It's amusing to watch Deep Research through its reasoning, struggling with medium.com sites and finally deciding it should just ignore them. Emergent intelligence
2
1
24
In the Shazeer 2019 paper "Fast Transformer Decoding: One Write-Head is All You Need", section 2.4 there's the MultiheadSelfAttentionIncremental function The calculation of new_K (and new_V) mentions M, which isn't defined. This is a typo, s/M/x/, correct?
11
(h/t to @SingularMattrix for his excellent educational resources on the topic of autograd - these were immensely helpful here. This HN comment by Matthew is a good list: news.ycombinator.com/item?id…
8
As a follow-up on my blog post about reverse mode automatic differentiation from earlier this week (eli.thegreenplace.net/2025/r…), here's a new project: github.com/eliben/radgrad - an educational implementation of Autograd-like reverse mode AD via tracing
2
2
24
Part 5 of my Raft blog series is out. This one is about fixing linearizability issues that arise from client retries and achieving exactly-once delivery eli.thegreenplace.net/2024/i… This concludes the series! Thanks for reading
1
26
162
A post I wanted to write for a while now: How to do ML inference in Go through a Python sidecar server eli.thegreenplace.net/2024/m…
1
19
After a brief hiatus of 4.5 years, I've finally published Part 4 of my "Implementing the Raft consensus algorithm in Go" series. In this part, we build a simple but strict serializable key/value database on top of Raft. eli.thegreenplace.net/2024/i…
2
17
106
I added a new sample to Notes on running Go in the browser with WebAssembly (eli.thegreenplace.net/2024/n…) Talking on a Web Socket with Go -- Go server and client talk via web sockets, the client side in WASM using browser APIs (new WebSocket(...)) Code: github.com/eliben/code-for-b…
1
12
Building LLM-powered applications in Go go.dev/blog/llmpowered The level of abstraction in ML applications has shifted up, making them similar to cloud applications. Go excels at this.
22
134
If you're curious how technologies like Ollama, llama.cpp and GGUF make it possible to run powerful LLMs locally on your machine, check out eli.thegreenplace.net/2024/t…
1
4
30
I've now reimplemented the same RAG tool using Google's Gemini model (full code: github.com/eliben/code-for-b…) The model is on par with OpenAIs, but ai.google.dev has a generous free quota for its API so the sample can be run without paying $$$ to OpenAI!
10
Google's new Gemini models now have an API for developers with a generous free tier (60 queries per minute). It's easy to use from Go with the new Generative AI Go SDK, and easy to switch to the Vertex (GCP project - based) version as well. Here's how: eli.thegreenplace.net/2023/u…
2
17
Many common LLM-based applications like RAG (Retrieval Augmented Generation) are just data pipeline problems, and Go is really great at those. Implementing RAG is Go is very easy - see eli.thegreenplace.net/2023/r… and the accompanying code at github.com/eliben/code-for-b… for an example
3
4
56
The Go team has published official guidelines for organizing / structuring a Go project: go.dev/doc/modules/layout This includes recommended project layouts for different kinds of projects: a simple package, a command-line tool, a mix of the two, etc.
This sample runs successfully with wasmtime, wazero and Node Shout out wazero for their amazingly useful -hostlogging flag
1
1
7
I was nerd-sniped to create a sample of reading a file using WASI from hand-coded WebAssembly Text. Wrangling syscalls from an assembly-that-was-really-designed-to-be-a-compilation-target with an extra helping of WASI capability security; it was fun 😅 github.com/eliben/wasm-wat-s…
1
1
15
Cute weekend toy: eliben.github.io/lsystem/ Simulation for L-systems in JS: generating arbitrary recursive fractals from a formal context-free grammar
11