From 22631f80a01a04b06398952e77d7890ab660ab10 Mon Sep 17 00:00:00 2001 From: Isotr0py Date: Mon, 27 Apr 2026 15:27:06 +0800 Subject: [PATCH] [Bugfix] Remove invalid deepstack boundary check for Qwen3-VL (#40932) Signed-off-by: Isotr0py --- vllm/model_executor/models/qwen3_omni_moe_thinker.py | 11 ----------- vllm/model_executor/models/qwen3_vl.py | 11 ----------- 2 files changed, 22 deletions(-) diff --git a/vllm/model_executor/models/qwen3_omni_moe_thinker.py b/vllm/model_executor/models/qwen3_omni_moe_thinker.py index ff8962c82b..44e656ab82 100755 --- a/vllm/model_executor/models/qwen3_omni_moe_thinker.py +++ b/vllm/model_executor/models/qwen3_omni_moe_thinker.py @@ -1778,11 +1778,6 @@ class Qwen3OmniMoeThinkerForConditionalGeneration( return None # If vision tower is skipped if getattr(self, "deepstack_input_embeds_num_tokens", 0) == 0: return None - if num_tokens > self.deepstack_input_embeds_num_tokens: - raise ValueError( - "Requested more deepstack tokens than available in buffer: " - f"{num_tokens=} > {self.deepstack_input_embeds_num_tokens=}" - ) # get deepstack_input_embeds from buffer, and clear the buffer return IntermediateTensors( @@ -1824,12 +1819,6 @@ class Qwen3OmniMoeThinkerForConditionalGeneration( # clear deepstack_input_embeds in buffer if num_tokens > 0: - if num_tokens > self.deepstack_input_embeds_num_tokens: - raise ValueError( - "Requested to clear more deepstack tokens than available in " - "buffer: " - f"{num_tokens=} > {self.deepstack_input_embeds_num_tokens=}" - ) for idx in range(self.deepstack_num_level): self.deepstack_input_embeds[idx][:num_tokens].zero_() self.deepstack_input_embeds_num_tokens = 0 diff --git a/vllm/model_executor/models/qwen3_vl.py b/vllm/model_executor/models/qwen3_vl.py index 4a0de15f02..2575f634be 100644 --- a/vllm/model_executor/models/qwen3_vl.py +++ b/vllm/model_executor/models/qwen3_vl.py @@ -1716,11 +1716,6 @@ class Qwen3VLForConditionalGeneration( return None # If vision tower is skipped if getattr(self, "deepstack_input_embeds_num_tokens", 0) == 0: return None - if num_tokens > self.deepstack_input_embeds_num_tokens: - raise ValueError( - "Requested more deepstack tokens than available in buffer: " - f"{num_tokens=} > {self.deepstack_input_embeds_num_tokens=}" - ) # get deepstack_input_embeds from buffer, and clear the buffer return IntermediateTensors( @@ -1762,12 +1757,6 @@ class Qwen3VLForConditionalGeneration( # clear deepstack_input_embeds in buffer if num_tokens > 0: - if num_tokens > self.deepstack_input_embeds_num_tokens: - raise ValueError( - "Requested to clear more deepstack tokens than available in " - "buffer: " - f"{num_tokens=} > {self.deepstack_input_embeds_num_tokens=}" - ) for idx in range(self.deepstack_num_level): self.deepstack_input_embeds[idx][:num_tokens].zero_() self.deepstack_input_embeds_num_tokens = 0