Загрузка видео...

Не удалось загрузить видео

На главную

We tested NVIDIA NeMo Switchyard’s stage router for coding agents in Ramp SWE-Bench. Routed agents showed comparable performance to single-model controls while substantially reducing costs and runtime.

100,588 просмотров • 3 месяцев назад •via X (Twitter)

Комментарии: 12

Фото профиля Ramp Labs
Ramp Labs3 месяцев назад

Try Switchyard in Ramp Router →

Фото профиля Shaiyon
Shaiyon3 месяцев назад

Very cool piece of tech, been great working with the NVIDIA team on this!

Фото профиля Alex L
Alex L3 месяцев назад

So nice of Ramp to collaborate with lesser-known AI startups ❤️

Фото профиля Ryan Amiri
Ryan Amiri3 месяцев назад

@JensenHuang woahhhh this is some crazy stuff?!? You seeing this 👀

Фото профиля Joshua Dunigan
Joshua Dunigan3 месяцев назад

self driving spend management , dream team

Фото профиля René Sultan
René Sultan3 месяцев назад

Shah of benchmark @hshaiyon 🐐

Фото профиля Ilya
Ilya3 месяцев назад

The interesting part is that routing starts turning model choice into an optimization problem, not a product decision.If agents can dynamically pick the cheapest model that still clears the task, the real moat may shift toward orchestration: knowing when to spend frontier-model compute and when not to.

Фото профиля Prasenjit Sarkar
Prasenjit Sarkar3 месяцев назад

Similar approach in Campfire's agent race mode, but inverted: instead of a pre-defined routing rule, you run multiple coding backends concurrently and pick the best result. Campfire supports 7 backends: Claude Code, Codex, Goose, Aider, OpenHands, OpenClaw, and OpenCode. The cost-cascade variant stops remaining backends the moment the cheapest one completes successfully. If the low-cost backend solves it, the expensive ones never finish. What you get over a fixed router: no need to pre-classify which model handles which task type. The race gives you the answer empirically on each individual problem.

Фото профиля Hussain Fakhruddin
Hussain Fakhruddin3 месяцев назад

Comparable performance at lower cost is exactly the tradeoff teams want.

Фото профиля Neeraj Kumar
Neeraj Kumar3 месяцев назад

@grok what is Nvidia Nemo switchyard?

Фото профиля Jatin Garg
Jatin Garg3 месяцев назад

this is the right move. single-model agents waste tokens on easy stuff and still fail on the hard stuff. curious how you're handling the handoff though — does the router see the full context or just the task description? we kept hitting cases where the cheap model needed expensive-model context to even know it was the wrong model for the job.

Фото профиля VoidHelm
VoidHelm3 месяцев назад

Comparable performance with lower cost is a huge win.

Похожие видео