xfalcox

8 ms·

31.
by xfalcox 2y ago
I just got an invite to Google I/O, which I've always dreamed of attending, but as someone from South America I'm having nightmares about ICE. It's just a 4 days trip, but I have an entire family that needs me to come ba
32.
by xfalcox 2y ago
This is super awesome, congrats!
33.
by xfalcox 2y ago
I shipped this in Discourse a couple of weeks ago, and it's incredible tech. And fully supported in PostgreSQL via pgvector.
34.
by xfalcox 2y ago
Can I safely use Ractors inside Rails if my tasks have a mix of CPU and I/O ?
35.
by xfalcox 2y ago
I wish the author would do the same for image decoding. iOS inability to draw a canvas with 4096x4097 dimensions means you can't use an Off screen canvas to go from file to RGBA array.
36.
by xfalcox 2y ago
What about FP8? It is a target that is very popular for LLM inference.
37.
by xfalcox 2y ago
I've been daily driving it on arch since the day it released and it's quite usable, with minimal bugs for my needs. I just wish they add an applet that integrates Google Calendar stuff in their calendar, so I can know if I have a
38.
by xfalcox 2y ago
Is it just me or the 2 Llama 3.1 providers are the two know for running quantization levels that undermine Llama perf ?
39.
by xfalcox 2y ago
Can I just say that I was disappointed by the Web Codecs API leaving muxing the end result as a "draw the rest of the fucking owl" thing.
40.
by xfalcox 2y ago
I still get {"message":"Could not resolve the foundation model from the provided model identifier."} on us-west-2.
41.
by xfalcox 2y ago
They said it is available on Bedrock, but it isn't :(
42.
by xfalcox 3y ago
Worth noting that this also impacts the great https://github.com/huggingface/text-embeddings-inference , which allows anyone to run state of the art embeddings with great performance.
43.
by xfalcox 3y ago
Their other license also prohibits that > Cohere For AI Acceptable Use Policy > generating synthetic data outputs for commercial purposes, including to train, improve, benchmark, enhance or otherwise develop model derivatives, or any
44.
by xfalcox 3y ago
Does anyone have compiled this to WASM? I'm currently using MozJPEG via WASM for a project and would love to test replacing it by Jpegli.
45.
by xfalcox 3y ago
> Is that within the defined data processing purposes of all Discourse setups? It's an optional plugin that can be enabled / disabled by the site admin. Those modules are all disabled by default, and each need to be enabled by
46.
by xfalcox 3y ago
That's super cool, thanks for sharing! I will share this as an easy to follow example of what we can with AI. > Allowing a Q&A interface using these embeddings over the post contents could speed up research over the community po
47.
by xfalcox 3y ago
Adding a lazy, or going the extra mile of only loading a pure HTML div that is swapped by the iframe on click would be better to limit tracking. Do not load the iframe for people who don't want to play it.
48.
by xfalcox 3y ago
> Gemma 7B is a 9B model. The name is a lie Ohhh so that explains why I couldn't load it on my RTX 4090, while other 7B models load just fine!
49.
by xfalcox 3y ago
Context is 4096? My app db DDL is 19877 tokens (using Llama2 tokenizer) long, so that means we need to do a RAG for handling the DDL prompt injection. A model like this with a 32k long seq_len, like Mixtral, would be a killer for me.
50.
by xfalcox 3y ago
We've been looking for something exactly like this, thanks for sharing!
51.
by xfalcox 3y ago
There are many plugins, check https://meta.discourse.org/c/plugin/22 The flexibility of Ruby makes it kinda easy to change behavior. Full Disclosure, I work there.
52.
by xfalcox 3y ago
Can you share some of those ZigBee products? I'm looking into getting started on that, but Costa has been holding me back.
53.
by xfalcox 3y ago
I'm researching using LLMs for alt-text suggestion for forum users, can you share your finding so far? Outside of GPT-4V I had good first results with https://github.com/THUDM/CogVLM
54.
by xfalcox 3y ago
Embedding cost and model choice makes this a very compelling choice. I'm working on leveraging embeddings in https://github.com/discourse/discourse-ai where it powers offering related topics, semantic search, tag
55.
by xfalcox 3y ago
I wonder what the deal with Google Meet was. Did Firefox fix this? Did Google team fix this? If this is retroactively fixed for the last 3 versions it must be a google-side fix right?
56.
by xfalcox 3y ago
We are experimenting in this area, as a way to provide a way for community managers to query their Discourse PG database using natural language. Your product is very interesting. Maybe we can collaborate here, as Discourse is a standard Rai
57.
by xfalcox 3y ago
I know this isn't your focus, but could this be added as an index type to pgvector ? They recently added HNSW, and ship with IVFlat for a while, so I wonder if that would be possible.
58.
by xfalcox 3y ago
We have several customers who aren't using OpenAI API / Anthropic API because of privacy reasons. We are spinning up infrastructure and making the features who rely on those APIs to also wok with OS LLMs too.
59.
by xfalcox 3y ago
I added support for TGI endpoints in Discourse yesterday . That stings. I wonder if the license is aimed to block companies from building TGI-as-a-service only, or also applies for companies who use TGI internally for LLM inference that po
60.
by xfalcox 3y ago
São Paulo traffic is so bad that even Uber had an Helicopter option here for a brief period. https://flyflapper.com/stories/uber-air/

More ›