Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.
Python agentaichatgptdeepseekfine-tuninggemmaimage-generationllamallmllmsopenaipythonqwenreinforcement-learningself-hostedstable-diffusiontext-to-speechttsuiunsloth
// SPDX-License-Identifier: AGPL-3.0-only
// Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
import assert from "node:assert/strict";
import test from "node:test";
import {
ALLOW_OVERSIZED_LABEL,
allowOversizedField,
isMemoryEstimateRefusal,
MEMORY_REFUSAL_HEADER,
MEMORY_REFUSAL_KIND,
MemoryEstimateRefusalError,
shouldOfferGenerateAnyway,
shouldRunQueuedOversizedRetry,
} from "../src/features/images/lib/memory-refusal.ts";
import { readSrc, readText } from "./helpers/kit.ts";
test("only a 400 tagged as the memory estimate is the memory refusal", () => {
assert.equal(isMemoryEstimateRefusal(400, MEMORY_REFUSAL_KIND), true);
assert.equal(isMemoryEstimateRefusal(400, " Memory-Estimate "), true);
assert.equal(isMemoryEstimateRefusal(400, null), false);
assert.equal(isMemoryEstimateRefusal(400, "something-else"), false);
assert.equal(isMemoryEstimateRefusal(500, MEMORY_REFUSAL_KIND), false);
assert.equal(isMemoryEstimateRefusal(409, MEMORY_REFUSAL_KIND), false);
});
test("Generate anyway is offered for the refusal, and not again once the override was sent", () => {
const refusal = new MemoryEstimateRefusalError("Generating at 2048x2048 needs about 34 GB");
assert.equal(shouldOfferGenerateAnyway({ error: refusal, allowOversizedSent: false }), true);
assert.equal(shouldOfferGenerateAnyway({ error: refusal, allowOversizedSent: true }), false);
assert.equal(
shouldOfferGenerateAnyway({ error: new Error("width is required"), allowOversizedSent: false }),
false,
);
});
test("allow_oversized is sent only when the setting or the one-shot retry asks for it", () => {
assert.equal(allowOversizedField(false, false), undefined);
assert.equal(allowOversizedField(true, false), true);
assert.equal(allowOversizedField(false, true), true);
});
test("the setting label matches the one the backend quotes in its refusal", () => {
const backend = readText("../../backend/core/inference/diffusion_memory.py");
assert.ok(
backend.includes(`OVERSIZED_GENERATE_SETTING_LABEL = "${ALLOW_OVERSIZED_LABEL}"`),
"the refusal would point at a setting the page does not show",
);
assert.ok(backend.includes(`IMAGE_REFUSAL_HEADER = "${MEMORY_REFUSAL_HEADER}"`));
assert.ok(backend.includes(`IMAGE_REFUSAL_MEMORY_ESTIMATE = "${MEMORY_REFUSAL_KIND}"`));
const main = readText("../../backend/main.py");
assert.match(main, /expose_headers = \[[^\]]*"X-Unsloth-Refusal"/);
});
test("the page wires the setting, the request field and the toast action", () => {
const page = readSrc("features/images/images-page.tsx");
assert.match(page, /usePersistedToggle\(\s*"unsloth_images_allow_oversized"/);
assert.match(page, /allow_oversized: allowOversizedSent \? true : undefined/);
assert.match(page, /label: GENERATE_ANYWAY_LABEL/);
assert.match(page, /checked=\{allowOversized\}/);
const api = readSrc("features/images/api.ts");
assert.match(api, /allow_oversized\?: boolean;/);
assert.match(api, /throw new MemoryEstimateRefusalError\(/);
});
test("Generate anyway clicked during the refused run's cleanup waits for busy to clear", () => {
assert.equal(shouldRunQueuedOversizedRetry({ queued: true, busy: "generating" }), false);
assert.equal(shouldRunQueuedOversizedRetry({ queued: true, busy: null }), true);
assert.equal(shouldRunQueuedOversizedRetry({ queued: false, busy: null }), false);
const page = readSrc("features/images/images-page.tsx");
// The refused run already consumed Vary's one-shot: the retry gets it back, or it renders an unrelated image.
assert.match(page, /onClick: \(\) => \{\s*varyOnce\.current = vary;\s*setOversizedRetryQueued\(true\);/);
assert.match(
page,
/shouldRunQueuedOversizedRetry\(\{ queued: oversizedRetryQueued, busy \}\)[\s\S]*?\}, \[oversizedRetryQueued, busy, handleGenerateWithRecall\]\);/,
);
});
test("Generate anyway on an unloaded model keeps the override for the recalled generation", () => {
const page = readSrc("features/images/images-page.tsx");
assert.match(page, /load: loadSeq\.current \+ 1,[\s\S]*?allowOversized: oversizedOnce\.current,/);
assert.match(
page,
/oversizedOnce\.current = pending\.allowOversized === true;\s*void handleGenerate\(\)/,
);
});