Top all-time · #9389
Lightly
Company blog · lightly.ai/postaiprogramming
Lightly is the company blog at lightly.ai, writing about ai and programming. Since 2020, 8 different Hacker News readers have linked it 32 times, most often “Nvidia B200 vs. H100: Independent Training Performance Tests”. No feed was found, so read it on the site itself.
Too few sampled comments (3) to say how often HN recommends it.?
Why HN links it
HN readers link Lightly's blog for its GPU and training benchmarks, such as the B200 vs H100 comparison, and its LightlyTrain launch post on training vision models with medical and agricultural data.
Summarised by Claude from a sample of 3 HN comments that link here. Each comment is marked recommending, critical or neutral (citing without judging); how this works.
Improve machine learning models by pretraining them on your data and curating vision data for fine-tuning.
Most-linked posts?
- Nvidia B200 vs. H100: Independent Training Performance Tests4 people · 4 mentions · first linked 2025-04-10
- Active Learning Strategies Compared for YOLOv8 on Lincolnbeet2 people · 2 mentions · first linked 2023-04-20
- /blog/introducing-lightly-train2 people · 2 mentions · first linked 2025-04-15
What Hacker News says?
We tried it on medical data and agricultural data as mentioned in our launch post: https://www.lightly.ai/blog/introducing-lightly-train But we will keep on exploring more domains and downstream tasks, you can be certain of that!
Comment linking /blog/introducing-lightly-train
on “LightlyTrain: Better Vision Models, Faster – No Labels Needed”Submitted Segment Anything Model and Friends as a story · 110 points
HN discussionHi HN, I’m Igor, co-founder of Lightly AI ( https://www.lightly.ai/ ). We just released LightlyTrain, a new open-source Python package (AGPL-3.0, free for research and educational purpose) for self-supervised pretraining of computer vision models: https://github.com/lightly-ai/lightly-train Standard vision models pretrained on generic datasets like ImageNet or COCO often underperform on specific domains (e.g., medic…
Comment linking /blog/introducing-lightly-train
on “LightlyTrain: Better Vision Models, Faster – No Labels Needed”Submitted Self-Supervised Learning for Videos as a story · 91 points
HN discussion- nl · 2026-01-15
> I am of the opinion that Nvidia's hit the wall with their current architecture Based on what? Their measured performance on things people care about keep going up, and their software stack keeps getting better and unlocking more performance on existing hardware Inference tests: https://inferencemax.semianalysis.com/ Training tests: https://www.lightly.ai/blog/nvidia-b200-vs-h100 https://newsletter.semianalysis.com…
Comment linking /blog/nvidia-b200-vs-h100
on “Furiosa: 3.5x efficiency over H100s” Submitted Nvidia B200 vs. H100: Independent Training Performance Tests as a story · 6 points
HN discussion> Realistically, serving 300 users per GPU you'll spend a lifetime cost of about $133 per user, plus the datacenter/upkeep bill. What is the operational cost and when does it become more expensive than the upfront capex? The B200 tops out at 1000W and idles around 140W. It averages around 600W. https://www.lightly.ai/blog/nvidia-b200-vs-h100 U.S. average electricity cost is $.14 per kWh in March. https://www.eia.gov…
Comment linking /blog/nvidia-b200-vs-h100
on “Inference cost at scale with napkin math”Submitted Self-Supervised Learning at ECCV 2024 as a story · 6 points
HN discussionSubmitted Image Retrieval using your fine-tuned DINO model as a story · 6 points
HN discussionSubmitted Train Test Split in Deep Learning as a story · 6 points
HN discussionSubmitted Active Learning Using Detectron2 as a story · 5 points
HN discussionSubmitted LightlyTrain X DINOv2: Smarter Self-Supervised Pretraining, Faster as a story · 4 points
HN discussionSubmitted Clip and Friends: How Vision-Language Models Evolved as a story · 4 points
HN discussionSubmitted Albumentations vs. PIL: 2x Speedup for Model Training Pipelines as a story · 4 points
HN discussionSubmitted Self-Supervised Learning for Autonomous Driving as a story · 4 points
HN discussionSubmitted Brief Introduction to Vision Language Models as a story · 4 points
HN discussionSubmitted How redundant is your dataset? as a story · 4 points
HN discussionSubmitted NVIDIA B200 low power usage for inference AI workload as a story · 3 points
HN discussionSubmitted Self-Supervised Models Are More Robust and Fair as a story · 3 points
HN discussionSubmitted The Advantage of Self-Supervised Learning as a story · 3 points
HN discussionSubmitted Using Self-Supervised Learning for Dense Prediction Tasks as a story · 2 points
HN discussionSubmitted Knowledge Distillation Trends as a story · 2 points
HN discussionSubmitted Active Learning Strategies Compared for YOLOv8 on Lincolnbeet as a story · 2 points
HN discussionSubmitted Embedded Covid mask detection on an Arm Cortex-M7 processor using PyTorch as a story · 2 points
HN discussionSubmitted Brief Introduction to Contrastive Learning as a story · 1 points
HN discussionSubmitted Data Curation Demystified for Stable Video Diffusion as a story · 1 points
HN discussionSubmitted Apple M1 and M2 Performance for Training SSL Models as a story · 1 points
HN discussionSubmitted Improve Your Large Language Models (LLMs) with Active Learning as a story · 1 points
HN discussionSubmitted Boosting YOLOv8 Accuracy and Saving 77% in Labeling Costs with Active Learning as a story · 1 points
HN discussionSubmitted Self-supervised learning trends and what to expect in 2023 as a story · 1 points
HN discussion