From 27448490f1c35b05abb40029d6267fd293054308 Mon Sep 17 00:00:00 2001 From: bk-201 Date: Thu, 11 Dec 2025 06:46:53 +0000 Subject: [PATCH] update argument name Signed-off-by: bk-201 --- tests/lora/test_qwenvl.py | 4 ++-- vllm/config/lora.py | 2 +- vllm/engine/arg_utils.py | 8 ++++---- vllm/lora/model_manager.py | 2 +- 4 files changed, 8 insertions(+), 8 deletions(-) diff --git a/tests/lora/test_qwenvl.py b/tests/lora/test_qwenvl.py index 2780f5e71dc21..4b3b92447789f 100644 --- a/tests/lora/test_qwenvl.py +++ b/tests/lora/test_qwenvl.py @@ -15,7 +15,7 @@ class TestConfig: max_num_seqs: int = 2 max_loras: int = 2 max_lora_rank: int = 32 - enable_mm_lora: bool = True + enable_tower_connector_lora: bool = True max_model_len: int = 8192 gpu_memory_utilization: float = 0.85 mm_processor_kwargs: dict[str, int] | None = None @@ -50,7 +50,7 @@ class Qwen2VLTester: enable_lora=True, max_loras=self.config.max_loras, max_lora_rank=self.config.max_lora_rank, - enable_mm_lora=self.config.enable_mm_lora, + enable_tower_connector_lora=self.config.enable_tower_connector_lora, trust_remote_code=True, gpu_memory_utilization=self.config.gpu_memory_utilization, mm_processor_kwargs=self.config.mm_processor_kwargs, diff --git a/vllm/config/lora.py b/vllm/config/lora.py index 23a46b9632cd1..97915ce925bc8 100644 --- a/vllm/config/lora.py +++ b/vllm/config/lora.py @@ -55,7 +55,7 @@ class LoRAConfig: per prompt. When run in offline mode, the lora IDs for n modalities will be automatically assigned to 1-n with the names of the modalities in alphabetic order.""" - enable_mm_lora: bool = False + enable_tower_connector_lora: bool = False """If `True`, LoRA support for multimodal models will be enabled. Currently, only the qwenvl series models support this feature. The default is False.""" diff --git a/vllm/engine/arg_utils.py b/vllm/engine/arg_utils.py index 2ba4cb0fbdca0..00b439a6ab9a1 100644 --- a/vllm/engine/arg_utils.py +++ b/vllm/engine/arg_utils.py @@ -486,7 +486,7 @@ class EngineArgs: fully_sharded_loras: bool = LoRAConfig.fully_sharded_loras max_cpu_loras: int | None = LoRAConfig.max_cpu_loras lora_dtype: str | torch.dtype | None = LoRAConfig.lora_dtype - enable_mm_lora: bool = LoRAConfig.enable_mm_lora + enable_tower_connector_lora: bool = LoRAConfig.enable_tower_connector_lora ray_workers_use_nsight: bool = ParallelConfig.ray_workers_use_nsight num_gpu_blocks_override: int | None = CacheConfig.num_gpu_blocks_override @@ -1008,8 +1008,8 @@ class EngineArgs: **lora_kwargs["lora_dtype"], ) lora_group.add_argument( - "--enable-mm-lora", - **lora_kwargs["enable_mm_lora"], + "--enable-tower-connector-lora", + **lora_kwargs["enable_tower_connector_lora"], ) lora_group.add_argument("--max-cpu-loras", **lora_kwargs["max_cpu_loras"]) lora_group.add_argument( @@ -1687,7 +1687,7 @@ class EngineArgs: default_mm_loras=self.default_mm_loras, fully_sharded_loras=self.fully_sharded_loras, lora_dtype=self.lora_dtype, - enable_mm_lora=self.enable_mm_lora, + enable_tower_connector_lora=self.enable_tower_connector_lora, max_cpu_loras=self.max_cpu_loras if self.max_cpu_loras and self.max_cpu_loras > 0 else None, diff --git a/vllm/lora/model_manager.py b/vllm/lora/model_manager.py index c308e2958cc25..479dcf88d02b3 100644 --- a/vllm/lora/model_manager.py +++ b/vllm/lora/model_manager.py @@ -128,7 +128,7 @@ class LoRAModelManager: if self.supports_mm and vllm_config is not None: model_config: ModelConfig = vllm_config.model_config self.mm_mapping: MultiModelKeys = self.model.get_mm_mapping() - if self.lora_config.enable_mm_lora: + if self.lora_config.enable_tower_connector_lora: self.info = MULTIMODAL_REGISTRY.create_processor(model_config).info self.supports_mm_lora = self.supports_mm and hasattr( self.info, "get_num_mm_encoder_tokens"