Ted Grönberg

[Ted Grönberg @GronbergTed]

@MiaAI_lab @MiaAI_lab anything new in your pipeline regarding RTX 5080 users like me. I am having a blast setting up my Qwen 2.1 to LTX 2.5 pipeline producing great videos. With your EXL3 Qwen 3.0 as the orchestrar via comfyUI MCP with Hermes

[Ted Grönberg @GronbergTed]

@UnslothAI It's my new primary driver when it comes to images I used to use the flux before but this is much better and also all the different notes that you can use to to really get the results that you want

[Ted Grönberg @GronbergTed]

@yume_arasaki 55 t/s on my laptop RTX 5080. 3.0 117.800CTV Mia labs version

[Robin Ebers @robinebers]

Jev is a much bigger deal than people believe

the average vibe coder has no fucking idea

this is the first time in a very long time that engineers regain a real advantage, because they can see the value of what this model does

it takes more architectural/systems thinking to understand where this could be useful, because it's not just a "drop in model replacement" but a completely different beast

but holy fuck is this thing amazing

huge kudos to @typesafeai for this

[Ted Grönberg @GronbergTed]

@PKU1898 Call of duty

[Ted Grönberg @GronbergTed]

@lightsilver323 You need it

[Ted Grönberg @GronbergTed]

@Lifeinvestmoney Lottery

[Ted Grönberg @GronbergTed]

@GuoShenCN7 Fu money....

[Ted Grönberg @GronbergTed]

@LeoBuilds_ Everyday. Claude to Qwen 27B local. But soon we maybe won't need Claude anymore:)

[Ted Grönberg @GronbergTed]

@HermesWatcher Wrong tool calls or skills loading up.

[Ted Grönberg @GronbergTed]

@Tech2Wild @Alibaba_Qwen I'm running @MiaAI_lab a 3.0 model Qwen 27b +vision +MTP on my RTX 5080 16VRAM laptop at 55/60 t/s. At 117.500 CTV. It's amazing. I'm the bottleneck now in my system.

[Mia @MiaAI_lab]

If you've been wondering how good it is...

Qwen3.8-27B on RTX 5080 laptop

From Q3 GGUF with 65k context, now running EXL3 with 148k context at 50 tok/s.

Try it.

https://t.co/TpJsbfLPUy

[Ted Grönberg @GronbergTed]

@MiaAI_lab Installed on my 5080 laptop. I got the 3.0 version. 148.000 ctv 50t/s this is a mad update from my Q3 gguf with 65k ctv. I can't thank you enough. You are a superstar. I tried pushing for more speed by twerking but it just got slower. Thanks 🙏

[Ted Grönberg @GronbergTed]

@MiaAI_lab I'm trying this today. Fantastic. Your are a superstar

[Ted Grönberg @GronbergTed]

@ZachChmael I built three harnesses into the workflow:Pre-approved tasks – the agent can run these freely
Flexible mode – more freedom, but only on non-sensitive code anything new or higher-risk is queued and held for my review in the morning (together with Claude as supervisor)

[Ted Grönberg @GronbergTed]

Before: solid vibe-coding only on days off, morning to https://t.co/uLgknBfL1j with Hermes + Telegram + local Qwen 27B I can push progress during work hours too.
And Cron keeps the agent working overnight.Local setup completely changed when and how much I can code.

[Ted Grönberg @GronbergTed]

@MiaAI_lab @jack746560359 @eth_brucewayne Yes! I can't wait and I'm having my vacation perfect timing. Qwen 4 will be the next game changer.

[Ted Grönberg @GronbergTed]

@ChristianXCesar I started with Q4 NS but changed to Q3 XL. I'm on a 5080 rtx. The quality drop is notable but it is so much faster on t/s tool calls is still ok. Have to /steer some times and supervise the start so it's on the right track. Prompt are super important.

[Ted Grönberg @GronbergTed]

Used to vibe-code: Claude → copy-paste → terminal. Weeks on bugs.
Now Hermes Agent + Telegram + local Qwen 27B. Reads code, tool-calls, finds issues, cleans up. I just review.
Claude only steers. Rest fully local.
Wild but night-and-day. Local agents finally work. Love MCP.

[Ted Grönberg @GronbergTed]

@lunkertw I still have the problem of my model timing out on the Claude MCP. Do you have any Trix. I'm cheap on the free tier still token maxing Claude.

[Ted Grönberg @GronbergTed]

@jack746560359 @eth_brucewayne @MiaAI_lab Me too