The open‑source LLaMA inference library adds a CUDA debug buffer pool and updates binary builds for macOS, iOS and Linux.
NoteBy OossaPublished by Oossa: 1 min read
ggml‑org released version b12318 of llama.cpp on 2026‑11‑12. The update adds a CUDA debug buffer pool to help developers track memory use with compute‑sanitizer. It also includes new pre‑built binaries for Apple Silicon, Intel macOS, iOS, and several Ubuntu CPU and Vulkan targets.
Developers running LLaMA models on NVIDIA GPUs can now more easily debug memory issues, reducing crashes during inference.
Models
AWS adds three open‑weight DeepSeek language models to SageMaker, offering long‑context and low‑cost options for developers.
Models
Developers can now run a copy of many AWS services on their laptop or CI server, with no cloud account needed.
Models
The new Gemini Decisions API can handle up to 1,000 closed‑ended questions in a single pass and costs $0.10 per million input tokens.
Oossa · Newsletter
The week in AI, explained
Every Monday: the stories worth knowing, in plain language. Free, no spam.
