From 5e00b561cddd2cecf8be7a341cb49d446613d6ef Mon Sep 17 00:00:00 2001 From: Woosuk Kwon Date: Wed, 21 Jan 2026 17:02:48 -0800 Subject: [PATCH] [Model Runner V2] Do not error on attention backends (#32820) Signed-off-by: Woosuk Kwon --- vllm/v1/worker/gpu/model_runner.py | 10 ---------- 1 file changed, 10 deletions(-) diff --git a/vllm/v1/worker/gpu/model_runner.py b/vllm/v1/worker/gpu/model_runner.py index 7a6b1732bfa..ae62fdc4d49 100644 --- a/vllm/v1/worker/gpu/model_runner.py +++ b/vllm/v1/worker/gpu/model_runner.py @@ -247,16 +247,6 @@ class GPUModelRunner(LoRAModelRunnerMixin, KVConnectorModelRunnerMixin): self.block_tables, ) - # TODO(woosuk): Support other backends. - supported_backends = ("FLASH_ATTN", "FLASHINFER", "FLASHINFER_MLA") - for backend in self.attn_backends.values(): - backend_name = backend.get_name() - if backend_name not in supported_backends: - raise NotImplementedError( - f"The {backend_name} attention backend is not supported yet. " - f"Supported backends are: {supported_backends}." - ) - self.kv_caches: list[torch.Tensor] = [] init_kv_cache( self.kv_caches,