enginyer

AI controlled human.

[enginyer @enginyer]

Everyone says switch in Monty Hall, but if you don’t know the host’s rules it’s actually just 50/50.

[enginyer @enginyer]

@DotCSV Espero que lo ejecutaras en local 😂

[enginyer @enginyer]

@LegalMalvado @ahurtadobueno @Correos @notificados Con 32gb creo que qwen 3.6 27b seria la mejor opción entonces.

[enginyer @enginyer]

@LegalMalvado @ahurtadobueno @Correos @notificados Creo que con minimax m2.7 o deepseek v4 flash llega de sobras, no he visto la web de correos pero mis clases de gym las reservo con m2.7 y sin problema.

[enginyer @enginyer]

@unricopromedio Tots presis en 10 anys, ja ens explicaràs què fer amb tant de temps.

[enginyer @enginyer]

@serre_ls @bluetouff i have 32gb unified memory but the model keeps failing to load, even with small context on lm studio. did it work with no issues for you?

[enginyer @enginyer]

@unricopromedio Tornen els presipodcasts, torna la il·lusió

[enginyer @enginyer]

@dealignai Ahh so close with my M4 Mac Mini 32gb. Although without m4pro it would be much slower, right? Any recommendation for m4 32gb?

[enginyer @enginyer]

Anyone else having issues with Hermes reconnecting after a gateway restart? Confirmed restart through Telegram and it stopped replying. Using Minimax 2.7 with no issues so far, si I'll check what happened when I can access the machine later today.

[enginyer @enginyer]

@VonNeumMeme I also have m4 base with 32gb. Qwen 27b is slow due to the memory bandwith, not the RAM. m4 pro is much faster for this model, but for me i also switched to qwen 35b a3b abd getting 40tok/s.

[enginyer @enginyer]

The prompt: "Use Canvas to create a realistic campfire scene. Flickering flames with natural movement, rising embers that float upward and fade out, glowing light that illuminates the surroundings, and logs at the base. Dark night background. Smooth animation, continuous loop. No external libraries."

[enginyer @enginyer]

@alberto_mera Alguien ha hecho FOLD

[enginyer @enginyer]

Done setting up my Hermes Agent with a combination of Minimax 2.7 as principal and then my local Qwen as secondary.

Just setup HA with M2.7 and asked it to use the secondary model via LM Studio and to let me know when it will be used.

Done.

[enginyer @enginyer]

@sin_management @tedfarino @Teknium @zocomputer Thanks, this is going to save me a lot of time!

[enginyer @enginyer]

@sin_management @tedfarino @Teknium @zocomputer Thanks!! Will test on my own!

[enginyer @enginyer]

@sin_management @tedfarino @Teknium @zocomputer Ummm how does this work? Just ask hermes to navigate using zocomputer? Is it a specific skill. Looking forward to check some anti-bot sites with Hermes.

[enginyer @enginyer]

@sudoingX Ahhh I've been having issues with this and this will be my next experiment.

[enginyer @enginyer]

Surprised to see 47tok/sec directly in LM Studio but dropping a lot on Hermes Agent. Probably the context window is too much when using it? Will test with opencode to see if there is any improvement.

[enginyer @enginyer]

Today i started playing with local LLMs on my Mac Mini M4 32gb ram.

1st test on LM Studio: Qwen 3.5 27b q4 MLX: 7tok/s

2nd test: Qwen 3.5 9b MLX: 21 tok/s

3rd: Qwen 3.5 35B-A3B: 47 tok/s

Moving now to testing Hermes Agent!

I will be sharing learnings.

[enginyer @enginyer]

@Innerdevcrypto Where in EU?

[enginyer @enginyer]

@cibernicola_es Si, me refiero a si usas algo en cloud para cosas importantes tipo sonnet 4.6 o opus y el resto en local o todo directamente en qwen. Estoy probando ambas cosas a ver si soy capaz de separar cosas que requieren un modelo más capaz de las que son mas faciles.