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
"""Wait limit and prefill lease; Infinity is unlimited in deadline math, None for sockets; durable
runs renew the lease during permitted waits without disabling orphan cleanup"""
from __future__ import annotations
import math
import time
from contextlib import contextmanager
from contextvars import ContextVar
from dataclasses import dataclass
DEFAULT_MINUTES = 20
MAX_MINUTES = 10080 # One week; 0 explicitly means Unlimited.
SETTING_KEY = "first_token_timeout_minutes"
def get_minutes() -> int:
try:
from storage.studio_db import get_app_setting
from utils.account_context import OWNER, run_as
value = run_as(OWNER, get_app_setting, SETTING_KEY, DEFAULT_MINUTES)
if type(value) is int and 0 <= value <= MAX_MINUTES:
return value
except Exception:
pass # An unavailable settings store must not break inference.
return DEFAULT_MINUTES
def set_minutes(value: int) -> None:
if type(value) is not int or not 0 <= value <= MAX_MINUTES:
raise ValueError(f"Use 0 for Unlimited or 1–{MAX_MINUTES} whole minutes.")
from storage.studio_db import upsert_app_settings
from utils.account_context import OWNER, run_as
run_as(OWNER, upsert_app_settings, {SETTING_KEY: value})
def timeout_seconds() -> float:
minutes = get_minutes()
return minutes * 60.0 if minutes else math.inf
def socket_timeout(seconds: float) -> float | None:
return seconds if math.isfinite(seconds) else None
@dataclass
class FirstTokenWait:
deadline: float | None = None
def active(self) -> bool:
# The worker thread can finish prefill while the supervisor checks it.
deadline = self.deadline
return deadline is not None and time.monotonic() < deadline
_current_wait: ContextVar[FirstTokenWait | None] = ContextVar("first_token_wait", default = None)
@contextmanager
def track_first_token_wait():
state = FirstTokenWait()
token = _current_wait.set(state)
try:
yield state
finally:
state.deadline = None
_current_wait.reset(token)
def start_first_token_wait(seconds: float | None = None) -> float:
deadline = time.monotonic() + (timeout_seconds() if seconds is None else seconds)
state = _current_wait.get()
if state is not None:
state.deadline = deadline
return deadline
def finish_first_token_wait() -> None:
state = _current_wait.get()
if state is not None:
state.deadline = None