unslothai/unsloth · warning · RuntimeError

VirusTotal request failed after {max_attempts} attempt(s): {

Error message

VirusTotal request failed after {max_attempts} attempt(s): {last_error}

What it means

Thrown by use-chat-model-runtime.ts:1142-1144 via getTrustRemoteCodeRequiredMessage (line 276-278) when validateModel flags requires_trust_remote_code or requires_security_review and the user does not complete the remote-code consent dialog — confirmRemoteCodeIfNeeded resolved approved=false. Models with custom modeling code (trust_remote_code) require an explicit review/approval that records a fingerprint; without approval the worker refuses the load. The dialog fires even when trustRemoteCode is preset, because the worker needs the matching fingerprint only the dialog produces.

Source

Thrown at scripts/virustotal_scan.py:441

                if attempt < max_attempts:
                    self._backoff(backoff * (2 ** (attempt - 1)), deadline)
                continue
            if status == 0 or status >= 500:
                last_error = last_error or f"HTTP {status}"
                if attempt < max_attempts:
                    self._backoff(backoff * (2 ** (attempt - 1)), deadline)
                continue
            if status >= 400 and status not in allow_status:
                raise RuntimeError(f"VirusTotal returned HTTP {status} for {_redact_url(url)}")

            if not payload:
                return status, None
            try:
                return status, json.loads(payload.decode("utf-8", "replace"))
            except json.JSONDecodeError:
                return status, None

        raise RuntimeError(
            f"VirusTotal request failed after {max_attempts} attempt(s): {last_error}"
        )

    def lookup_hash(
        self,
        sha256: str,
        deadline: float | None = None,
    ) -> object | None:
        """Return the existing file report, or None when VirusTotal has never seen it.

        Doing this first is both a quota saving and a disclosure saving: a bundle that
        VirusTotal already holds gains nothing from being uploaded again.
        """
        status, payload = self.request(
            "GET",
            f"{API_ROOT}/files/{sha256}",
            allow_status = (404,),
            deadline = deadline,

View on GitHub (pinned to 203007d190)

Solutions

Load the model again, open the review dialog, read the custom code, and approve it — the recorded fingerprint lets the load proceed.
If the code looks unsafe, do not approve: choose a different model or a transformers-native/GGUF variant that needs no custom code.
Keep the studio backend updated so more architectures are natively supported and the consent gate disappears for them.

Defensive patterns

Strategy: try-catch

Validate before calling

async function needsRemoteCodeConsent(modelId: string): Promise<boolean> {
  const v = await validateModel({ model_path: modelId });
  return Boolean(v.requires_trust_remote_code || v.requires_security_review);
}

Type guard

function isRemoteCodeApprovalError(e: unknown): e is Error {
  return e instanceof Error && /custom code was not approved/.test(e.message);
}

Try / catch

try {
  await loadModel(selection);
} catch (error) {
  if (isRemoteCodeApprovalError(error)) {
    // safe to retry: re-run load, complete the review dialog, approve
  } else throw error;
}

Prevention

Approve the code-review dialog only after actually reading the custom modeling files.
Cache the approval fingerprint so repeat loads of the same model don't re-prompt.
Prefer transformers-native or GGUF builds of architectures you don't trust.

When it happens

Trigger: Loading a Hub model with custom code (or a flagged unsafe file) where validation sets requires_trust_remote_code / requires_security_review, and the user cancels the code-review dialog or rejects the code after reviewing it.

Common situations: Newer/niche architectures not yet in the installed transformers (Phi-style custom models); users wary of remote code; the security review flagging a file in the repo; dialog dismissed with Escape.

Related errors

AI-assisted analysis of unslothai/unsloth@203007d190 (2026-08-15). Data as JSON: /api/errors/4dda80e66b9ce121. Report an issue: GitHub.