- Blablador-news - lists.fz-juelich.de

GLM 4.7 FULL
by Strube, Alexandre 22 Jan '26

22 Jan '26

After the great news of GLM-Flash, I decided to try the full, 385 billion parameter GLM 4.7. And guess what? It’s running! Compared to Minimax, it is slower, but better at math, and multi-modal. It can understand text, video, images and audio. While Minimax is theoretically capable of one million tokens, GLM is limited to 204 thousand. Given that is just slightly better than Minimax, while being slower, I am not sure if I will keep running it. What do you think? Leeeeeeet’s bark! Alex

1 1

What a difference a small patch makes...
by Strube, Alexandre 22 Jan '26

22 Jan '26

Following a hint found on Twitter [1], a little patch on VLLM allowed me to increase the context length on the GLM 4.7-Flash models from 8000 to 200000 (!) tokens. Now, those models get good rankings, but they REALLY overthink. The good thing is that if this patch works with other models, I can run the big GLM 4.7 on bigger nodes! Let’s bark! [1] https://x.com/TheAhmadOsman/status/2013881920099062163

1 1

GLM 4-7 Flash in two variants!
by Strube, Alexandre 20 Jan '26

20 Jan '26

Hello Blabladores! Some of you might have heard of the best medium-sized model released last night, GLM-4.7 Flash. You know the story: It’s slightly better than last week’s model of similar capacity. It’s a 31 billion parameter model, a mixture-of-experts with 3 billion activated. As we hear more and more about strongly quantizing models, I decided to run an experiment. I am running not one, but two versions of GLM 4.7-Flash: One with quantization and one without. Given that they are running on identical hardware (4x RTX 3090 each), something has to give: the not-quantized model has a smaller context length (around 18k tokens, vs 68k tokens for the quantized one). The quantized one is slightly faster, but not by much. I might have to optimize it still. In any case, they are both available for you to compare. What is better: Quantized but bigger context or not-quantized but smaller? You decide. Let’s bark! Alex

1 0

Ghristmas Mrrey!
by Strube, Alexandre 26 Dec '25

26 Dec '25

I found this jacket in a clothes stall at one of the Vienna’s Christmas market couple years ago. I regret to this day not having bought it. Anyway, I’m writing to let you know that NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 is running since day one, and from tonight, Minimax 2.1 is replacing Minimax 2 as our alias-huge model. Have some rest and go hug someone. Let’s bark, Alex, el Blablador

1 0

And we are back!
by Strube, Alexandre 03 Dec '25

03 Dec '25

So, the updates on the JSC infrastructure went pretty smoothly, and we are already back for some hours! While some instability might happen tomorrow, for now, we’re barking as loud as ever! PS: Mistral released their latest models, Mistral 3, yesterday. It’s already on Blablador for you to try with, and with function calling! Keep barking! Alex, el Blablador

1 0

Big models unavailable Wednesday and Thursday
by Strube, Alexandre 01 Dec '25

01 Dec '25

Dec 3, 2025, 07:00 - Dec 4, 2025, 19:00 A planned maintenance for the network backbone at the Jülich Supercomputing Centre (JSC) datacenter is scheduled from 07:00 CET on 2025-12-03 to 19:00 CET on 2025-12-04. During this period, all HPC, data, and cloud systems at JSC will be unavailable. Sorry for that :-( Alex, el Blablador

1 1

Blablador is DOWN
by Strube, Alexandre 03 Nov '25

03 Nov '25

The whole VM is down. I can’t access it myself. I’m going to the institute as soon as I can. Sorry for that. Alex

1 0

Phasing out Llama3-405
by Strube, Alexandre 27 Oct '25

27 Oct '25

Hello Blabladores, in order to improve the reliability of BLABLADOR, I am having to do some drastic changes, which required me to shutdown pretty much every single model and the server itself. OF COURSE while this happened, some GPUs decided do fail, which forces a full node reboot. I am bringing everything back up. If you are not working at the evening, just ignore this part. They are basically all up. Another thing is that I would like to phase out Llama3-405. It’s an obsolete model, it’s slow, and takes too many resources. Yesterday, Minimax-M2 came out, and it’s one of the best models out there. It’s ranked among models such as ChatGPT 5 and Claude 4.5. It is already running. It’s aliased as “alias-huge”. But as this is a reasoning model, you might filter what comes between the <think> and </think> on your code. Another thing is that this model might be able to generate images and music, but I need to figure it out. Let’s bark! Alex

1 0

Blabla fork for private use?
by Mario.Muellhaeuser＠dlr.de 17 Oct '25

17 Oct '25

Dear Alexandre, thank you for providing Blablador! I use it frequently for business. Now I am looking for a privacy friendly AI for friendly private use. I searched the internet and only found quite costly providers like mindverse, offering 250 tokens per month for 19 EUR. Can you suggest a more affordable AI aas provider, which offers privacy? Or are there plans or possibilities to open Blablador or a clone for public use, of course for a fee? Regards Mario -------------------------- 20 Jahre Forschungshubschrauber ACT/FHS - DLR Portal<https://www.dlr.de/content/de/artikel/news/2022/04/20221202_20-jahre-forsch…> DLR-Forschungshubschrauber: 20 Jahre Innovative Hubschrauberforschung - YouTube<https://www.youtube.com/watch?v=z_7RL3IsGrs&t=211s>

1 0

And we are good!
by Strube, Alexandre 16 Oct '25

16 Oct '25

Seems that most of the problems had been solved for now! Uhuu!

1 1