What causes "load error: {backend.load_progress()}" in unslothai/unsloth?

Enter a realistic output-token cap (typically 1024–32768, any value < 9007199254740991).

unslothai/unsloth · warning · RuntimeError

load error: {backend.load_progress()}

Error message

load error: {backend.load_progress()}

What it means

Thrown by parseMaxOutputTokens (chat-providers-dialog.tsx:500-502) when the digit string parses to a number outside Number.isSafeInteger — i.e. beyond 2^53-1 (9007199254740991). The earlier regex already guarantees digits, so this is purely a magnitude guard against absurd token limits that would silently lose precision as floats.

Source

Thrown at scripts/perf_verify.py:80

    out.mkdir(parents = True, exist_ok = True)
    backend = DiffusionBackend()
    token = os.environ.get("HF_TOKEN")

    def load(mode_speed = None, mode_mem = None):
        backend.begin_load(
            args.model,
            gguf_filename = args.gguf,
            hf_token = token,
            speed_mode = mode_speed,
            memory_mode = mode_mem,
        )
        deadline = time.time() + 2400
        while time.time() < deadline:
            ph = backend.load_progress().get("phase")
            if ph == "ready":
                return backend.status()
            if ph == "error":
                raise RuntimeError(f"load error: {backend.load_progress()}")
            time.sleep(0.5)
        raise RuntimeError("load timed out")

    def gen():
        torch.cuda.synchronize()
        t0 = time.time()
        img = backend.generate(
            prompt = args.prompt,
            width = args.width,
            height = args.height,
            steps = args.steps,
            guidance = 0.0,
            seed = args.seed,
            batch_size = 1,
        )["images"][0]
        torch.cuda.synchronize()
        return img, time.time() - t0

View on GitHub (pinned to 203007d190)

Solutions

Enter a realistic output-token cap (typically 1024–32768, any value < 9007199254740991).
Clear the field to use the provider default.
If you meant context length, set that in the model/context settings, not the provider Max Tokens field.

Example fix

// before
99999999999999999999

// after
32768

Defensive patterns

Strategy: validation

Validate before calling

function isSafeIntegerToken(input: string): boolean {
  const t = input.trim();
  return t === '' || (/^\d+$/.test(t) && Number.isSafeInteger(Number(t)));
}

Prevention

Cap the input's maxLength (e.g. 7-9 digits covers any sane token limit).
Clamp on blur rather than rejecting absurd values outright.
Guide users toward realistic values (1024-32768) with helper text.

When it happens

Trigger: Entering a digit string longer than ~16 characters, e.g. '99999999999999999999' or a pasted model context length like '18446744073709551615'. Number(trimmed) exceeds MAX_SAFE_INTEGER and the guard fires.

Common situations: Copy-pasting huge numbers from model spec sheets (e.g. theoretical context windows); holding a digit key; misunderstanding the field as a context-length setting.

Related errors

AI-assisted analysis of unslothai/unsloth@203007d190 (2026-08-15). Data as JSON: /api/errors/8287408978aa810d. Report an issue: GitHub.