From 43c55bb7e2fc6933c9ffbd8566413a31f86225fc Mon Sep 17 00:00:00 2001 From: Concedo <39025047+LostRuins@users.noreply.github.com> Date: Tue, 30 Jul 2024 22:18:22 +0800 Subject: [PATCH] hack to fix bad unicode fragments corrupting streamed output --- koboldcpp.py | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/koboldcpp.py b/koboldcpp.py index befdad114..9af8c9ddf 100644 --- a/koboldcpp.py +++ b/koboldcpp.py @@ -1435,7 +1435,8 @@ class ServerRequestHandler(http.server.SimpleHTTPRequestHandler): newbyte = ctypes.string_at(token) incomplete_token_buffer += bytearray(newbyte) tokenSeg = incomplete_token_buffer.decode("UTF-8","ignore") - if tokenSeg!="": + badFragment = (tokenSeg==" " and len(incomplete_token_buffer)>1) #partial incomplete unicode + if tokenSeg!="" and not badFragment: incomplete_token_buffer.clear() tokenStr += tokenSeg