diff --git a/vllm/model_executor/models/commandr.py b/vllm/model_executor/models/commandr.py
index 317269ec3b..66adb9a3ca 100644
--- a/vllm/model_executor/models/commandr.py
+++ b/vllm/model_executor/models/commandr.py
@@ -56,6 +56,7 @@ from vllm.sequence import IntermediateTensors
 from .interfaces import SupportsLoRA, SupportsPP, SupportsQuant
 from .utils import (
     AutoWeightsLoader,
+    WeightsMapper,
     extract_layer_index,
     is_pp_missing_parameter,
     make_empty_intermediate_tensors_factory,
@@ -397,6 +398,9 @@ class CohereForCausalLM(nn.Module, SupportsLoRA, SupportsPP, SupportsQuant):
     }
     # LoRA specific attributes
     embedding_modules = {"embed_tokens": "input_embeddings"}
+    # ModelOpt NVFP4 checkpoints carry raw quantizer-module state
+    # (e.g. "*.weight_quantizer._double_scale"); drop them before loading. See #41925.
+    hf_to_vllm_mapper = WeightsMapper(orig_to_new_substr={"_quantizer.": None})
 
     def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
         super().__init__()
@@ -453,4 +457,4 @@ class CohereForCausalLM(nn.Module, SupportsLoRA, SupportsPP, SupportsQuant):
         loader = AutoWeightsLoader(
             self, skip_prefixes=["lm_head", "rotary_emb.inv_freq"]
         )
-        return loader.load_weights(weights)
+        return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)