unslothai/unsloth · error · HTTPException

AI assist failed: {scrubbed}

Error message

AI assist failed: {scrubbed}

What it means

HTTP 500 from the AI-assist mapping advisor when its multi-pass LLM analysis raised an exception that fell through the mapped branches (hf_error_status, FileNotFoundError→404, ValueError→400). Like 807, the message is secret-scrubbed and both returned to the client and logged. The AI assist loads a 7B helper model, so failures are often model/runtime/resource related rather than dataset related.

Source

Thrown at studio/backend/hub/services/datasets/formatting.py:592

                warning = result.get("warning"),
            )

        return AiAssistMappingResponse(
            success = False,
            warning = "AI could not determine column roles. Please assign them manually.",
        )

    except Exception as e:
        scrubbed = download_registry.scrub_secrets(str(e), hf_token = hf_token)
        status = hf_error_status(e)
        if status is None and isinstance(e, FileNotFoundError):
            status = 404
        elif status is None and isinstance(e, ValueError):
            status = 400
        if status is not None:
            raise HTTPException(status_code = status, detail = scrubbed)
        logger.error("AI assist mapping failed: %s", scrubbed)
        raise HTTPException(
            status_code = 500,
            detail = "AI assist failed: " + scrubbed,
        )

View on GitHub (pinned to 203007d190)

Solutions

1. Check the logged `AI assist mapping failed: ...` line — it carries the scrubbed root cause.
2. Ensure the helper model is already cached locally (pre-download it) or that the model host is reachable from the backend.
3. Free GPU/CPU memory: close other training jobs before running AI assist.
4. If the underlying error is ValueError/FileNotFound from the dataset itself, fix the dataset reference first — those surface as 4xx and are actionable.

# Pre-flight: the same dataset ref must pass a cheap format check before the expensive AI pass
fmt = check_dataset_format(client, req)  # if this 4xx/5xx, AI assist will fail too

try:
    resp = run_ai_assist(client, req)
except HTTPStatusError as e:
    if e.response.status_code >= 500:
        disable_ai_assist_with_reason(e.response.json()["detail"])  # model/env issue, not dataset
    else:
        fix_dataset_ref_and_retry(e)

When it happens

Trigger: The helper model fails to download or load (network, disk), inference OOMs on a GPU-poor machine, the datasets-derived prompt triggers an unexpected library error, or the hub raises an unmapped error while fetching the dataset.

Common situations: First run of AI assist with no cached model and a slow/blocked model host; small VRAM boxes; proxies blocking the model download; concurrent assist runs exhausting memory.

Related errors

• The inference worker stopped unexpectedly while loading the
• Failed to export GGUF model
• deadline reached while pacing before {method} {_redact_url(u
• VirusTotal returned HTTP {status} for {_redact_url(url)}
• VirusTotal request failed after {max_attempts} attempt(s): {

AI-assisted analysis of unslothai/unsloth@203007d190 (2026-08-15). Data as JSON: /api/errors/d31eaa0241884592. Report an issue: GitHub.