Show HN: Pip – Your Git history puts you on a Performance Improvement Plan
Article URL: https://github.com/littlestjames82-sys/pip-pip Comments URL: https://news.ycombinator.com/item?id=50556132 Points: 1 # Comments: 0
Article URL: https://github.com/littlestjames82-sys/pip-pip Comments URL: https://news.ycombinator.com/item?id=50556132 Points: 1 # Comments: 0
Article URL: https://github.com/gaidardzhiev/vertigo Comments URL: https://news.ycombinator.com/item?id=50558380 Points: 1 # Comments: 1
Article URL: https://spectralorbit.com/blog/satelital_game_engine Comments URL: https://news.ycombinator.com/item?id=50556174 Points: 1 # Comments: 0
Hi. I made small project called Zpup. It takes small 3D scene written in JSON and gives back a flat SVG from any angle. There is no JavaScript in the file, so it works in img-tag or README.I needed 3D for small project at my company, and Zdog make big impression on me. But Zdog has not had a release for several years, so I wanted to make modern version.One thing works different. Zdog sorts each shape by one point. so small thing on big floor gets covered when you turn it. Zpup compares the shapes that overlap. You can see both in the README.I wrote it with Claude Code Fable 5.2.No canvas renderer, no text shape, and it is 0.2 version.Playground: https://rxolve.github.io/zpupFeedback is always welcome :) Comments URL: https://news.ycombinator.com/item?id=50556534 Points: 3 # Comments: 0
ComfyUI v0.46.0 introduces new partner nodes from AlphaGen, MiniMax, Gemini-LLM, HeyGen, Kling, Bria, and others, updates workflow templates to v0.11.111, adds support for single channel PNG saving, video deflicker, PAN2, and Qwen3-1.7B text generation, fixes issues with workflows, VRAM buffer leaks, and CUDA version warnings, removes DirectML support, and enhances functionality for video processing and model integration.
Vercel 在 Vercel CLI 中引入了统一的计费命令结构,将与计费相关的功能归入 'vercel billing' 命令下。用户现在可以查看计费概览、管理发票并预览即将产生的费用。新功能包括访问地址和税号等计费详情、列出及下载发票,以及预估下一个计费周期的费用。此次更新要求使用 CLI 65.17.0 或更高版本,并支持 JSON 输出以用于自动化。现有命令已添加新前缀进行更新,但仍可作为别名使用。
arXiv:2611.23428v1 Announce Type: new Abstract: We show that the importance of a LoRA adapter's initial input-subspace direction is regime-dependent, and we characterize this dependence in theory and experiment. This refines the recent view that spectral initializers help through update magnitude alone. Direction matters most when labeled data are scarce relative to the representation dimension and the activation subspace used for initialization captures supervised task structure. In this regime, the adapter's row span remains sensitive to its starting point, making task-aligned initialization especially useful. A spiked-covariance perturbation analysis characterizes the finite-sample stability of the estimated subspace. A first-step analysis of LoRA training explains why the subspace's alignment with the task matters and motivates an alignment coefficient $\rho$ that closely tracks gains across classification tasks. We test this account with PivotLoRA, which initializes adapters from activation PCA at prediction-relevant positions and applies one output-preserving orthogonal pivot early in training. In few-shot BERT-base classification, PivotLoRA achieves the best average rank among 13 methods and improves over vanilla LoRA in 29 of 30 settings, even though it starts from a smaller input-factor initialization norm. Additional classification experiments extend the evaluation to 7B and 8B Llama decoders. Classification gains diminish with more labeled data, consistent with the proposed regime dependence. A targeted extension to completion-only reasoning also shows improvements on GSM8K and StrategyQA. These findings identify conditions under which initialization direction matters, with subspace-task alignment offering a low-cost diagnostic before training for classification.
A study analyzing 272 production supervised fine-tuning jobs on NVIDIA H200 GPUs found that traditional FLOP-based estimates explain only 31% of actual training duration (R² = 0.31, MAE 12.41 h). In contrast, a CatBoost regressor using pre-dispatch job configuration features—including evaluation cadence, checkpoint policy, and per-GPU load—achieved R² = 0.98 with a mean absolute error of 1.71 h and median error of 0.51 h in cross-validation. The model outperforms analytical baselines by capturing overheads beyond compute, such as I/O contention and non-linear interactions between configuration variables. Results highlight the importance of training-specific signals and suggest that scheduling accuracy improves significantly when predictors account for operational overheads.
Learn why coding agent bills spiral out of control — and how to trace, compare, and govern spend across Claude Code, Cursor, Copilot, and more in one place.
The author explores streaming MoE (Mixture of Experts) model experts directly from SSD to achieve near-Strata performance on a 32GB MacBook M5, using a custom implementation in Hugging Face Transformers. By optimizing expert caching, leveraging Metal kernels, and minimizing host-GPU synchronization delays, the approach reaches ~14 tok/s—within 6% of Strata's performance—demonstrating that efficient expert streaming is feasible on consumer hardware without C++ extensions.