From 9f667ee1512fc67b8416ed8ff1e06a7c2f0af31e Mon Sep 17 00:00:00 2001 From: amy-why-3459 Date: Fri, 27 Mar 2026 11:46:24 +0800 Subject: [PATCH] [BugFix][Qwen3-Omni]Fixed the issue of incorrect answers for single words. (#2239) Signed-off-by: amy-why-3459 --- vllm_omni/benchmarks/patch/patch.py | 3 +-- vllm_omni/model_executor/models/qwen3_omni/qwen3_omni.py | 4 +--- 2 files changed, 2 insertions(+), 5 deletions(-) diff --git a/vllm_omni/benchmarks/patch/patch.py b/vllm_omni/benchmarks/patch/patch.py index 4e13dface..539af11f8 100644 --- a/vllm_omni/benchmarks/patch/patch.py +++ b/vllm_omni/benchmarks/patch/patch.py @@ -376,9 +376,8 @@ async def benchmark( limit_per_host=max_concurrency or 0, ttl_dns_cache=300, use_dns_cache=True, - keepalive_timeout=60, enable_cleanup_closed=True, - force_close=False, + force_close=True, ssl=ssl_setting, ) diff --git a/vllm_omni/model_executor/models/qwen3_omni/qwen3_omni.py b/vllm_omni/model_executor/models/qwen3_omni/qwen3_omni.py index 2b5bcd60a..5a22ce024 100644 --- a/vllm_omni/model_executor/models/qwen3_omni/qwen3_omni.py +++ b/vllm_omni/model_executor/models/qwen3_omni/qwen3_omni.py @@ -1047,9 +1047,7 @@ class Qwen3OmniMoeForConditionalGeneration( dim=0, ) else: - trailing_text_hidden = torch.zeros( - tts_eos_embed.shape, device=tts_eos_embed.device, dtype=tts_eos_embed.dtype - ) + trailing_text_hidden = tts_eos_embed input_embeds = assistant_text_hidden + assistant_codec_hidden input_ids = torch.full(