Noah Cylich

Machine learning researcher and engineer, final year @ucla cs

[Noah Cylich @noah_cylich]

@GregoryDiamos Done this with needle 2 (v3 soon) from @cactuscompute, entire binary is 14mb

[Noah Cylich @noah_cylich]

@kisinepuchakya @cactuscompute Not yet! We recommend nvidias parakeet STT models, they’re the best in our experience, and we do support them with our engine

[Noah Cylich @noah_cylich]

@josephpollack @cactuscompute @max_paperclips Check hf: https://t.co/29XGITAAIu

[Noah Cylich @noah_cylich]

@gusnion @cactuscompute 25 MB of ram is enough!

[Noah Cylich @noah_cylich]

@kysstalol @cactuscompute Those models are like 1000x bigger, so I think largest we do is maybe lfm 1.2B

[Noah Cylich @noah_cylich]

@prodhi_code @cactuscompute CQ2 cuts file size in half and memory requirements by at least 1/3, SAN is how we optimize for so few params. They both go hand in hand a lot

[Noah Cylich @noah_cylich]

Insane work from my friends and colleagues. They pushed this thing to the absolute limits, 2 bit quant out of the box 🔥

[Cactus Compute @cactuscompute]

We release Needle 2: A 14MB agentic LLM for phones, wearables, smart home, robots and microcontroller. The whole model is a single 14MB binary that runs a full session in 28MB of RAM. It is built on our Simple Attention Network findings, compressed to CQ2-bit with Cactus Quants, and baked into its own engine.

Needle 2 has 45m parameters trained from the ground up on 140B tool call, device use structured generation tokens. On mobile device use benchmarks, Needle 2 trades wins with frontier small LLMs like LFM2.5 230M, Apple FM Gemma-270m, at 5× to 70× smaller, and 2 bits against their f16.

Needle hits 500 tokens/sec decode speed on a Raspberry Pi 5, between 400–1,500 tokens/sec on VR devices like Meta Quest 3S and Apple Vision Pro, and ranges 300–700 on sub-$200 phones such as the Samsung A-Series. Needle also runs on newer microcontrollers like ESP32.

A conventional transformer of Needle's width and depth spends 164 MFLOPs per token, and even one squeezed down to Needle's parameter count spends 87, Needle spends 70. Even on a high-end phone, an always-on assistant lives inside a power budget; every MFLOP is milliwatt-hours, and Needle spends 7x to 85x fewer of them per token than the smallest performant LLMs.

Read more: https://t.co/dLK2tIKXQu

[Noah Cylich @noah_cylich]

@Nintendeal Walmart keeps cancelling all my orders even tho I don’t have any active ones for the switch from any account (I had to cancel 1 previous order tho)

[Noah Cylich @noah_cylich]

@jakerandall What happens if you have multiple tabs open for the same retailer in same browser and same device BUT they’re each for a different bundle (ie switch and switch + mkw both open and all of Best Buy’s bundles open too)

[Noah Cylich @noah_cylich]

@jakerandall Source?

[Noah Cylich @noah_cylich]

@jakerandall @LeBlorstOfTimes I just called Best Buy customer support and they said the preorder time is 12AM but he didn’t seem like the most knowledgeable guy in the world

[Noah Cylich @noah_cylich]

@TheOnlyLisaRhea @RetroAgent12 undocumented immigrants are not the ones making streets unsafe 😭. Job prospects are horrible rn with recession fears at an all time high. “Safer Border” ig but what good does a safer border do when you’re not safe from ur own gov. Gas is down bc world fears a recession 😂

[Noah Cylich @noah_cylich]

@jakerandall The GameStop post is deleted?

[Noah Cylich @noah_cylich]

@IAmTheJMan2 @smc429 I get the sentiment but bro EVERYONE IS A “POTENTIAL CRIMINAL”

[Noah Cylich @noah_cylich]

@Drclaw411 @toxic2277 @jakerandall No I think they’d import as much as possible in the 90 days to supply the rest of the year in the US

[Noah Cylich @noah_cylich]

@jakerandall @Switch2Stock He did Germany yesterday

[Noah Cylich @noah_cylich]

@jakerandall @Switch2Stock Bro it’s been sold out for like a day anyways 😭 who knew that it would be cancelled until yesterday came and people’s orders failed

[Noah Cylich @noah_cylich]

@Infinite_Vibez7 @jakerandall @Stealth40k I don’t 100% follow. But the reason it’s not so simple as re-directing shipping networks is because the product’s country with respect to tariffs is the last country where major part of manufacturing was completed

[Noah Cylich @noah_cylich]

@NintendoSega54 @jakerandall @Stealth40k Well the US is having its own economic situation now and given the uproar, raising prices will lead to a much less successful system here

[Noah Cylich @noah_cylich]

@Felspawned @jakerandall @Stealth40k Couldn’t they still make more Japanese switches just for the US market though. Like even with Japan’s ~25% tariff, the cost would still be below the USs MSRP by like $100 for MKW