Quick question to all the TRL and *PO wizards. Has anyone tried resetting the weights of pretrained models and still able to properly RL the model? Any reference paper/code/blogs?

#llm #reinforcementlearning #policy #nlproc #machinelearning

Mark 3 should be stable code now!! #neuralempty #llm #nlproc

An adaptation of the model initialization part of T5Gemma that converts decoder-only models to encoder-decoder

But the UL2 feature for preprocessing data is missing though.

github.com/alvations/AED

Back to work... How do we evaluate translation quality in unsegmented video captions?

x.com/alvations/st...

#neuralempty #llm #nlproc #videostreaming #translation #ComputerVision #GenAi #VideoCaptioning

Here goes, I’ll start posting some more non #nlproc post. Promise that I’ll only share if it’s “hō-kāng”

Love Brit-humor / Monty #Python? Planning to watch Fackham Hall next week? Here’s a discount code from @LegionMOfficial

Code: Z1PD95W1Y

www.atomtickets.com/movies/fackh...

I’m back at the whole #neuralempty eval thingy. Reply with your recent QE or MT eval and I’ll put them on my paper blitz list and I promise to tag you and your co-authors when I post my blitzing posts/summary 😁

What’s the vibe on QE and #LLM as a judge these days?

There should be 2 internets. One for #llm bots, one for humans.

www.404media.co/ai-scraping-...

"Vibing" reward models, cool! arxiv.org/abs/2502.187...

#llm #nlproc

Interesting, Moore's law also has hyperparams knobs to tune. #llm #hardware

www.asml.com/en/news/stor...

It works!

This is the default channel for this user. The profile is running but no content has been added yet.

#nlproc #llm #python #geekout