A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Bugfix] Fix Qwen3-VL regression from #24982 (#25814)
Signed-off-by: Roger Wang <hey@rogerw.io>
Roger Wang committed
Sep 27, 2025 at 20:21 UTC
61447540146602992116ae1068e45419e6bb1a5b
1 file changed
+4
-4
vllm/model_executor/models/qwen3_moe.py
+4
-4
@@ -107,7 +107,7 @@ class Qwen3MoeSparseMoeBlock(nn.Module):
107
):
108
super().__init__()
109
110
- config = vllm_config.model_config.hf_config
110
+ config = vllm_config.model_config.hf_text_config
111
parallel_config = vllm_config.parallel_config
112
quant_config = vllm_config.quant_config
113
@@ -293,7 +293,7 @@ class Qwen3MoeDecoderLayer(nn.Module):
293
def __init__(self, vllm_config: VllmConfig, prefix: str = "") -> None:
294
super().__init__()
295
296
- config = vllm_config.model_config.hf_config
296
+ config = vllm_config.model_config.hf_text_config
297
cache_config = vllm_config.cache_config
298
quant_config = vllm_config.quant_config
299
@@ -372,7 +372,7 @@ class Qwen3MoeModel(nn.Module):
372
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
373
super().__init__()
374
375
- config = vllm_config.model_config.hf_config.get_text_config()
375
+ config = vllm_config.model_config.hf_text_config
376
quant_config = vllm_config.quant_config
377
parallel_config = vllm_config.parallel_config
378
eplb_config = parallel_config.eplb_config
@@ -586,7 +586,7 @@ class Qwen3MoeForCausalLM(nn.Module, SupportsPP, SupportsLoRA,
586
587
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
588
super().__init__()
589
- config = vllm_config.model_config.hf_config
589
+ config = vllm_config.model_config.hf_text_config
590
quant_config = vllm_config.quant_config
591
self.config = config
592
self.quant_config = quant_config