ultralytics v8.4.210 — v8.4.210 - ultralytics 8.4.210 Use PyTorch device names for OpenVINO and LiteRT Intel CPU, GPU and NPU (#27136)
Source
github.com
Author
github-actions[bot]
Date
Why it matters
Device strings for OpenVINO and LiteRT now match PyTorch naming, and the old intel: aliases warn. INT8 export keeps detection heads in FP32 to preserve box and score precision, which matters for edge deployments.
Key takeaways · AI-distilled
Ultralytics 8.4.210 lets OpenVINO and LiteRT accept PyTorch-style device names (cpu, xpu, npu) for Intel hardware, keeping intel:cpu/gpu/npu as deprecated aliases. Mapping still depends on the runtime: PyTorch npu means Ascend.
The OCR solution moves to PP-OCRv6 on ONNX Runtime with YOLO26x-Text as the default detector, reading rotated or upside-down pages in reading order without PaddlePaddle. The optimized recognizer needs Python 3.11+ and ONNX Runtime 1.26+.
LiteRT and NXP INT8 exports now keep detection, segmentation, pose and OBB head outputs in FP32 to preserve confidence scores and box precision, and RT-DETR's transformerThe neural network architecture behind modern AI models, built on attention — letting every word directly consider every other word in parallel.Full definition → parts stay out of INT8 to avoid severe accuracy drops on TensorRT 11.3 and 11.4.
Terms in this piece · Glossary
transformer — The neural network architecture behind modern AI models, built on attention — letting every word directly consider every other word in parallel.
