open-mmlab/mmdetection · error · NotImplementedError

If you want to reduce GPU memory usage,

Error message

If you want to reduce GPU memory usage,                     please install fairscale by executing the                     following command: pip install fairscale.

What it means

GroundingDinoTransformerEncoder raises NotImplementedError when num_cp > 0 but fairscale is missing. Both the encoder layers and the text-image fusion layers would need checkpoint_wrapper from fairscale.

Source

Thrown at mmdet/models/layers/transformer/grounding_dino_layers.py:160

    def _init_layers(self) -> None:
        """Initialize encoder layers."""
        self.layers = ModuleList([
            DeformableDetrTransformerEncoderLayer(**self.layer_cfg)
            for _ in range(self.num_layers)
        ])
        self.text_layers = ModuleList([
            DetrTransformerEncoderLayer(**self.text_layer_cfg)
            for _ in range(self.num_layers)
        ])
        self.fusion_layers = ModuleList([
            SingleScaleBiAttentionBlock(**self.fusion_layer_cfg)
            for _ in range(self.num_layers)
        ])
        self.embed_dims = self.layers[0].embed_dims
        if self.num_cp > 0:
            if checkpoint_wrapper is None:
                raise NotImplementedError(
                    'If you want to reduce GPU memory usage, \
                    please install fairscale by executing the \
                    following command: pip install fairscale.')
            for i in range(self.num_cp):
                self.layers[i] = checkpoint_wrapper(self.layers[i])
                self.fusion_layers[i] = checkpoint_wrapper(
                    self.fusion_layers[i])

    def forward(self,
                query: Tensor,
                query_pos: Tensor,
                key_padding_mask: Tensor,
                spatial_shapes: Tensor,
                level_start_index: Tensor,
                valid_ratios: Tensor,
                memory_text: Tensor = None,
                text_attention_mask: Tensor = None,
                pos_text: Tensor = None,

View on GitHub (pinned to cfd5d3a985)

Solutions

1. pip install fairscale
2. Or set num_cp=0

Example fix

# before
num_cp=1  # no fairscale installed
# after
pip install fairscale  # or num_cp=0

if num_cp > 0 and importlib.util.find_spec('fairscale') is None:
    raise RuntimeError('install fairscale or set num_cp=0')

When it happens

Trigger: Configuring Grounding DINO with num_cp>0 in the encoder cfg while `checkpoint_wrapper` failed to import because fairscale is not installed.

Common situations: Running Grounding DINO training configs tuned for low-memory GPUs in a bare environment; docker images missing optional deps.

Related errors

• If you want to reduce GPU memory usage,
• If you want to reduce GPU memory usage,
• imagecorruptions is not installed
• albumentations is not installed
• There is not post_norm in {name}

AI-assisted analysis of open-mmlab/mmdetection@cfd5d3a985 (2026-08-27). Data as JSON: /api/errors/d28cb8d89b3d2a16. Report an issue: GitHub.