Emiliano Conti @Rekeyea

Software Engineer. Lawful Good. #WindowsInsiders He/Him Joined September 2013

Tweets

2K
Followers

141
Following

954
Likes

4K

Emiliano Conti @Rekeyea

2 days ago

@EnoReyes Is this already available? Not seeing it :(

0 0 1 75 0

View Details

Excited to launch Luce Spark: now a 35B MoE runs on a 16GB GPU, with no offload tax. An A3B model fires ~8 of its 256 experts per token, but to keep it resident you pay VRAM for all 256. Spark pins the experts your traffic actually hits, offloads the rest to CPU, and decodes the whole token in one fused graph, so offload stops costing speed. ▸ Qwen3.6 35B-A3B: ~20.5 → 13.3 GiB ▸ Laguna XS.2 33B-A3B: 18.8 → 14.6 GiB Decode holds ~100 tok/s, close to the 119 you get with every expert resident on a 24 GB card. No calibration step. It tunes itself from live traffic.

Sandro @pupposandro

a week ago

x.com/i/article/2063…

10 16 159 69K 211

14 16 150 19K 101

View Details

Emiliano Conti @Rekeyea

2 weeks ago

@VictorTaelin May I suggest droid missions? Or is that a path you already followed?

0 0 0 244 0

View Details

Emiliano Conti @Rekeyea

3 weeks ago

@skcd42 Thnx for the copy/paste fix it was so annoying

0 0 1 17 0

View Details

Emiliano Conti @Rekeyea

3 weeks ago

@LottoLabs Also for the 3090 TI? A little bit confused because of the nvlink not being supported?

0 0 0 24 0

View Details

Emiliano Conti @Rekeyea

4 weeks ago

@skcd42 We need higher limits

0 0 0 29 0

View Details

Emiliano Conti @Rekeyea

4 weeks ago

@sudoingX I'm not finding much info (or don't know where to look at). Would you say a double 3090ti is a good local AI rig (I already have one) or should I go for a bigger gpu (or more than one like Arc B70)?

0 0 0 19 0

View Details

Emiliano Conti @Rekeyea

4 weeks ago

@0xSero I'd suggest @orca_build as best ADE

0 0 0 275 0

View Details

Emiliano Conti @Rekeyea

4 weeks ago

@thsottiaux can we have non-reasoning in codex? Like I can in Pi?

0 0 0 5 0

View Details

Emiliano Conti @Rekeyea

a month ago

@davideciffa @poolsideai Amazing!!!

0 0 2 20 0

View Details

Emiliano Conti @Rekeyea

a month ago

@orca_build is everything I wanted for an ADE. Loving the ergonomics and the freedom to use the best tool for the job while adapting perfectly to my workflow

0 0 0 6 0

View Details

Emiliano Conti @Rekeyea

a month ago

@thsottiaux Obviously Frontend. Also I miss my workflow in Pi where I can tell de harness to Self correct whenever I find something that is missing or need to configure. Worktrees workflows could be improved too to make it clearer. And lastly an editor with LSP

0 0 1 319 0

View Details

Emiliano Conti @Rekeyea

2 months ago

@0xSero Amazing! I will try to replicate then. Any guidance on this would also be awesome. And thanks for the work you been doing lately 🙏

0 0 0 65 0

View Details

Emiliano Conti @Rekeyea

2 months ago

@kskrygan Yessss

0 0 0 6 0

View Details

Emiliano Conti @Rekeyea

7 months ago

@shehab_amins Wow! I think Sail goes the right direction. Would you say it is ready for streaming capabilities?

1 0 0 139 0

View Details

Emiliano Conti @Rekeyea

a year ago

@mgechev The first one :)

0 0 2 112 0

View Details

Steve Sanderson @stevensanderson

2 years ago

Introducing Microsoft.AI.Extensions - a library to streamline use of AI in .NET apps. Glad to see this go live today! The library is something I and a few smart people have been working on in recent months. devblogs.microsoft.com/dotnet/introdu…

12 109 393 38K 176

View Details

Emiliano Conti @Rekeyea

4 years ago

@mgechev I do like self closing tags. And being able to use the component by name. I would allow to rename it if People want just to keep backward compatibility. And agree that JSX is not a Good idea. As many People also said, funcional components for presentational would be amazing too