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
/**
* Local GGUF, MLX and managed engine auto-compaction is on or off only; the server chooses how it
* compacts (UNSLOTH_CONTEXT_POLICY, default "checkpoint").
*/
export const DEFAULT_AUTO_COMPACT_ENABLED = true;
export function ggufCompactionRequestFields(options: {
isGguf: boolean;
isMlx?: boolean;
isManagedEngine?: boolean;
autoCompactEnabled: boolean;
// /compact: asked for by the user, so it applies with auto-compaction off too
compactNow?: boolean;
}): {
context_overflow?: "error" | "truncate_oldest";
compact_context?: true;
} {
if (!options.isGguf && !options.isMlx && !options.isManagedEngine) return {};
if (options.compactNow) {
return { context_overflow: "truncate_oldest", compact_context: true };
}
if (!options.autoCompactEnabled) {
// explicit error avoids the UNSLOTH_CONTEXT_OVERFLOW fallback, which may compact
return { context_overflow: "error" };
}
// omit context_policy so the server applies UNSLOTH_CONTEXT_POLICY
return { context_overflow: "truncate_oldest" };
}
// leaves 25% of the window free, matching _EXTERNAL_COMPACTION_HEADROOM for self-hosted servers
const API_COMPACTION_HEADROOM = 0.25;
// request schema ceiling for compaction_threshold
const API_COMPACTION_THRESHOLD_MAX = 2_000_000;
export function apiCompactionRequestFields(options: {
autoCompactEnabled: boolean;
contextLength: number | null | undefined;
}): {
context_overflow?: "truncate_oldest";
compaction_threshold?: number;
context_window?: number;
} {
const { autoCompactEnabled, contextLength } = options;
if (!autoCompactEnabled) return {};
// without a catalogued window, the backend reads the self-hosted server's reported window
if (!contextLength || contextLength <= 0) return { context_overflow: "truncate_oldest" };
return {
context_overflow: "truncate_oldest",
compaction_threshold: Math.min(
API_COMPACTION_THRESHOLD_MAX,
Math.floor(contextLength * (1 - API_COMPACTION_HEADROOM)),
),
context_window: contextLength,
};
}