> vLLM and SGLang are complex, and bugs are common
This for me is the heart of the issue. Feature creep will lead to the downfall of all these frameworks. Today, we can conjure our own bespoke inference engine for our own hardware in no time. It need only support a few modern model architectures. The code can be audited too. I think the article highlights an important gap area for the industry.
No need to appease me. I am on the lookout for a Trello replacement, but may end up self-hosting or vibe coding my own. I have a dream of marrying plain HTML _details elements with _textareas where the _summary is an editable _input.
Same here with FreeTube. YT has become a toxic experience on any browser and recently it has stopped working on my laptop with Firefox and uBlock Origin, no matter if I surf it as logged in in my main google account or not. The same arrangement works on my main desktop, but i mostly use FreeTube there as well.
Adding that once one accidentally clicks on a short all the results are skewed in that direction afterwards. I have to close the browser, run BleachBit and start over.
Folks, this is what bad stats look like. All bot accounts look and behave like this. They didn't even do a systematic study to prove their claims against other bot acconts (which are not associated with deepseek at all).
I don't see any network analysis on this page. What network analysis do you see?
I do see generic statements like "boosting each other", and I see vaguely-drawn lines in the primary diagram with no further explanation, but that hardly counts as network analysis, right?
interesting comment... what are you doing with base models? Are you a "finetuner"? I have been trying my hand with finetunes on instruct models and the results have been ok, but not awesome. I have a base model downloading now to give that a proper shot.
I'm not them but I still prefer a text completion style of prompting rather than a baked in pre-prompt structure assuming only a 'chat' style metaphor of interaction.
