From 3300247e97fe5726821ba47299e289bcb3351b23 Mon Sep 17 00:00:00 2001
From: Fabio Rossini Sluzala <fabio3rs@gmail.com>
Date: Sat, 1 Apr 2023 17:01:12 -0300
Subject: [PATCH] Fix for temp == 0

---
 llama.cpp | 7 +++++++
 1 file changed, 7 insertions(+)

diff --git a/llama.cpp b/llama.cpp
index bed24207d..853427a96 100644
--- a/llama.cpp
+++ b/llama.cpp
@@ -3,6 +3,7 @@
 #include "ggml.h"
 
 #include <cinttypes>
+#include <float.h>
 #include <fstream>
 #include <random>
 #include <map>
@@ -1757,6 +1758,12 @@ llama_token llama_sample_top_p_top_k(
     // TODO: avoid this ...
     const auto last_n_tokens = std::vector<llama_token>(last_n_tokens_data, last_n_tokens_data + last_n_tokens_size);
 
+    if (std::abs(temp) < FLT_EPSILON) {
+        temp = 0.8f;
+        top_k = 1.0f;
+        top_p = 0.0f;
+    }
+
     result = llama_sample_top_p_top_k(
             *ctx,
             last_n_tokens,