Cato Nooka

swe, cat, gym, ai, solo leveling

[Cato Nooka @catonooka]

GML-5.3 is a very good model to try, not having vision, but the reasoning is great, and the experience talking with with the model is generally better than most model!

[Z.ai @Zai_org]

GLM-5.3 is now open-weight.

Our most capable model for agentic coding and cyber defense is now available to download, run, and customize.

Weights: https://t.co/v1IbWMXxg4
Tech blog: https://t.co/ekQkO83jCv

[Cato Nooka @catonooka]

@TechMDAI ahh sorey, I thought 2 sparks

[Cato Nooka @catonooka]

@TechMDAI what?
145.8????

[Cato Nooka @catonooka]

@karanC_12 I bet flash is very small, and do better for RL

[Cato Nooka @catonooka]

@MiaAI_lab hey hermes, launch research session about EXL3, with this github repo ;))

[Cato Nooka @catonooka]

@djcows then X worth is current + 862B, which also adds up to his total again

[Cato Nooka @catonooka]

@LuminaBench so everyone is doing RL, or something like 'sparse' RL

[Cato Nooka @catonooka]

@pixelprayer

[Cato Nooka @catonooka]

@jauneclaude that's interesting, will try to see if it can handle 1-3 items/batch

[Cato Nooka @catonooka]

@himekako3150 LGTM!

[Cato Nooka @catonooka]

@patrickassale @dreamina_ai I have an idea:
- Gen the video until found satisfied game play
- Extract that video to game assets + upscale to 4k
- Start to build the actual game

[Cato Nooka @catonooka]

@MiaAI_lab @TechMDAI agree, I just read posts and do experiment too much and done almost none realistic work 😑

[Cato Nooka @catonooka]

@mirochill at this speed I can be a movie director for a small film production, just with a spark

[Cato Nooka @catonooka]

oh shit, I need to grab another key for Hermes now, I reposted some of their posts but didn't know they are the East HuggingFace

[GDP @bookwormengr]

How come these people do not know China has their own competitor to Huggingface called Modelscope?

You can download GLM-5.3-Flash, for example, from here.

https://t.co/9io9TuA9Px

[Cato Nooka @catonooka]

@superalesha running on a spark, the recent update seem very promising, 3x3090 will run, and 4x3090 will make some room for context

[Cato Nooka @catonooka]

@witcheer at this pace, you can basically register an account on Nous Research for Hermes, and one for OpenCode, then you can span tons of free token all day.

[Cato Nooka @catonooka]

@yachimat_manga can't wait ;))

[Cato Nooka @catonooka]

@LeTechLead no worry, if Nvidia add alot of filter and rules that changing the current HF, people will build a new one, even with option like hugging bay