Stability-AI/generative-models · error · ValueError

Decay must be between 0 and 1

Error message

Decay must be between 0 and 1

What it means

LitEma stores exponential-moving-average shadow weights and requires a decay factor in [0, 1]; the constructor validates this up front and raises ValueError otherwise. Decay of 0 would make the EMA meaningless and negative or >1 values are mathematically invalid for an EMA.

Source

Thrown at sgm/modules/ema.py:9

import torch
from torch import nn


class LitEma(nn.Module):
    def __init__(self, model, decay=0.9999, use_num_upates=True):
        super().__init__()
        if decay < 0.0 or decay > 1.0:
            raise ValueError("Decay must be between 0 and 1")

        self.m_name2s_name = {}
        self.register_buffer("decay", torch.tensor(decay, dtype=torch.float32))
        self.register_buffer(
            "num_updates",
            torch.tensor(0, dtype=torch.int)
            if use_num_upates
            else torch.tensor(-1, dtype=torch.int),
        )

        for name, p in model.named_parameters():
            if p.requires_grad:
                # remove as '.'-character is not allowed in buffers
                s_name = name.replace(".", "")
                self.m_name2s_name.update({name: s_name})
                self.register_buffer(s_name, p.clone().detach().data)

        self.collected_params = []

View on GitHub (pinned to e8cd657656)

Solutions

1. Pass decay as a fraction in [0,1], e.g. decay=0.9999
2. Fix the value in the model config dict/YAML that feeds LitEma
3. Clamp or validate the config value before constructing the model

Example fix

// before
LitEma(model, decay=99.9)
// after
LitEma(model, decay=0.999)

decay = config.get('decay', 0.9999)
if not isinstance(decay, (int, float)) or not (0.0 <= decay <= 1.0):
    raise ValueError(f'decay must be in [0,1], got {decay}')
ema = LitEma(model, decay=float(decay))

try:
    ema = LitEma(model, decay=cfg.model.decay)
except ValueError as e:
    logging.error('bad EMA decay in config: %s', cfg.model.decay)
    raise

When it happens

Trigger: Instantiating LitEma(model, decay=...) with decay < 0.0 or decay > 1.0, e.g. passing a percentage like 99.9 instead of 0.999, or a typo such as 0.99999.9999.

Common situations: Config YAML files holding decay as 9999 or 0.9999e2; unit misinterpretation (percent vs fraction); loading an old config where decay was expressed differently.

Related errors

• unknown merge strategy {self.merge_strategy}
• Order {order} too high for step {i}
• unknown merge strategy {merge_strategy}
• rearranging not available for {len(in_shape)}-dimensional in
• Unknown loss type {self.loss_type}

AI-assisted analysis of Stability-AI/generative-models@e8cd657656 (2026-08-29). Data as JSON: /api/errors/f5ccd3ceb49b2a1e. Report an issue: GitHub.