Remove workaround

llama_cpp/llama.py

@@ -176,11 +176,6 @@ class Llama:
             The sampled token.
         """
         assert self.ctx is not None
-        # Temporary workaround for https://github.com/ggerganov/llama.cpp/issues/684
-        if temp == 0.0:
-            temp = 1.0
-            top_p = 0.0
-            top_k = 1
         return llama_cpp.llama_sample_top_p_top_k(
             ctx=self.ctx,
             last_n_tokens_data=(llama_cpp.llama_token * self.last_n_tokens_size)(