Expose DecodeSpecialTokens through IInferenceParams for StatelessExecutor

LLama.Web/Common/InferenceOptions.cs

@@ -21,5 +21,8 @@ public class InferenceOptions
 
         /// <inheritdoc />
         public ISamplingPipeline SamplingPipeline { get; set; } = new DefaultSamplingPipeline();
+
+        /// <inheritdoc />
+        public bool DecodeSpecialTokens { get; set; }
     }
 }

LLama/Abstractions/IInferenceParams.cs

@@ -28,5 +28,13 @@ public interface IInferenceParams
 		/// Set a custom sampling pipeline to use.
 		/// </summary>
 		ISamplingPipeline SamplingPipeline { get; set; }
+
+		/// <summary>
+		/// If true, special characters will be converted to text. If false they will be invisible.
+		/// </summary>
+		/// <remark>
+		/// Controls the behavior of decoders like <see cref="StreamingTokenDecoder" />
+		/// </remark>
+		public bool DecodeSpecialTokens { get; set; }
 	}
 }

LLama/Common/InferenceParams.cs

@@ -30,6 +30,9 @@ public record InferenceParams
 
         /// <inheritdoc />
         public ISamplingPipeline SamplingPipeline { get; set; } = new DefaultSamplingPipeline();
+
+        /// <inheritdoc />
+        public bool DecodeSpecialTokens { get; set; }
     }
 
     /// <summary>

LLama/LLamaStatelessExecutor.cs

@@ -88,7 +88,10 @@ public async IAsyncEnumerable<string> InferAsync(string prompt, IInferenceParams
                 throw new ArgumentOutOfRangeException(nameof(inferenceParams), $"TokensKeep ({inferenceParams.TokensKeep}) cannot be larger than ContextSize ({Context.ContextSize})");
 
             // Create decoders for the token stream
-            var decoder = new StreamingTokenDecoder(Context);
+            var decoder = new StreamingTokenDecoder(Context)
+            {
+                DecodeSpecialTokens = inferenceParams.DecodeSpecialTokens,
+            };
             var antiprocessor = new AntipromptProcessor(inferenceParams.AntiPrompts);
 
             if (ApplyTemplate)