From c091c0a58898b8a0a76e18bd6724732d80fcfc28 Mon Sep 17 00:00:00 2001 From: Harry Mellor <19981378+hmellor@users.noreply.github.com> Date: Wed, 26 Mar 2025 14:26:48 +0000 Subject: [PATCH] Improve validation of TP in Transformers backend (#15540) Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com> --- vllm/model_executor/models/transformers.py | 5 ++++- 1 file changed, 4 insertions(+), 1 deletion(-) diff --git a/vllm/model_executor/models/transformers.py b/vllm/model_executor/models/transformers.py index 6ea149506581c..bdc390689104e 100644 --- a/vllm/model_executor/models/transformers.py +++ b/vllm/model_executor/models/transformers.py @@ -229,7 +229,10 @@ class TransformersModel(nn.Module): Apply the model's tensor parallelization plan. Currently only supports linear layers. """ - if self.tp_size > 1 and self.config.base_model_tp_plan is None: + if not self.model.supports_tp_plan: + if self.tp_size <= 1: + return + raise ValueError( f"{type(self.model)} does not support tensor parallel yet!")