# New Release: Ultralytics v8.3.213

**URL:** <https://community.ultralytics.com/t/new-release-ultralytics-v8-3-213/1547>\
**Category:** Discussion\
**Tags:** releases, announcements, ultralytics-official\
**Created:** [October 13, 2025, 1:14pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-3-213/1547 "2025-10-13T13:14:03Z")\
**Posts on this page:** 1\
**Page:** 1

<div class="post-metadata">

**Author:** ![glenn-jocher](https://sea1.discourse-cdn.com/flex001/user_avatar/community.ultralytics.com/glenn-jocher/32/83_2.png) [@glenn-jocher](https://community.ultralytics.com/u/glenn-jocher)\
**Post date:** [October 13, 2025, 1:14pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-3-213/1547/1 "2025-10-13T13:14:04Z")

</div>

# Ultralytics v8.3.213 — Stability first: automatic NaN recovery, faster Objects365, unified dataset URLs 🚑🔁⚡

A stability-focused release that automatically recovers from NaN/Inf training issues, cleans up resume logic, speeds up Objects365 setup, and unifies dataset download URLs. Upgrade with confidence and keep training moving.

- Version: v8.3.213
- Focus: Training robustness, dataset reliability, and setup speed
- Release: See the full notes on the [v8.3.213 release page](https://github.com/ultralytics/ultralytics/releases/tag/v8.3.213)

## TL;DR

- Auto-recovery from NaN/Inf losses and metric collapse during training (DDP-aware, capped retries) 🛡
- Safer resumes with centralized checkpoint loading and scheduler reset
- Objects365 preparation runs significantly faster with multithreading 🚀
- Unified `ASSETS_URL` across dataset YAMLs for more reliable downloads 🔗

* * *

## Highlights

### Training robustness and resume improvements

- Automatically detects NaN/Inf losses and fitness collapse, then safely restores from the latest checkpoint with capped retries (up to 3).
- DDP-aware broadcasting keeps distributed training in sync during recovery.
- Validates checkpoint weights to avoid reloading corrupted EMA states.
- Centralizes checkpoint loading via `_load_checkpoint_state()` and uses it in `resume_training()` for consistency and less drift.
- Resets the scheduler state after recovery to maintain the intended LR schedule.
- Includes a new test `test_nan_recovery` that injects a NaN to verify the recovery path. ✅  
Details are in the [NaN epoch recovery PR (#22352)](https://github.com/ultralytics/ultralytics/pull/22352) by [Glenn Jocher](https://github.com/glenn-jocher).

### Dataset YAMLs: unified asset URLs

- Replaced hardcoded links with a centralized `ASSETS_URL` across VOC, COCO, COCO-Pose, VisDrone, and LVIS for consistent, maintainable downloads.  
See the [Use ASSETS\_URL in dataset YAMLs PR (#22361)](https://github.com/ultralytics/ultralytics/pull/22361) by [Glenn Jocher](https://github.com/glenn-jocher).

### Objects365 setup speedups

- Parallelized downloads, file moves, and label generation using `ThreadPoolExecutor`.
- Increased threads and refactored annotation handling to boost throughput.  
See the [Improve Objects365.yaml PR (#22362)](https://github.com/ultralytics/ultralytics/pull/22362) by [Glenn Jocher](https://github.com/glenn-jocher).

### CI and tests

- Temporarily disabled Jetson JetPack 5 Docker build while the new recovery path stabilizes.
- Skips training tests on Jetson/Raspberry Pi to reflect that edge devices are not training targets.

* * *

## Improvements

- More resilient training at scale with automatic recovery from NaN/Inf losses or sudden metric collapse.
- Safer, more consistent resume logic for optimizer, scaler, EMA, and best-fitness states.
- Faster dataset preparation, especially for Objects365, thanks to multithreading.
- More reliable dataset downloads via centralized hosting and unified URLs.

## Bug Fixes

- Prevents reloading corrupted EMA states by validating checkpoint weights during recovery.
- Avoids scheduler desynchronization by resetting LR scheduler after recovery.

* * *

## Quick start

- Upgrade:

```bash
pip install -U ultralytics

```

- Train (recovery is automatic; no extra flags needed):

```bash
yolo detect train model=yolo11n.pt data=coco128.yaml epochs=100 imgsz=640

```

- Resume anytime:

```bash
yolo detect train resume=True

```

Learn more about training options in the [Train mode documentation](https://docs.ultralytics.com/modes/train/).

* * *

## Model guidance

- YOLO11 is our latest stable and recommended default for all tasks; explore the [YOLO11 model docs](https://docs.ultralytics.com/models/yolo11/).
- Community models YOLO12 and YOLO13 are not recommended; see notes in the [YOLO12 model docs](https://docs.ultralytics.com/models/yolo12/).
- Ultralytics R&D for YOLO26 is underway; follow the [YOLO26 R&D preview](https://docs.ultralytics.com/models/yolo26/).

* * *

## What’s changed (PRs and authors)

- Unified dataset URLs with `ASSETS_URL` in YAMLs, see the [dataset URL unification PR (#22361)](https://github.com/ultralytics/ultralytics/pull/22361) by [Glenn Jocher](https://github.com/glenn-jocher).
- Objects365 preparation speedups, see the [Objects365 improvements PR (#22362)](https://github.com/ultralytics/ultralytics/pull/22362) by [Glenn Jocher](https://github.com/glenn-jocher).
- Automatic NaN epoch recovery and resume logic cleanup, see the [NaN epoch recovery PR (#22352)](https://github.com/ultralytics/ultralytics/pull/22352) by [Glenn Jocher](https://github.com/glenn-jocher).

Review the [complete changelog for v8.3.212 → v8.3.213](https://github.com/ultralytics/ultralytics/compare/v8.3.212...v8.3.213).

* * *

## Try it and share feedback

Please upgrade, put the new recovery path through its paces, and let us know how it performs in your workflows. You can start a conversation in [Ultralytics GitHub Discussions](https://github.com/orgs/ultralytics/discussions) or open issues with reproducible examples. Your feedback helps the YOLO community and Ultralytics team keep improving.