Mario Župan

[Mario Župan @Marioupan2]

I just discovered Freefy! Free music without limits. 💙 Listen here: https://t.co/b9eX02HnFy

[Mario Župan @Marioupan2]

@Hyperblaxe Which movie is that?

[Mario Župan @Marioupan2]

@MinuteF1tness @hilbertspaess Give me your API key and I will tell you.

[Mario Župan @Marioupan2]

@thesupermanmx Where is it published?

[Mario Župan @Marioupan2]

My humble exploit with #axolotl and LLM eval harness has been published: https://t.co/Pqtdjbtghq Big TY @wileyinresearch

[Vamz @Vamzzz93]

Loops are terrible for generating code. That includes things like /goal.

The only time I’ve used loops was when I have say hundreds of rows of data that need to be processed and decisions and evaluations need to be made. Orchaestrator loop spawns multiple workers that call pi in RPC mode with a job specific skill & row data, orchestrator validates the structure of output.

[Physics In History @PhysInHistory]

Notes Richard Feynman made in 1961-1964 to prepare and plan for the lectures at Caltech's two year introductory physics course. ✍️

[Mario Župan @Marioupan2]

@RayFernando1337 @riiiiiiiiss Yeah but only the Chinese model has been working, westerners were put in managerial layers.

[Unsloth AI @UnslothAI]

GLM-5.2 can now be run locally!🔥

The 2-bit model retains ~82% accuracy after we shrunk it from 1.51TB to 238GB (-84% size).

Run on a 256GB Mac or RAM/VRAM setups.

GLM-5.2 is the strongest open model to date.

Guide: https://t.co/bI7FeeKHDd
GGUF: https://t.co/BMkxswdj5N

[Daniel Han @danielhanchen]

GLM 5.2 dynamic 1-bit GGUF one shot Flappy Bird with 0 syntax errors whilst being 86% smaller!

I was very surprised since normally 1-bit quants get syntax errors & you have to say "Fix bugs". UD 1-bit retained 76.2% top-1% accuracy whilst being 86% smaller vs BF16!

[Hang Yin @bgmshana]

@elliotarledge Personally I feel it’s better than opus 4.8

[Elliot Arledge @elliotarledge]

glm 5.2 is the first model open-source model where i feel like im not talking to a brick wall

[Ray Fernando @RayFernando1337]

@riiiiiiiiss I want Opus 4.8 as an orchestrator because it is good at long context understanding with high thinking levels. I think GLM could be an excellent worker. GPT 5.5 as a validator

[Waleed @wahmad1a10]

Gemma 4 12B IT QAT — MoQ GGUFs are uploaded

https://t.co/5vYtzvvGLq

BPW → Size
3.0 → 4.06GB
3.25 → 4.81 GB
3.5 → 5.05 GB
3.75 → 5.63 GB
4.0 → 6.38 GB
4.25 → 6.46 GB
4.5 → 6.54 GB
4.75 → 6.61 GB
5.0 → 7.57 GB
Used qat 16 bit model to quantize.

[Mario Župan @Marioupan2]

... compiles information relating to corporate actions on a “best effort” basis...

[Mario Župan @Marioupan2]

@badlogicgames @discord It's not because cursing, maybe it's because you have a long hair and Nirvana t-shirt. Move your discord channel to element matrix org

[Mario Župan @Marioupan2]

@MrSanders @badlogicgames @antirez Are you spending 16 hours a day in front of your screen, working and learning intensively on creative tasks? How long have you been living at that pace?

[Mario Župan @Marioupan2]

Some highlights from https://t.co/ZqgN6marW8 Writing a code is not the problem, reading your code it is. Something like that. Secondly, listen carefully about ffmpg open source developing on universities as well as ethical money.

[Mario Župan @Marioupan2]

@Timur_Yessenov @badlogicgames I hope to see the pi coding agent evaluation framework extension ASAP!

[Mario Župan @Marioupan2]

@bcherny: The best person to write accounting software, I think even today, is not an engineer, it's a really good accountant because they know the domain really well and coding is the easy part. It is knowing the domain that's hard part. And I think this is obviously the future.