Hacker News: xfalcox

New comment by xfalcox in "A Jev that can ask Opus for help gives Opus accuracy at 4x lower cost"

xfalcox — Fri, 06 Nov 2026 21:08:19 +0000

Been doing this on Discourse for Spam and locale detection, mostly doing d3-turbo with Qwen 3.8 Flash Next Help.

New comment by xfalcox in "Show HN: Corgi, a helpful {agent, terminal} multiplexer"

xfalcox — Sat, 24 Oct 2026 04:07:40 +0000

I've been using herdr and this looks like a v2 of it. Will give it a try.

New comment by xfalcox in "Decision models like Jev don't beat LLM-as-a-judge or traditional classifiers"

xfalcox — Mon, 05 Oct 2026 10:23:26 +0000

Doesn't the article covers the speed part by showing that Qwen 3.6 35A3B has lower latency and same accuracy?

New comment by xfalcox in "vLLM v0.28.0"

xfalcox — Sat, 29 Aug 2026 20:46:05 +0000

Yeah, DeepSeek 4 Flash on vLLM has been an adventure indeed. It finally stabilized for me on 2 x H200 using a commit a few days before 0.28, so this release should be good for you.

New comment by xfalcox in "Qwen3.8 27B scores 52 on Artificial Analysis"

xfalcox — Mon, 17 Aug 2026 22:28:25 +0000

Have you tried running it on a single 5090? Dual 5090 require https://github.com/aikitoria/open-gpu-kernel-modules for higher perf. Are you using TP?

New comment by xfalcox in "Designing emoji for the way we communicate today"

xfalcox — Fri, 17 Jul 2026 17:47:18 +0000

I'm wondering the same! How that article has no links is beyond me.

New comment by xfalcox in "Show HN: Getting GLM 5.2 running on my slow computer"

xfalcox — Thu, 09 Jul 2026 21:08:52 +0000

Question to the OP, have you tested this on a machine where the entire model and context fit in RAM ?

New comment by xfalcox in "Show HN: Getting GLM 5.2 running on my slow computer"

xfalcox — Thu, 09 Jul 2026 21:08:00 +0000

README covers that

https://github.com/JustVugg/colibri#ssd-wear-warning

New comment by xfalcox in "Use your Nvidia GPU's VRAM as swap space on Linux"

xfalcox — Tue, 02 Jun 2026 23:50:47 +0000

Given my dev machine has 32GB of RAM and 32GB of VRAM that sits mostly idle when I'm not running AI models, this is not that bad of an idea.

New comment by xfalcox in "Google releases Gemma 4 open models"

xfalcox — Thu, 02 Apr 2026 16:53:22 +0000

Comparing a model you can downloads weights for with an API-only model doesn't make much sense.

New comment by xfalcox in "Claude wrote a full FreeBSD remote kernel RCE with root shell"

xfalcox — Wed, 01 Apr 2026 20:27:42 +0000

Our CEO did that at our company and found 33 CVEs. Rails also did that and found 7 or 8.

New comment by xfalcox in "Show HN: RatatuiRuby wraps Rust Ratatui as a RubyGem – TUIs with the joy of Ruby"

xfalcox — Thu, 22 Jan 2026 03:27:58 +0000

I just made a new installer for Discourse on CharmRuby, now I gotta check this out and see if porting is feasible. Hopefully this reduces the app size, that is quite large with CharmRuby

New comment by xfalcox in "Tell HN: The Google Tenor GIF API has been shut down"

xfalcox — Wed, 14 Jan 2026 17:49:20 +0000

That is a great fit for the GIF integration in Discourse.

I was able to quickly add support for it at https://github.com/discourse/discourse-gifs/pull/107

Love to see WEBP support. Do you plan on adding support for AVIF?

Also, this is used by many Discourse sites, we should talk.

New comment by xfalcox in "Sergey Brin's Unretirement"

xfalcox — Wed, 07 Jan 2026 17:49:59 +0000

First time I was in San Francisco and someone introduced themselves like that, going even beyond, was indeed a super weird experience being a brazilian.

New comment by xfalcox in "Z-Image: Powerful and highly efficient image generation model with 6B parameters"

xfalcox — Sat, 06 Dec 2025 18:20:18 +0000

We have vLLM for running text LLMs in production. What is the equivalent for this model?

New comment by xfalcox in "28M Hacker News comments as vector embedding search dataset"

xfalcox — Fri, 28 Nov 2025 19:28:14 +0000

I am partial to https://huggingface.co/Qwen/Qwen3-Embedding-0.6B nowadays.

Open weights, multilingual, 32k context.

New comment by xfalcox in "Honda: 2 years of ml vs 1 month of prompting - heres what we learned"

xfalcox — Fri, 14 Nov 2025 13:27:48 +0000

It's the Amazon own model. I'm baffled someone would pick it, even more that someone would test Llama 4 for a task in an age where Sonnet 4.5 is already out, so in the last 45 days.

Looks like they were limited by AWS Bedrock options.

New comment by xfalcox in "The Case Against PGVector"

xfalcox — Tue, 04 Nov 2025 03:55:57 +0000

> what does the rag for uploaded files do in discourse?

You can upload files that will act as RAG files for an AI bot. The bot can also have access to forum content, plus the ability to run tools in our sandboxed JS environment, making it possible for Discourse to host AI bots.

> also, when i run a discourse search does it really do both a regular keyword search and a vector search? how do you combine results?

Yes, it does both. In the full page search it does keyword first, then vector asynchronously, which can be toggled by the user in the UI. It's auto toggled when keyword has zero results now. Results are combined using reciprocal rank fusion.

In the quick header search we simply append vector search to keyword search results when keyword returns less than 4 results.

> does all discourse instances have those features? for example, internals.rust-lang.org, do they use pgvector?

Yes, all use PGvector. In our hosting all instances default to having the vector features enabled, we run embeddings using https://github.com/huggingface/text-embeddings-inference

New comment by xfalcox in "The Case Against PGVector"

xfalcox — Mon, 03 Nov 2025 22:47:44 +0000

We host thousands of forums but each one has its own database, which means we get a sort of free sharding of the data where each instance has less than a million topics on average.

I can totally see that at a trillion scale for a single shard you want a specialized dedicated service, but that is also true for most things in tech when you get to the extreme scale .

New comment by xfalcox in "The Case Against PGVector"

xfalcox — Mon, 03 Nov 2025 22:30:43 +0000

I was taken back when I saw what was basically zero recall loss in the real world task of finding related topics, by doing the same thing you described where we over capture with binary embeddings, and only use the full (or half) precision on the subset.

Making the storage cost of the index 32 times smaller is the difference of being able to offer this at scale without worrying too much about the overhead.