hiyouga/LlamaFactory · error · RuntimeError

Flash Linear Attention and FSDPTurbo are required for this k

Error message

Flash Linear Attention and FSDPTurbo are required for this kernel.

What it means

FlashLinearAttentionKernel.check_deps() tries importing fla.ops.gated_delta_rule, fsdp_turbo.ops.fla, the fsdp_turbo op registry, and its patch helper. If any is missing it raises RuntimeError, chaining the original ImportError so the exact missing module is visible.

Source

Thrown at src/llamafactory/v1/plugins/model_plugins/kernels/ops/linear_attention/fla.py:58

@KernelPlugin("flash-linear-attention").register()
class FlashLinearAttentionKernel(BaseKernel):
    """Install selected FLA callables through FSDPTurbo's device operator registry."""

    @staticmethod
    def check_device() -> None:
        current = get_current_accelerator().type
        if current not in (DeviceType.CUDA, DeviceType.NPU):
            raise RuntimeError(f"FlashLinearAttentionKernel requires CUDA or NPU, current accelerator is {current}.")

    @staticmethod
    def check_deps() -> None:
        try:
            import fla.ops.gated_delta_rule  # noqa: F401
            import fsdp_turbo.ops.fla  # noqa: F401
            from fsdp_turbo.ops.registry import get_op  # noqa: F401
            from fsdp_turbo.utils.patch import patch_model_members  # noqa: F401
        except ImportError as exc:
            raise RuntimeError("Flash Linear Attention and FSDPTurbo are required for this kernel.") from exc

    @staticmethod
    def _apply(**kwargs) -> HFModel:
        model = kwargs["model"]
        config = kwargs.get("config") or {}
        include_kernels = config.get("include_kernels", "auto")
        chunk_size = config.get("chunk_size", 64)

        if include_kernels == "auto" or include_kernels is True:
            selected = list(FLASH_LINEAR_ATTENTION_KERNELS)
        elif isinstance(include_kernels, str):
            selected = [name.strip() for name in include_kernels.split(",") if name.strip()]
        else:
            raise TypeError("kernel_config.include_kernels must be 'auto' or a comma-separated string.")

        if not selected:
            raise ValueError("kernel_config.include_kernels must select at least one FLA kernel.")

View on GitHub (pinned to f28afaf635)

Solutions

1. Install both: `pip install flash-linear-attention fsdp-turbo`.
2. Verify each import the check performs: `python -c "import fla.ops.gated_delta_rule, fsdp_turbo.ops.fla; from fsdp_turbo.ops.registry import get_op"`.
3. Upgrade fsdp-turbo if the fla submodule is missing in your version.
4. Remove the kernel from config if it was unintended.

Example fix

# before
# missing deps, kernel_config.name: flash-linear-attention

# after
pip install flash-linear-attention fsdp-turbo
python -c "import fsdp_turbo.ops.fla"  # verify

import importlib.util
required = ['fla.ops.gated_delta_rule', 'fsdp_turbo.ops.fla', 'fsdp_turbo.ops.registry', 'fsdp_turbo.utils.patch']
missing = [m for m in required if importlib.util.find_spec(m) is None]
assert not missing, f'pip install flash-linear-attention fsdp-turbo (missing: {missing})'

def fla_deps_installed() -> bool:
    """True when fla and fsdp_turbo modules the kernel needs are importable."""
    return not missing  # from validationCode's probe

try:
    model = KernelPlugin('flash-linear-attention').apply(model=model)
except RuntimeError as e:
    if 'required for this kernel' in str(e) and e.__cause__ is not None:
        raise SystemExit(f'pip install flash-linear-attention fsdp-turbo: {e.__cause__}') from None
    raise

When it happens

Trigger: Selecting 'flash-linear-attention' without installing both optional packages flash-linear-attention and fsdp-turbo (or with incompatible versions where fsdp_turbo.ops.fla is absent).

Common situations: Optional-extras not installed in the base env; pip resolving an old fsdp-turbo without the fla ops module; reinstalling the env without kernel extras.

Related errors

• Liger kernel is not installed.
• FlashLinearAttentionKernel requires CUDA or NPU, current acc
• The installed Transformers-KT does not provide `TrainingArgu
• The installed Transformers-KT does not provide `configure_kt
• HFModel instance is required for {cls.__name__}.

AI-assisted analysis of hiyouga/LlamaFactory@f28afaf635 (2026-08-14). Data as JSON: /api/errors/bdb93c41f39d6457. Report an issue: GitHub.