ultralytics 8.3.192

v8.3.192 - `ultralytics 8.3.192` Distributed Hyperparameter Tuning with MongoDB Atlas Integration (#21882)

on Python PyPI

🌟 Summary

Distributed hyperparameter tuning lands in ultralytics 8.3.192 with optional MongoDB Atlas coordination, delivering scalable HPO across machines, a steadier progress ETA, faster GPU data transfers, and more reliable CSV parsing — plus a clear docs warning on YOLOv5 compatibility. 🚀

📊 Key Changes

• Distributed Hyperparameter Tuning (primary)
  • Optional MongoDB Atlas integration to coordinate tuning across multiple machines.
  • New args: mongodb_uri, mongodb_db (default: ultralytics), mongodb_collection (default: tuner_results).
  • Robust connection handling: retries, pooling, and an automatic fitness index.
  • Shared results: reads best runs from MongoDB, writes new runs back, and syncs to CSV for plotting/resume.
  • Early stopping when the shared collection reaches the target iteration count.
  • Improved mutation logic, safer bounds, stronger resume behavior, and refreshed examples/docs.
• Progress ETA (tqdm)
  • More stable remaining-time estimates, reducing jitter in progress bars. ⏱️
• Performance: non-blocking GPU transfers
  • Applied .to(device, non_blocking=True) across train/val for Classification, Detection, Pose, Segmentation, YOLO-World (text embeddings), and YOLOE (text/visual prompts) to reduce data-transfer bottlenecks. ⚡
• CSV parsing reliability
  • Polars now infers schema from entire files (infer_schema_length=None) across dataset conversion, training results, and plotting, preventing type mix-ups and improving plot/metric accuracy. 📈
• Docs update
  • Clear warning: models trained in ultralytics/yolov5 are not compatible with the ultralytics/ultralytics library (YOLOv5u is anchor-free). ⚠️

🎯 Purpose & Impact

• Scale your HPO effortlessly
  • Distribute tuning jobs across machines using MongoDB Atlas — speed up experimentation and find better hyperparameters sooner. 🌐
• Smoother training experience
  • More accurate ETA and faster input pipelines translate to improved GPU utilization and potentially shorter epoch times.
• More trustworthy metrics and plots
  • Full-file CSV schema inference reduces parsing errors for local runs and Ultralytics HUB exports.
• Reduced confusion for YOLOv5 users
  • Clear compatibility guidance helps teams avoid loading errors and plan migrations to YOLOv5u within ultralytics/ultralytics.

Quick start for distributed tuning:

from ultralytics import YOLO

model = YOLO("yolo11n.pt")
model.tune(
    data="coco8.yaml",
    epochs=10,
    iterations=300,
    mongodb_uri="mongodb+srv://user:pass@cluster.mongodb.net/",
    mongodb_db="ultralytics",
    mongodb_collection="tuner_results",
    plots=False,
    save=False,
    val=False,
)

Helpful links:

• Read the Hyperparameter Tuning guide: https://docs.ultralytics.com/guides/hyperparameter-tuning
• See ultralytics/yolov5 vs ultralytics/ultralytics note: https://github.com/ultralytics/ultralytics/tree/main/docs/en/models/yolov5.md

What's Changed

• fix: 🐞 improve CSV reading performance by disabling schema inference by @onuralpszr in #21909
• Add non_blocking=True for additional train/val loaders by @glenn-jocher in #21912
• Add warning that legacy YOLOv5 models are not compatible with Ultralytics by @Y-T-G in #21915
• ultralytics 8.3.192 Distributed Hyperparameter Tuning with MongoDB Atlas Integration by @glenn-jocher in #21882

Full Changelog: v8.3.191...v8.3.192