maciejgryka

6 ms·

1.
by maciejgryka 1mo ago
Does anyone know what the license for this model is? Specifically any word on restrictions about what it can be used for?
2.
by maciejgryka 2mo ago
That’s like asking whether fuel consumption was productive or leisure as the number of cars on the road increased. It’s both! I don’t think you can separate one from the other in any reasonable way.
3.
by maciejgryka 2mo ago
Thank you fur writing this nichochar. I have many similar thoughts, but one thing that worries me is also trying too hard to do the optimal thing. I often find that my own expectations collide with reality (kids’ moods, random events necess
4.
by maciejgryka 2mo ago
Ha that’s a really interesting comparison! I guess the difference is that OR directly sends revenue to model providers, so maybe they’re more likely to continue working together? But I can totally see it go the other way once a provider fee
5.
by maciejgryka 2mo ago
I’d bet there are as many stories of businesses failing because of inability to ship quickly as there are about focusing too much on your tooling instead of delivering value to customers. Both failures are dangerous! And depending on your m
6.
Gemma 4 Models(huggingface.co)
5 points by maciejgryka 6mo ago|1 comments
7.
The 10x inference tax you don't have to pay(distillabs.ai)
1 points by maciejgryka 7mo ago|0 comments
8.
Benchmarking the best base small model for fine-tuning(distillabs.ai)
2 points by maciejgryka 7mo ago|0 comments
9.
Show HN: Small "AI slop" classifier running in a browser extension(github.com)
2 points by maciejgryka 8mo ago|0 comments
10.
Show HN: Distilled 0.6B text-to-SQL model(github.com)
5 points by maciejgryka 9mo ago|0 comments
11.
Which small model is best for fine-tuning? We tested 12 of them on 8 tasks(distillabs.ai)
8 points by maciejgryka 10mo ago|1 comments
12.
by maciejgryka 10mo ago
We benchmarked which small language models are most tunable and which deliver best performance after fine-tuning. Tested 12 models (Qwen, Llama, Gemma, Granite, SmolLM) on 8 tasks. TL;DR Qwen3 family is the best overall choice, small Llamas
13.
by maciejgryka 10mo ago
Huh works fine for me, even when not logged in to Github. Can you try again?
14.
by maciejgryka 10mo ago
We've been experimenting with small models for structured tool calling tasks and just released gitara. Both the 3B and 1B models turn natural language instructions into valid git commands and run fully locally through Ollama. Highlight
15.
Gitara: A small, local Git agent(github.com)
5 points by maciejgryka 10mo ago|4 comments
16.
by maciejgryka 11mo ago
I think it’s going to be a while before we see small models (defined roughly as “runnable on reasonable consumer hardware”) do a good job at general coding tasks. It’s a very broad area! You can do some specific tasks reasonably well (eg I
17.
by maciejgryka 11mo ago
I think this is a description of how things are today, but not an inherent property of how the models are built. Over the last year or so the trend seems to be moving from “more data” to “better data”. And I think in most narrow domains (wh
18.
Beads – A memory upgrade for your coding agent(github.com)
12 points by maciejgryka 1y ago|0 comments
19.
We Trained a 3B Function-Calling Git Agent for Local Use(distillabs.ai)
6 points by maciejgryka 1y ago|2 comments
20.
Dashbit Plans for 2025(dashbit.co)
3 points by maciejgryka 2y ago|0 comments
21.
by maciejgryka 2y ago
I’d encourage everyone, who finds this appealing to check out how Ecto works in Elixir. It’s all functional & immutable goodies and pipelines are built into the language and idiomatic. Definitely looks weird at first glance (and Ecto is
22.
by maciejgryka 2y ago
Highest-end fidget spinner I’ve ever seen. Instantly appealing to my inner 6-year-old.
23.
by maciejgryka 2y ago
Yes, compute is absolutely the limiting factor today. Not only because the space of hyperparameters is huge and having more compute would make it easier/possible to explore. But also, weirdly, because inference becomes increasingly imp
24.
by maciejgryka 2y ago
> “Retrieval-Augmented Generation” is nothing more than a fancy way of saying “including helpful information in your LLM prompt.”
25.
by maciejgryka 2y ago
There's hype and FOMO for sure and you're right that there's lots to learn from information retrieval work. But why be dismissive of the whole thing? People learning from past research and applying it in new contexts seems li
26.
by maciejgryka 2y ago
For sure, it's only worth doing if you actually have so much relevant data that it doesn't fit in the context! This is definitely the case for us for this problem, but it's not universal.
27.
by maciejgryka 2y ago
This is one of the things we learned recently about building production workflows with LLMs. Happy to answer any questions/feedback here <3
28.
Surprise, your data warehouse can RAG(rainforestqa.com)
34 points by maciejgryka 2y ago|3 comments
29.
by maciejgryka 2y ago
I have no actual info on this, but I always assumed they'd compute some mutlimodal embeddings of the screenshots to then retrieve semantically-relevant ones by text? And yeah, they'd have to do it using on-device models, which doe
30.
by maciejgryka 2y ago
I recorded myself trying to read through and understand the high-level of this if anyone's interested in following along: https://maciej.gryka.net/papers-in-public/#scaling-monoseman...

More ›