unslothai/unsloth · error · ValueError

flow_shift must be a positive number or 'auto', got {self.fl

Error message

flow_shift must be a positive number or 'auto', got {self.flow_shift!r}

What it means

flow_shift accepts a positive number or the string 'auto' (family-default dynamic shift). If a string is supplied that is neither 'auto' nor parseable by float(), the config raises with the original value echoed. Numbers pass through a separate finite/positive check.

Source

Thrown at studio/backend/core/training/diffusion_train_common.py:1151

            # scheme cleared only for rendering reach a trainer.
            from core.inference.diffusion_transformer_quant import _family_train_denied

            if _family_train_denied(resolved_family, base_precision):
                raise ValueError(
                    f"base_precision={base_precision!r} is not validated for training "
                    f"{resolved_family}. Use 'nf4', 'int8', 'bf16', or 'auto'."
                )
        # flow_shift: None resolves to the family default ("auto" only for qwen-image, whose scheduler skips its static shift under use_dynamic_shifting); an explicit value is validated and kept.
        flow_shift = self.flow_shift
        if flow_shift is None:
            flow_shift = "auto" if resolved_family in AUTO_FLOW_SHIFT_FAMILIES else 1.0
        if isinstance(flow_shift, str):
            flow_shift = flow_shift.strip().lower()
            if flow_shift != "auto":
                try:
                    flow_shift = float(flow_shift)
                except ValueError as exc:
                    raise ValueError(
                        f"flow_shift must be a positive number or 'auto', got {self.flow_shift!r}"
                    ) from exc
        if not isinstance(flow_shift, str):
            flow_shift = float(flow_shift)
            # isfinite as well as positive: JSON accepts 1e309, which floats to inf and would poison every sampled sigma while progress looks normal.
            if not math.isfinite(flow_shift) or flow_shift <= 0:
                raise ValueError(
                    "flow_shift must be a finite number > 0 (1.0 disables the shift), or 'auto'"
                )
        try:
            cfg_dropout = float(self.cfg_dropout or 0.0)
        except (TypeError, ValueError) as exc:
            raise ValueError(f"cfg_dropout must be a number, got {self.cfg_dropout!r}") from exc
        if not 0.0 <= cfg_dropout <= 1.0:
            raise ValueError("cfg_dropout must be between 0 and 1")
        weighting_scheme = str(self.weighting_scheme or "none").strip().lower()
        if weighting_scheme not in ("none", "bell"):
            raise ValueError("weighting_scheme must be one of none / bell")

View on GitHub (pinned to 203007d190)

Solutions

1. Set flow_shift to a positive float (e.g. 1.0) or the literal string 'auto'.
2. Omit flow_shift (None) to take the family default: 'auto' for AUTO_FLOW_SHIFT_FAMILIES like qwen-image, 1.0 otherwise.
3. If the value comes from user input, normalize locale decimal separators and strip whitespace before assigning.

Example fix

# before
cfg.flow_shift = '1,5'  # or 'default'
# after
cfg.flow_shift = 1.5  # or 'auto', or None for the family default

def normalize_flow_shift(v):
    if v is None:
        return None
    if isinstance(v, str):
        s = v.strip().lower()
        if s == 'auto':
            return 'auto'
        v = float(s)  # raises here instead of inside the trainer
    return float(v)

def is_valid_flow_shift(v) -> bool:
    if v is None:
        return True
    if isinstance(v, str):
        s = v.strip().lower()
        if s == 'auto':
            return True
        try:
            v = float(s)
        except ValueError:
            return False
    return isinstance(v, (int, float))

When it happens

Trigger: Setting flow_shift to a non-numeric string such as 'default', '1,0' (locale decimal comma), or '' (empty string after strip/lower) in DiffusionLoraConfig.

Common situations: Studio UI or hand-edited YAML/JSON config with a typo'd or localized number; passing None-like placeholder strings; copy-pasting a scheduler name into the field.

Related errors

• base_precision={base_precision!r} trains in bf16 compute; se
• flow_shift must be a finite number > 0 (1.0 disables the shi
• cfg_dropout must be a number, got {self.cfg_dropout!r}
• cfg_dropout must be between 0 and 1
• weighting_scheme must be one of none / bell

AI-assisted analysis of unslothai/unsloth@203007d190 (2026-08-15). Data as JSON: /api/errors/e04430ebb1fe1124. Report an issue: GitHub.