xai-org/x-algorithm · error · ValueError

Unknown domain: {domain} for {name!r}

Error message

Unknown domain: {domain} for {name!r}

What it means

_convert_domains accepts, per tensor name, a dict (JSON IndexDomain), a ts.DimExpression, or a ts.IndexDomain. Anything else raises this ValueError naming the tensor and the unsupported domain object.

Source

Thrown at phoenix/python/training/xai-checkpointing/xai_checkpointing/load.py:193

                cur, cur_bytes = [], 0
            cur.append(item)
            cur_bytes += nbytes
        if cur:
            batches.append(cur)
    elif plan:
        batches.append(plan)
    return batches


def _convert_domains(domains: dict[str, Any], host_state: dict[str, jax.Array]) -> dict[str, Any]:
    for name, domain in domains.items():
        assert host_state.get(name) is not None, f"Cannot restrict domain of skipped tensor {name}"
        if isinstance(domain, dict):
            domains[name] = ts.IndexDomain(json=domain)
        elif isinstance(domain, ts.DimExpression):
            domains[name] = ts.IndexDomain(shape=host_state[name].shape)[domain]
        elif not isinstance(domain, ts.IndexDomain):
            raise ValueError(f"Unknown domain: {domain} for {name!r}")
    return domains


def _open_tensor(
    checkpoint_name: str,
    name: str,
    path: pathlib.Path,
    use_zarr3: bool,
    ts_context: ts.Context,
    dest: jax.Array,
    has_domain: bool,
    tspec_transform: Callable[[dict[str, Any]], dict[str, Any]] | None,
) -> ts.TensorStore:
    info = ocp.type_handlers.ParamInfo(
        name=checkpoint_name,
        path=path / checkpoint_name,
        parent_dir=path,
        is_ocdbt_checkpoint=True,

View on GitHub (pinned to 24c60942c5)

Solutions

1. Wrap the restriction in ts.DimExpression (e.g. ts.d[:4]) or pass ts.IndexDomain(shape=...)
2. If using a dict, ensure it's valid IndexDomain JSON
3. Use 'no_loading' for tensors you want skipped entirely
4. Check tensorstore version compatibility if passing its types

Example fix

# before
load_checkpoint(..., domains={"embedding": (slice(0, 1024),)})
# after
import tensorstore as ts
load_checkpoint(..., domains={"embedding": ts.IndexDomain(shape=[V, H])[ts.d[0][:1024]]})

import tensorstore as ts

def valid_domain(d):
    return isinstance(d, (dict, ts.IndexDomain)) or isinstance(d, ts.DimExpression)

def is_supported_domain(domain) -> bool:
    import tensorstore as ts
    return isinstance(domain, (dict, ts.DimExpression, ts.IndexDomain))

When it happens

Trigger: Passing load_checkpoint(..., domains={"params": (slice(0,4),)}) or a list/str/tuple as the domain for a tensor; only the three supported types are recognized.

Common situations: Intuitively passing Python slices/tuples expecting them to work, or passing a TensorStore type from a different tensorstore version with a changed class layout so isinstance checks fail.

Understand the failure class

Background: Invalid argument type errors: "must be of type string", "expected X, got Y", and ERR_INVALID_ARG_TYPE explained — this error's family across 15 libraries.

Related errors

• unexpected head param layout {sorted(params)} (expected {sor
• type checking expression %s failed: invalid argument type: %
• Non-optional parameter %s must be declared before optional p
• Duplicated argument name %s
• field name %s is not found in struct tuple class %s

AI-assisted analysis of xai-org/x-algorithm@24c60942c5 (2026-08-28). Data as JSON: /api/errors/685c7305d6b32b03. Report an issue: GitHub.