Samuel Castillo

[Samuel Castillo @samuel_571]

@MaziyarPanahi Thank you sir

[Samuel Castillo @samuel_571]

@zanoga Thanks for sharing! When you say:

“ … processing a lot of data every day.��

Would you mind sharing a bit more on the type of data and how you exploit it commercially? Trying to better understand the implied ROI

[Samuel Castillo @samuel_571]

@tomosman @NickADobos Looking fwd!

Thanks for sharing

[Samuel Castillo @samuel_571]

@jun_song > MB Air M5 24gb : Supergemma4-12B, mostly for remote works

Say more about “remote works”?

Thx for sharing!

[Samuel Castillo @samuel_571]

@mirrash7 @roboflow Congratulations! Have followed your work for a bit now, looking forward for what you share next

[Samuel Castillo @samuel_571]

@mirrash7 @roboflow Very clear, thank you!

[Samuel Castillo @samuel_571]

@soumitrashukla9 What’s you humble MacBook?

[Samuel Castillo @samuel_571]

@sudoingX What’s the equivalent starter stack for Mac users?

[Samuel Castillo @samuel_571]

@astridwilde1 @NigelHiggs7 What’s your local setup?

Thanks for this, very useful

[Samuel Castillo @samuel_571]

@MaziyarPanahi Does/could this support Spanish?

Thanks for your work!

[Samuel Castillo @samuel_571]

@PerceptualPeak @tobi @TomDavenport very cool stuff!

did you ultimately manage to make it work as you intended?

[Samuel Castillo @samuel_571]

@GeeFingBeeMan @Ex0byt What spec(s) fall short?

[Samuel Castillo @samuel_571]

@sudoingX @Teknium The exact comparison you mention - 30B-A3B vs qwen 35B-A3B

https://t.co/YBgRTspl1N

[Wei Ping @_weiping]

🚀 Introducing Nemotron-Cascade 2 🚀

Just 3 months after Nemotron-Cascade 1, we’re releasing Nemotron-Cascade 2: an open 30B MoE with 3B active parameters, delivering best-in-class reasoning and strong agentic capabilities.

🥇 Gold Medal-level performance on IMO 2025, IOI 2025, and ICPC World Finals 2025:
• Capabilities once thought achievable only by frontier proprietary models (e.g. Gemini Deep Think) or frontier-scale open models (i.e. DeepSeek-V3.2-Speciale-671B-A37B).
• Remarkably high intelligence density with 20× fewer parameters.

🏆 Best-in-class across math, code reasoning, alignment, and instruction following:
• Outperforms the latest Qwen3.5-35B-A3B (2026-02-24) and even larger Qwen3.5-122B-A10B (2026-03-11).

🧠 Powered by Cascade RL + multi-domain on-policy distillation:
• Significantly expand Cascade RL across a much broader range of reasoning and agentic domains than Nemotron-Cascade 1, while distilling from the strongest intermediate teacher models throughout training to recover regressions and sustain gains.

🤗 Model + SFT + RL data:
👉 https://t.co/4QJqfTOt6I

📄 Technical report:
👉 https://t.co/dFC00m6RZU

[Samuel Castillo @samuel_571]

@OfficialLoganK Pls bring logprobs back!

https://t.co/130ecKDhFs

[DEJAN @dejanseo]

request to bring back logprobs for 3 and 3.1 and have some consistency between vertex and gemini api 🙏 @OfficialLoganK @googleaidevs

https://t.co/zDouKwJXjo

[Samuel Castillo @samuel_571]

@heynavtoor What specific models did they use? I only see “Claude” or “Gemini”, but of course that doesn’t tell me anything

[Samuel Castillo @samuel_571]

@zackbshapiro Great read, thanks. Do you run out of context window / get degraded performance from getting close to its limit? How have you solved it?

[Samuel Castillo @samuel_571]

@RuskEating @atiorh +1

[Samuel Castillo @samuel_571]

@llorellama @signulll The red Coca Cola chairs proudly showcased in the gathering are also a big giveaway 🤣