ray

transformers

[ray @rayzarion]

@jayair Fable fumbled hard

[ray @rayzarion]

@slime_framework Slime is everywhere lol

[ray @rayzarion]

@kimbochen @yifanzhang_ Seperate rollout engine for teacher

[ray @rayzarion]

@rahaincrease @OpenAI Hitting codex limit faster than Claude. the irony.

[ray @rayzarion]

@bridgemindai Also slow af

[ray @rayzarion]

@mweinbach Quantization rollout started

[ray @rayzarion]

@sheriyuo Exactly

[ray @rayzarion]

@dejavucoder releasing a sota model on the first attempt is pretty hard.

[ray @rayzarion]

@bcherny by how much?

[ray @rayzarion]

@liranringel Will this work with mtp ?

[ray @rayzarion]

@ClementDelangue this model must hallucinate a lot 😂

[ray @rayzarion]

@jackminong @ShuaichenChang Lora is a better option for RL, https://t.co/BAn0Q4BMIv

[ray @rayzarion]

@jxmnop Building a frontier model is not gonna happen in one iteration

[ray @rayzarion]

@levelsio @jeremyphoward using aliases makes it easier

[ray @rayzarion]

@casper_hansen_ probably also used indexcache

[ray @rayzarion]

@cloneofsimo cursor is biased towards openai and they are really bad at hiding it lol

[ray @rayzarion]

@eliebakouch @claudeai love it

[elie @eliebakouch]

Muon optimizer explained by the new @claudeai interactive feature, this is so nice

[ray @rayzarion]

@m_sirovatka @samsja19 Any guess on compute req for Lora ?