llm: add back check for empty token cache

fcf4d60e · jmorganca · e33d5c2d · fcf4d60e
Commit fcf4d60e authored Apr 30, 2024 by jmorganca
Hide whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

llm/ext_server/server.cpp llm/ext_server/server.cpp +1 -1

No files found.
--- a/llm/ext_server/server.cpp
+++ b/llm/ext_server/server.cpp
@@ -1032,7 +1032,7 @@ struct llama_server_context
            slot.has_next_token = false;
        }

-        if (llama_token_is_eog(model, result.tok))
+        if (!slot.cache_tokens.empty() && llama_token_is_eog(model, result.tok))
        {
            slot.stopped_eos = true;
            slot.has_next_token = false;