafeLlamaModelHandle model, LLamaToken llamaToken, Span<byte> buffer)
+        public static int llama_token_to_piece(SafeLlamaModelHandle model, LLamaToken llamaToken, Span<byte> buffer, bool special)
         {
             unsafe
             {
                 fixed (byte* bufferPtr = buffer)
                 {
-                    return llama_token_to_piece_native(model, llamaToken, bufferPtr, buffer.Length);
+                    return llama_token_to_piece_native(model, llamaToken, bufferPtr, buffer.Length, special);
                 }
             }
 
             [DllImport(libraryName, CallingConvention = CallingConvention.Cdecl, EntryPoint = "llama_token_to_piece")]
-            static extern unsafe int llama_token_to_piece_native(SafeLlamaModelHandle model, LLamaToken llamaToken, byte* buffer, int length);
+            static extern unsafe int llama_token_to_piece_native(SafeLlamaModelHandle model, LLamaToken llamaToken, byte* buffer, int length, bool special);
         }
 
         /// <summary>

LLama/Native/SafeLlamaModelHandle.cs

@@ -361,6 +361,15 @@ namespace LLama.Native
         /// <returns></returns>
         [DllImport(NativeApi.libraryName, CallingConvention = CallingConvention.Cdecl)]
         private static extern int llama_token_eot(SafeLlamaModelHandle model);
+
+        /// <summary>
+        /// Check if the token is supposed to end generation (end-of-generation, eg. EOS, EOT, etc.)
+        /// </summary>
+        /// <param name="model"></param>
+        /// <param name="token"></param>
+        /// <returns></returns>
+        [DllImport(NativeApi.libraryName, CallingConvention = CallingConvention.Cdecl)]
+        private static extern bool llama_token_is_eog(SafeLlamaModelHandle model, LLamaToken token);
         #endregion
 
         #region LoRA
@@ -402,10 +411,11 @@ namespace LLama.Native
         /// </summary>
         /// <param name="token">Token to decode</param>
         /// <param name="dest">A span to attempt to write into. If this is too small nothing will be written</param>
+        /// <param name="special">If true, special charcters will be converted to text. If false they will be invisible.</param>
         /// <returns>The size of this token. **nothing will be written** if this is larger than `dest`</returns>
-        public uint TokenToSpan(LLamaToken token, Span<byte> dest)
+        public uint TokenToSpan(LLamaToken token, Span<byte> dest, bool special = false)
         {
-            var length = NativeApi.llama_token_to_piece(this, token, dest);
+            var length = NativeApi.llama_token_to_piece(this, token, dest, special);
             return (uint)Math.Abs(length);
         }
 
@@ -623,6 +633,16 @@ namespace LLama.Native
             /// Codellama end of infill middle
             /// </summary>
             public LLamaToken? EOT => Normalize(llama_token_eot(_model));
+
+            /// <summary>
+            /// Check if the given token should end generation
+            /// </summary>
+            /// <param name="token"></param>
+            /// <returns></returns>
+            public bool IsEndOfGeneration(LLamaToken token)
+            {
+                return llama_token_is_eog(_model, token);
+            }
         }
     }
 }