From 988aa34f3d84ba077311cbf68eb921ed44825d43 Mon Sep 17 00:00:00 2001 From: Nicolas Patry Date: Fri, 31 May 2024 22:56:31 +0000 Subject: [PATCH] Fix non decoding paths. --- server/text_generation_server/layers/attention/cuda.py | 1 + server/text_generation_server/layers/attention/rocm.py | 1 + 2 files changed, 2 insertions(+) diff --git a/server/text_generation_server/layers/attention/cuda.py b/server/text_generation_server/layers/attention/cuda.py index a564237c..d87d9199 100644 --- a/server/text_generation_server/layers/attention/cuda.py +++ b/server/text_generation_server/layers/attention/cuda.py @@ -159,6 +159,7 @@ def paged_attention( "auto", 1.0, ) + return out try: diff --git a/server/text_generation_server/layers/attention/rocm.py b/server/text_generation_server/layers/attention/rocm.py index c5c485de..36db12d0 100644 --- a/server/text_generation_server/layers/attention/rocm.py +++ b/server/text_generation_server/layers/attention/rocm.py @@ -129,6 +129,7 @@ def paged_attention( "auto", 1.0, ) + return out if ENGINE != "triton":