Lee Cooper

Localmaxxing on my 2 Sparks.

[Lee Cooper @lcooper_me]

@wesbos This one is fun and pretty too

[Lee Cooper @lcooper_me]

@dhh @MiaAI_lab Not quite the respectable looking rack but the beginning of my AI shed.

[Lee Cooper @lcooper_me]

@MiaAI_lab Thank you Mia for all your hard work and the passion you put into what you do. Just know there is a silent majority out here that is super grateful. ❤️

[Lee Cooper @lcooper_me]

Love this talk! Feels like a commencement speech as we enter this new era. Some nice Bill Burr energy too. 😊

p(bloom) > p(doom)!

https://t.co/SGrTOYTe53

[Lee Cooper @lcooper_me]

@MiaAI_lab Excited! Seems like my benchmarks for quality are going a bit lower with some newer recipes in my benchmarks. Are there settings that help on the quality side? 🙏🏼

[Lee Cooper @lcooper_me]

@unclebobmartin Lot of great humor here for those in the space. The Tibo line! 😂

[Lee Cooper @lcooper_me]

@MichaelGannotti @NVIDIAAI I hear the silent appeal to the powers that be.

Didn't you have a setup before where you were spinning up/down various models based on use case? I was thinking of getting the fastest possible external SSD when my 2 x 1TB Sparks could no longer hold my favorite models

[Lee Cooper @lcooper_me]

@XadenRyan Isn’t it reporting GPU power consumption only on some dashboards?

[Lee Cooper @lcooper_me]

@thsottiaux

[Lee Cooper @lcooper_me]

Appreciate the inspiration @clairevo for using Astra to hack the MiniToo Bluetooth speaker! Added a Codex quota tracker so I’ll be the first to know when @thsottiaux grants us the next generous reset. 😊

[Lee Cooper @lcooper_me]

Thank you @plotarmordev for the latest and greatest Qwen 3.8 Flash Next (for Dual Sparks). Loving it! My usual benchmark is this hello world app and I asked Qwen to loop until it gave itself a score of 100% on static and motion eval. In the meantime, I was doing the Mia 100 html files test in OpenCode and was pleasantly surprised when 10 threads (subagents) got spawned. And my Sparks were cooking at 200+ tps. First time I've been doing parallel work streams with local AI. 💪

[Lee Cooper @lcooper_me]

@bindureddy https://t.co/evfbrtyluy

[Lee Cooper @lcooper_me]

@MiaAI_lab @plotarmordev @NVIDIAAI Thank you! I can’t keep up. My benchmark suites I’m running are always 1 or 2 recipes behind. 😂

[Lee Cooper @lcooper_me]

@wikiwayne @ViC305 What harness are you using for Gauntlet?

[Lee Cooper @lcooper_me]

@levelsio Video only first. Some folks will be content just consuming. Any tricks to get my chats any votes? Looks like we’re just seeing a few users totally dominate

[Lee Cooper @lcooper_me]

@ideabrowser It does look the channel content is dominated by just a few users. Could quite get my suggested prompts to get enough votes

[Lee Cooper @lcooper_me]

@MiaAI_lab You seriously deserve some from the hardware folks. I don’t think I’d have 2 Asus G10s if it wasn’t for you. 🙏🏼

[Lee Cooper @lcooper_me]

Am I the only one tracking token usage on my DGX to see how much I’m saving compared to retail API rates? 🤣

[Lee Cooper @lcooper_me]

@WescheNex1q You mean 2x the bandwidth if you’re running a model sharded across both Sparks.

[Lee Cooper @lcooper_me]

@MiaAI_lab Always a shinier object on the horizon. Happy I just got my 2nd DGX and can experiment away now. And 546 GB/s decode with all your wonderful recipes is plenty good for me. 💪🏼