Vítor de Oliveira

@Vitorinox

Commit activity

7 commits in the last 30 days

longest streak 2 dayscurrent 2

107 / 150 XP43 XP to operator

Today

PHP shipped its CRITICAL-rated security release today for all supported branches: 8.6.0, 8.5.13, 8.4.28, 8.3.37, 8.2.36. Fixes include CVE-2026-115560, an unlikely RCE. Can't update yet? Set session.upload_progress.enabled=0 and restart workers. https://www.php.net/

Yesterday

LatamGPT opened two Hugging Face datasets: 178.4M documents in Spanish, Portuguese and English. Firms above USD 1B in revenue or capital need CENIA's written OK, and CENIA says the compilation is not OSI open source. https://huggingface.co/datasets/latam-gpt/LatamGPT-Corpus-1.0
huggingface.colatam-gpt/LatamGPT-Corpus-1.0 · Datasets at Hugging FaceDataset with 143M rows (Image, Text). Explore, preview and download it on Hugging Face.

Mon, Nov 16

Qwen released Qwen4-Max, a 2.4T MoE with 71B active parameters, native vision and a 1M context. Weights are not open: API-only on QwenCloud and Novita. Qwen says smaller Qwen4 models will be open; I'm waiting for a ~27B to run locally. https://qwen.ai/blog?id=qwen4-max
qwen.aiQwenQwen offers comprehensive functionality spanning chatbot, image and video understanding, image generation, document processing, web search integration, tool utilization, and artifacts.

Fri, Nov 13

Aaron, thank you for writing to me. In June I reported that in Solo, on macOS, an accent with a dead key came out twice: ó became óo. You fixed it in 0.12.0 and emailed me about it. Nobody had ever written to me after fixing a bug I reported. It means a lot. Latest is 0.12.4. ❤️

Thu, Nov 12

Everyone is talking about Nvidia. I keep thinking about one thing: 250 GB of VRAM. Llama 3.1 405B in GGUF at 4-bit would fit (about 200 GB for the weights alone), with room left for context. Meanwhile I am still bargaining with llama.cpp over how many layers fit on my GPU.

That's @Vitorinox's root commit.