Friday, 11th October 2024
Providing validation, strength, and stability to people who feel gaslit and dismissed and forgotten can help them feel stronger and surer in their decisions. These pieces made me understand that journalism can be a caretaking profession, even if it is never really thought about in those terms. It is often framed in terms of antagonism. Speaking truth to power turns into being hard-nosed and removed from our subject matter, which so easily turns into be an asshole and do whatever you like.
This is a viewpoint that I reject. My pillars are empathy, curiosity, and kindness. And much else flows from that. For people who feel lost and alone, we get to say through our work, you are not. For people who feel like society has abandoned them and their lives do not matter, we get to say, actually, they fucking do. We are one of the only professions that can do that through our work and that can do that at scale.
HTML for People (via) Blake Watson's brand new HTML tutorial, presented as a free online book (CC BY-NC-SA 4.0, on GitHub). This seems very modern and well thought-out to me. It focuses exclusively on HTML, skipping JavaScript entirely and teaching with Simple.css to avoid needing to dig into CSS while still producing sites that are pleasing to look at. It even touches on Web Components (described as Custom HTML tags) towards the end.
The primary use of “misinformation” is not to change the beliefs of other people at all. Instead, the vast majority of misinformation is offered as a service for people to maintain their beliefs in face of overwhelming evidence to the contrary.
— Mike Caulfield, via Charlie Warzel
$2 H100s: How the GPU Bubble Burst. Fascinating analysis from Eugene Cheah, founder of LLM hosting provider Featherless, discussing GPU economics over the past 12 months.
TLDR: Don’t buy H100s. The market has flipped from shortage ($8/hr) to oversupplied ($2/hr), because of reserved compute resales, open model finetuning, and decline in new foundation model co’s. Rent instead.
lm.rs: run inference on Language Models locally on the CPU with Rust (via) Impressive new LLM inference implementation in Rust by Samuel Vitorino. I tried it just now on an M2 Mac with 64GB of RAM and got very snappy performance for this Q8 Llama 3.2 1B, with Activity Monitor reporting 980% CPU usage over 13 threads.
Here's how I compiled the library and ran the model:
cd /tmp
git clone https://github.com/samuel-vitorino/lm.rs
cd lm.rs
RUSTFLAGS="-C target-cpu=native" cargo build --release --bin chat
curl -LO 'https://huggingface.co/samuel-vitorino/Llama-3.2-1B-Instruct-Q8_0-LMRS/resolve/main/tokenizer.bin?download=true'
curl -LO 'https://huggingface.co/samuel-vitorino/Llama-3.2-1B-Instruct-Q8_0-LMRS/resolve/main/llama3.2-1b-it-q80.lmrs?download=true'
./target/release/chat --model llama3.2-1b-it-q80.lmrs --show-metrics
That --show-metrics option added this at the end of a response:
Speed: 26.41 tok/s
It looks like the performance is helped by two key dependencies: wide, which provides data types optimized for SIMD operations and rayon for running parallel iterators across multiple cores (used for matrix multiplication).
(I used LLM and files-to-prompt to help figure this out.)