From a03fe003de537bf4044ec6ca658bb33cb768c9f4 Mon Sep 17 00:00:00 2001
From: Martin Evans <martindevans@gmail.com>
Date: Mon, 23 Oct 2023 16:42:38 +0100
Subject: [PATCH] Fixed decoding of text "accumulating" over time (never
 properly clearing buffer)

---
 LLama/LLamaStatelessExecutor.cs      | 2 +-
 LLama/Native/SafeLlamaModelHandle.cs | 4 ----
 LLama/StreamingTokenDecoder.cs       | 7 ++++++-
 3 files changed, 7 insertions(+), 6 deletions(-)

diff --git a/LLama/LLamaStatelessExecutor.cs b/LLama/LLamaStatelessExecutor.cs
index f5cdfb2e..ab1e9bbc 100644
--- a/LLama/LLamaStatelessExecutor.cs
+++ b/LLama/LLamaStatelessExecutor.cs
@@ -100,7 +100,7 @@ namespace LLama
 
                 decoder.Add(id);
                 var decoded = decoder.Read();
-                yield return decoder.Read();
+                yield return decoded;
 
                 tokens.Clear();
                 tokens.Add(id);
diff --git a/LLama/Native/SafeLlamaModelHandle.cs b/LLama/Native/SafeLlamaModelHandle.cs
index 94ddca2d..b93c2b89 100644
--- a/LLama/Native/SafeLlamaModelHandle.cs
+++ b/LLama/Native/SafeLlamaModelHandle.cs
@@ -1,11 +1,7 @@
 using System;
-using System.Buffers;
 using System.Collections.Generic;
-using System.Diagnostics;
-using System.Runtime.InteropServices;
 using System.Text;
 using LLama.Exceptions;
-using LLama.Extensions;
 
 namespace LLama.Native
 {
diff --git a/LLama/StreamingTokenDecoder.cs b/LLama/StreamingTokenDecoder.cs
index fc459199..c5d9683e 100644
--- a/LLama/StreamingTokenDecoder.cs
+++ b/LLama/StreamingTokenDecoder.cs
@@ -155,7 +155,12 @@ public sealed class StreamingTokenDecoder
     /// <returns></returns>
     public string Read()
     {
-        return string.Join("", _characters);
+        if (_characters.Count == 0)
+            return "";
+
+        var str = string.Join("", _characters);
+        _characters.Clear();
+        return str;
     }
 
     /// <summary>