# New Release: Ultralytics v8.4.157 **URL:** https://community.ultralytics.com/t/new-release-ultralytics-v8-4-157/2228 **Category:** Discussion **Tags:** announcements, ultralytics-official, releases **Created:** [September 21, 2026, 1:09pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-157/2228 "2026-09-21T13:09:58Z") **Posts on this page:** 1 **Page:** 1
### 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: [September 21, 2026, 1:09pm UTC](https://community.ultralytics.com/t/new-release-ultralytics-v8-4-157/2228/1 "2026-09-21T13:09:58Z")
# πŸš€ Ultralytics v8.4.157: Faster TensorRT, Expanded YOLOE-26 Support, and Better Apple Silicon Performance Ultralytics v8.4.157 delivers **up to 20% faster TensorRT FP16 and INT8 inference** , expanded prompt-free YOLOE-26 support, major Apple Silicon optimizations, and more reliable training, validation, export, and dataset workflows. 🌟 Explore the [Ultralytics v8.4.157 release](https://github.com/ultralytics/ultralytics/releases/tag/v8.4.157) for the complete release details. ## 🌟 New Features ### ⚑ TensorRT inference up to 20% faster In [PR #26223](https://github.com/ultralytics/ultralytics/pull/26223), [@Y-T-G](https://github.com/Y-T-G) introduced several TensorRT optimizations: - FP16 conversion is calibrated with a real image instead of random noise. - SiLU activations are rewritten into a TensorRT-fusable form. - CUDA Graph replay reduces launch overhead for supported static engines. - FP16 and INT8 inference can be **up to 20% faster** , particularly with static-shape engines. CUDA Graph acceleration is excluded for dynamic engines, DLA execution, and engines with embedded NMS. ### 🎯 Expanded prompt-free YOLOE-26 inference In [PR #26201](https://github.com/ultralytics/ultralytics/pull/26201), [@ShuaiLYU](https://github.com/ShuaiLYU) added support for checkpoints containing both one-to-many and one-to-one LRPC detection heads. This update enables: - Selection between standard NMS and NMS-free inference through the `nms` setting. - Regeneration of both branches with `set_vocab()`. - More flexible custom prompt-free YOLOE-26 workflows. ## 🍎 Performance Improvements Apple Silicon users receive several CPU and MPS optimizations: - [PR #26221](https://github.com/ultralytics/ultralytics/pull/26221) by [@raimbekovm](https://github.com/raimbekovm) disables a slow NNPACK convolution path during CPU inference and training. - [PR #26228](https://github.com/ultralytics/ultralytics/pull/26228) by [@raimbekovm](https://github.com/raimbekovm) replaces costly box-IoU reductions for faster MPS validation. - [PR #26234](https://github.com/ultralytics/ultralytics/pull/26234) by [@raimbekovm](https://github.com/raimbekovm) accelerates `TaskAlignedAssigner` on CPU and MPS. - [PR #26239](https://github.com/ultralytics/ultralytics/pull/26239) by [@Y-T-G](https://github.com/Y-T-G) replaces problematic semantic segmentation operations with MPS-friendly alternatives, addressing slowdowns and buffer-size failures. - [PR #26233](https://github.com/ultralytics/ultralytics/pull/26233) by [@raimbekovm](https://github.com/raimbekovm) avoids repeated box and keypoint conversions in `Results.save_txt()`, plotting, and summaries. Together, these changes improve larger-batch training, validation, segmentation, result export, and visualization performance. ## πŸ§ͺ Training and Validation Reliability - [PR #26224](https://github.com/ultralytics/ultralytics/pull/26224) by [@fcakyon](https://github.com/fcakyon) prevents early AutoBatch probe failures from being incorrectly treated as memory limits. - [PR #26226](https://github.com/ultralytics/ultralytics/pull/26226) by [@cainiao33](https://github.com/cainiao33) ensures mosaic augmentation closes correctly during short training runs. - [PR #26225](https://github.com/ultralytics/ultralytics/pull/26225) by [@cainiao33](https://github.com/cainiao33) preserves the real checkpoint or configuration error when resume checks fail. - [PR #26236](https://github.com/ultralytics/ultralytics/pull/26236) by [@wizzseen](https://github.com/wizzseen) rejects invalid negative `patience` values during configuration validation. - [PR #26243](https://github.com/ultralytics/ultralytics/pull/26243) by [@cainiao33](https://github.com/cainiao33) floors `imgsz` at the model stride during image-size validation. These updates provide earlier validation and clearer failures for invalid or unsupported configurations. ## πŸ“ Export, Dataset, and Evaluation Fixes - [PR #26241](https://github.com/ultralytics/ultralytics/pull/26241) by [@cainiao33](https://github.com/cainiao33) keeps the exported image size for fixed-shape models across repeated `predict()` calls. - [PR #26244](https://github.com/ultralytics/ultralytics/pull/26244) by [@cainiao33](https://github.com/cainiao33) makes TAR extraction return the dataset’s actual top-level directory, matching ZIP behavior. - [PR #26227](https://github.com/ultralytics/ultralytics/pull/26227) by [@cainiao33](https://github.com/cainiao33) refreshes stale semantic mask caches when `nc` changes to `1`, preserving foreground pixels. - [PR #26240](https://github.com/ultralytics/ultralytics/pull/26240) by [@Nicholas022400701](https://github.com/Nicholas022400701) prevents standalone detection validation from crashing on labels with classes unsupported by the model. - [PR #26246](https://github.com/ultralytics/ultralytics/pull/26246) by [@Nicholas022400701](https://github.com/Nicholas022400701) keeps COCO size-specific metrics correctly associated with detection, segmentation, and pose tasks. ## πŸ“¦ Dependency Improvements In [PR #26219](https://github.com/ultralytics/ultralytics/pull/26219), [@hylreg](https://github.com/hylreg) corrected Python-version markers and export-extra conflicts, improving `uv lock` and `uv sync` dependency resolution. ## πŸ™Œ New Contributors A warm welcome to two first-time contributors: - [@wizzseen](https://github.com/wizzseen), contributing configuration validation through [PR #26236](https://github.com/ultralytics/ultralytics/pull/26236). - [@hylreg](https://github.com/hylreg), improving dependency resolution through [PR #26219](https://github.com/ultralytics/ultralytics/pull/26219). Thank you to every contributor and the wider YOLO community for helping make this release possible! πŸ’™ ## πŸš€ Try v8.4.157 Upgrade to the latest release: ```bash pip install -U ultralytics ``` TensorRT, YOLOE-26, and Apple Silicon users are especially encouraged to test this release and share feedback. You can review every change in the [v8.4.156 to v8.4.157 full changelog](https://github.com/ultralytics/ultralytics/compare/v8.4.156...v8.4.157).