From f3c1638927eee6ea31ad5a66e86e5b7ed6ebaa02 Mon Sep 17 00:00:00 2001 From: vllmellm Date: Thu, 13 Aug 2026 22:34:53 +0800 Subject: [PATCH] [ROCm] Enable V2 model runner for Kimi-K3 on ROCm (#51653) Signed-off-by: vllmellm --- vllm/config/vllm.py | 16 ---------------- 1 file changed, 16 deletions(-) diff --git a/vllm/config/vllm.py b/vllm/config/vllm.py index eeb2d840ff..34f5d2122d 100644 --- a/vllm/config/vllm.py +++ b/vllm/config/vllm.py @@ -83,26 +83,10 @@ DEFAULT_V2_MODEL_RUNNER_ARCHITECTURES = frozenset( } ) -# Architectures that default to V1 on ROCm: the V2 runner faults during the -# profile run. VLLM_USE_V2_MODEL_RUNNER=1 still forces V2. -# TODO: fix V2 enablement -ROCM_EXCLUDED_V2_MODEL_RUNNER_ARCHITECTURES = frozenset( - { - "KimiK3ForConditionalGeneration", - } -) - @lru_cache def default_v2_model_runner_architectures() -> frozenset[str]: """Architectures defaulting to the V2 model runner on this platform.""" - from vllm.platforms import current_platform - - if current_platform.is_rocm(): - return ( - DEFAULT_V2_MODEL_RUNNER_ARCHITECTURES - - ROCM_EXCLUDED_V2_MODEL_RUNNER_ARCHITECTURES - ) return DEFAULT_V2_MODEL_RUNNER_ARCHITECTURES