hiyouga/LlamaFactory · error · ValueError

kernel_config.include_kernels must select at least one FLA k

Error message

kernel_config.include_kernels must select at least one FLA kernel.

What it means

After parsing `include_kernels`, the FLA plugin verifies that at least one kernel name was actually selected. A string that contains only commas, whitespace, or is empty produces an empty selection list, and the plugin raises a ValueError rather than silently patching nothing.

Source

Thrown at src/llamafactory/v1/plugins/model_plugins/kernels/ops/linear_attention/fla.py:75

        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.")

        unsupported = set(selected).difference(FLASH_LINEAR_ATTENTION_KERNELS)
        if unsupported:
            raise ValueError(f"Unsupported Flash Linear Attention kernels: {sorted(unsupported)}")
        if isinstance(chunk_size, bool) or not isinstance(chunk_size, int) or chunk_size not in SUPPORTED_CHUNK_SIZES:
            raise ValueError(f"chunk_size must be one of {SUPPORTED_CHUNK_SIZES}, got {chunk_size!r}.")

        from fsdp_turbo.ops.registry import get_op
        from fsdp_turbo.utils.patch import patch_model_members

        patched = 0
        named_modules = tuple(model.named_modules())
        for op_name in selected:
            module_attribute = FLA_MODULE_ATTRIBUTES[op_name]
            op = get_op(op_name)
            configured_op = partial(op, chunk_size=chunk_size) if op_name == CHUNK_GATED_DELTA_RULE else op
            targets = {
                f"{type(module).__module__}.{type(module).__name__}.{module_attribute}"

View on GitHub (pinned to f28afaf635)

Solutions

1. Remove the include_kernels key entirely so the default "auto" applies
2. Or set it to at least one valid FLA kernel name
3. Audit templated/interpolated configs for blank values reaching kernel_config

Example fix

# before
kernel_config:
  include_kernels: ""

# after
kernel_config:
  include_kernels: "auto"

names = [n.strip() for n in include_kernels.split(",") if n.strip()]
assert names, "include_kernels selected nothing; use 'auto' or real kernel names"

When it happens

Trigger: Passing include_kernels as "" (empty string), " ", or ",, ," — strings that split into zero non-empty names. The strip/filter step removes every token, leaving `selected == []`.

Common situations: Empty YAML value that parses to an empty string instead of the default 'auto' (e.g. `include_kernels:` with no value in some loaders), templated configs where a variable interpolates to blank, or hand-edited configs where names were deleted but the key remained.

Related errors

• Unsupported Flash Linear Attention kernels: {sorted(unsuppor
• chunk_size must be one of {SUPPORTED_CHUNK_SIZES}, got {chun
• Unknown backend: {model_args.infer_backend}
• Plugin configuration must have a 'name' field.
• kernel_config.include_kernels must be 'auto' or a comma-sepa

AI-assisted analysis of hiyouga/LlamaFactory@f28afaf635 (2026-08-14). Data as JSON: /api/errors/1f9641852d2506fc. Report an issue: GitHub.