diff --git a/vllm/patches/deepseek_v4.py b/vllm/patches/deepseek_v4.py index 277e34a1..05662ab2 100644 --- a/vllm/patches/deepseek_v4.py +++ b/vllm/patches/deepseek_v4.py @@ -355,7 +355,8 @@ class DeepseekV4MegaMoEExperts(nn.Module): ) -> bool: # Progress bar for k8s/docker liveness during GPU upload if DeepseekV4MegaMoEExperts._weight_load_count == 0: - DeepseekV4MegaMoEExperts._weight_load_tqdm = tqdm( + from tqdm import tqdm as _tqdm + DeepseekV4MegaMoEExperts._weight_load_tqdm = _tqdm( total=self.num_local_experts * 20, # ~20 tensors per expert desc=" Loading Native NVFP4 Expert Weights", unit="tensor",