Stability-AI/generative-models · warning

Attention mode '{attn_mode}' is not available. Falling back

Error message

Attention mode '{attn_mode}' is not available. Falling back to native attention. This is not a problem in Pytorch >= 2.0. FYI, you are running with PyTorch version {torch.__version__}.

What it means

CrossAttention/MemoryEfficientAttention __init__ asserts attn_mode is a known mode, then warns and coerces the mode to 'softmax' if a non-default mode (e.g. 'xformers') was requested while xformers is not installed. The model still constructs, just with slower native attention.

Source

Thrown at sgm/modules/attention.py:478

    }

    def __init__(
        self,
        dim,
        n_heads,
        d_head,
        dropout=0.0,
        context_dim=None,
        gated_ff=True,
        checkpoint=True,
        disable_self_attn=False,
        attn_mode="softmax",
        sdp_backend=None,
    ):
        super().__init__()
        assert attn_mode in self.ATTENTION_MODES
        if attn_mode != "softmax" and not XFORMERS_IS_AVAILABLE:
            logpy.warn(
                f"Attention mode '{attn_mode}' is not available. Falling "
                f"back to native attention. This is not a problem in "
                f"Pytorch >= 2.0. FYI, you are running with PyTorch "
                f"version {torch.__version__}."
            )
            attn_mode = "softmax"
        elif attn_mode == "softmax" and not SDP_IS_AVAILABLE:
            logpy.warn(
                "We do not support vanilla attention anymore, as it is too "
                "expensive. Sorry."
            )
            if not XFORMERS_IS_AVAILABLE:
                assert (
                    False
                ), "Please install xformers via e.g. 'pip install xformers==0.0.16'"
            else:
                logpy.info("Falling back to xformers efficient attention.")
                attn_mode = "softmax-xformers"

View on GitHub (pinned to e8cd657656)

Solutions

1. Install xformers so the requested attention mode actually becomes available
2. Change attn_mode to 'softmax' explicitly in the config to silence the fallback warning
3. Verify XFORMERS_IS_AVAILABLE is True after importing sgm.modules.attention before relying on xformers speedups

Example fix

// before
params:
  attn_mode: "xformers"   # xformers not installed -> silent fallback
// after
pip install xformers
# or
params:
  attn_mode: "softmax"

import sgm.modules.attention as A
if cfg_attn_mode != "softmax" and not A.XFORMERS_IS_AVAILABLE:
    print(f"{cfg_attn_mode} unavailable, will fall back to softmax")

def attn_mode_usable(mode: str) -> bool:
    import sgm.modules.attention as A
    return mode == "softmax" or A.XFORMERS_IS_AVAILABLE

try:
    attn = CrossAttention(..., attn_mode=cfg_attn_mode)
finally:
    if attn.attn_mode != cfg_attn_mode:
        logger.warning("attention mode fell back to %s", attn.attn_mode)

When it happens

Trigger: Config `attn_mode: 'xformers'` (or 'torch-sdp'/'vanilla' variants requiring backends) in a CrossAttention with XFORMERS_IS_AVAILABLE False — i.e. xformers import failed at module load.

Common situations: Reusing SD configs that specify xformers attention on machines without xformers; CI/CPU environments where xformers wheels are unavailable.

Related errors

• We do not support vanilla attention anymore, as it is too ex
• unknown merge strategy {self.merge_strategy}
• Unknown loss type {self.loss_type}
• provide num_res_blocks either as an int (globally constant)
• need either 'input_key' or 'input_keys' for embedder {embedd

AI-assisted analysis of Stability-AI/generative-models@e8cd657656 (2026-08-29). Data as JSON: /api/errors/0c6585d01c6e61f2. Report an issue: GitHub.