Developer news without the noise — My Dev News in Telegram.
Tag
All blog posts with this tag.
How data becomes numbers, a model computes an answer, error points to the next adjustment, and held-out examples decide whether the network learned. From apartment prices to convolution, recurrence, and attention.
A Habr survey of hypernetworks, Kilobyte Models, and SeedLM — compact “recipes” instead of giant weight files, and what you pay for the savings.
How large training sets are laid out: JSON vs JSONL, shards, compression, tokens, batches, and epochs — and why a terabyte of data does not need a terabyte of RAM.
Computer Vision is not a PyTorch-vs-TensorFlow debate. Data, tensors, CNNs, classification, detection, segmentation, transfer learning, metrics, and production form one pipeline that survives an API change.
Why GPUs matter for neural networks, how CPU, TPU, and NPU differ, why VRAM and bandwidth matter, how multi-GPU setups work, quantization, and the stack from PyTorch down to silicon.
What an ML framework is, how tensors and autograd work, how PyTorch, TensorFlow, JAX, tinygrad, and micrograd differ, and what to pick for your task — from learning to production inference.
What ONNX and ONNX Runtime are, how to export from PyTorch, opsets and dynamic axes, why CRNN/YOLO exports break, and how to match Python before the browser.
How PyTorch and TensorFlow differ: tensors, autograd, the training loop, data pipelines, GPU/TPU, deployment, CV and LLM ecosystems — and how to pick a framework for your task in 2026.
Object detection in plain engineering language: how YOLO differs from classification and segmentation, what a box, confidence, and duplicate suppression actually mean, why pretrained weights and your own dataset matter, and how to connect a detector with OCR, lines, a graph, and a language model on process diagrams.
Field notes: how the pilot cuts a UT form photo into cells — morphology, projections, gated perspective — and which other sheet-splitting methods show up in industry practice.
Field notes: a local CRNN on thickness crops, CTC, 640k parameters, the empty-cell conflict, and why a Transformer is not the first step.
Third UT-pilot note: error diagnosis, one lever at a time, epochs, the dataset, and when the model should stop being squeezed for points.
AI for weld radiography: dataset, labels, detector, rules, and an expert. Why 88 GDXray images are not a plant dataset.
A pillar guide to supervised fine-tuning corpora: instruction→answer contracts, coverage, refusals, synthetic vs expert sources, pre-train gates, and eval leakage.
A practical path from single-digit classification to sequence recognition on real paper forms: CNN, CTC, CRNN, TrOCR, labels from your database, experiments, and a production pipeline without the wrong model.
A pillar guide to data for training and evaluating AI: golden records, golden editors, train/eval splits, leakage, manifests, and parameters by task and domain.
