The post discusses the training costs and capabilities of the Olmo3 models from the Allen Institute for AI, highlighting their open weights, training scripts, and datasets. It compares the energy consumption for training the 7B an...
About:
Alex Bradbury is an open source consultant and technical blogger with interests in open source projects and drawing.
Website:
Specializations:
Interests:
Outgoing Links:
Subscribe to RSS:
Energy consumption of querying large language models is analyzed, emphasizing the need for transparency and accurate benchmarking to understand environmental impacts.
The post discusses the analysis of code performance through instruction counting, particularly using a QEMU plugin. It outlines how to capture necessary data for performance analysis, the development of a patch for QEMU's plugin A...
The post analyzes the inference and training costs associated with the DeepSeek AI models, specifically focusing on the GPU hours required for training and inference. It highlights the opacity of big AI labs regarding their energy...
This post compares the performance of Hetzner's AX102 dedicated server and the CCX53 VPS, focusing on their specifications, costs, and benchmark results. The AX102 features a Ryzen 9 7950X3D CPU, 128GB RAM, and RAID 0 storage, whi...
An analysis of per-query energy consumption for various LLMs reveals insights into their efficiency and highlights the need for further testing with additional models.
5.5shandbox
2026-02-11 • security linux code sandbox namespaces sandbox
shandbox is a user-friendly Linux sandboxing script that isolates personal files while allowing easy command execution and management of mounts.
The text discusses the importance of setting parameters when accessing models via a hosted API, and the challenges of finding recommended settings for different models. It also covers the default behavior of vLLM and the recommend...
0suite-helper
2025-05-12 • python llvm bug hunting testing
The suite-helper is a Python script that helps with repetitive tasks when setting up, building, and analyzing LLVM test suite builds. It provides subtools like create, add-config, status, run, match-tool, get-ll, and reduce-ll. Th...
The Chuwi MiniBook X N150 is a lightweight, low-cost, ultra-portable system that is a positive upgrade from previous models. It has a newer, faster processor, but also some drawbacks such as reduced battery life, fan noise, and so...
The text discusses the problem of cache misses when building LLVM in different directories and provides a solution using ccache. It explains the steps to enable ccache and the testing of the solution, along with its limitations an...
The text explains how to directly commit files to a separate git branch in a current git repository without perturbing the current branch. The author provides a motivating use case and breaks down the different object types needed...
The text discusses the problem of producing a Debian-derived sysroot for cross-compilation or a full image for booting with QEMU or on real hardware. It presents a solution using debootstrap and fakeroot, and explains the process ...
The post documents the results of experiments in various ways to access the full tree - the source and the build artifacts from under qemu. It includes the requirements, results, notes on test setup, and details of different metho...
The text is a runbook for setting up Arch Linux on a remote server. It includes detailed instructions for setting up drives, creating filesystems, performing an initial bootstrap, and chrooting into the new rootfs. The author also...
0pwr
2024-07-01 • workflow rss specification rss feed reader nuclear reactor
Pwr (paced web reader) is a script and terminal-centric workflow for keeping up to date with various sources online. It provides a structured workflow for keeping up to date with a preferred set of sites in an incremental fashion....
The text is a detailed account of a panel session on the Carbon programming language at EuroLLVM 2024. It includes discussions on Carbon's design, governance, comparison with other languages like Rust, and implementation of the Ca...
The text discusses the need to clarify instruction semantics when working with disassembly for different architectures. It introduces a simple script to convert an encoded instruction to a target-independent description of its sem...
The author reflects on ten years of writing LLVM Weekly, a newsletter summarizing developments on projects under the LLVM umbrella. They discuss the motivation, readership, content, tools used, and future plans for the newsletter....
The text discusses the use of terminal bells as a visual indicator for long-running tasks, and how to configure them for different terminal emulators. It also highlights the limitations and support for terminal emulator visual bel...