Azeez

Building Atlas, pure Rust and CUDA/ROCm kernels on DGX Spark & Strix Halo | MLPerf Agentic Edge Taskforce | Ambassador @Alibaba_Qwen | Ambassador @DevinAI

[Azeez @AtlasInference]

@dad__vibes @RicursiveAI @AIatAMD DFlash2 definitely sped things up quite a lot. Let me know how it's running :)

[Azeez @AtlasInference]

@RicursiveAI's Ripple-65B-NVFP4 runs on ONE AMD Strix Halo desktop on your desk 🖥️

Fitting across 128 GB unified memory, with DFlash2 drafter this massive DENSE model decodes with pristine BFCL accuracy at 6.3 tok/s ⚙️

@AIatAMD's Strix Halos can run some serious models 🔥

[Azeez @AtlasInference]

Try it yourself!
git clone https://t.co/GdG5uesFhl && cd atlas ./build-amd.sh

DFLASH=1 DRAFT_MODEL=ricursiveAI/Ripple-65B-Preview-DFlash2 GPU_UTIL=0.95 MAX_SEQ_LEN=8192 MAX_BATCH=1 ATLAS_KV_EXTERNAL_RESERVE_GB=0 ./serve-amd.sh ricursiveAI/Ripple-65B-Preview-NVFP4 --dflash-window-size 16384

[Jason McCartney @jmacftw]

All of our SparkScore demos are real-time, recorded against a serving endpoint on one of our @NVIDIAAI DGX Sparks. SparkScore technology has been contributed to @AtlasInference - the O.G. of Rust-based inference engines.
https://t.co/Nm5NnTBAja

[Azeez @AtlasInference]

@PlsHoldMyHalo @alibaba_cloud @Alibaba_Qwen I've said it before but good things come to those who wait 😀

[Azeez @AtlasInference]

Give early-access Qwen4 an @alibaba_cloud models crew like Image, Wan-2.7, Audio, Omni and see 🎥

It storyboarded, shot 10 clips, scored/re-scored after hearing a first mix, caught and fixed its own broken first cut, and signed off 🫡

@Alibaba_Qwen is cooking... can't wait 🍿

[Jason McCartney @jmacftw]

We started an open effort to train a DFlash2 drafter for GLM 5.3 Flash. It's already 92-95% as fast as one of the best drafters for the model and totally open licensed. Plug and play with most model engines and works great with our SparkGLM and @AtlasInference .

[Jason McCartney @jmacftw]

We have a 27B decision model reading a 12,000-token handbook on every request and making a decision in 191ms on one @NVIDIAAI DGX Spark. It follows the rules every time and the model hasn't been fine-tuned.

Here's what we did. 🧵

[Azeez @AtlasInference]

@Wappy_x @jmacftw @Wappy_x that’s the whole point! We don’t intend to sacrifice any of the quality while upping the speeds, have strict evals in place to ensure this :)

[Jason McCartney @jmacftw]

“I wanna go fast!” - Ricky Bobby

We got a very smart 27B decision model answering in 146 ms. On one DGX Spark. On your desk.

The video shows real time serving of an actual endpoint. The possibilities are endless! @AtlasInference

[Azeez @AtlasInference]

SWE-2 is already fantastic, so excited to blaze through kernel dev with this flash variant 🔥

[Cognition @cognition]

Introducing SWE-2 Flash, our new fast & efficient coding model.

On leading evals, it outperforms all American open-weight models, even ones 3x larger.

To balance transparency and cyber risks, we are building a program to securely share the weights with partners & researchers.

[Azeez @AtlasInference]

@aijoey @DevinAI is amazing for making SWE-2 free through November 🎉

[Azeez @AtlasInference]

@Wbackdown @Alibaba_Qwen @jmacftw Long-horizon is holding up well, it's being utilized for lengthy agentic sessions for some of our users :)

[Azeez @AtlasInference]

FP4-Local=better quality/BFCL accuracy, smaller footprint 🤏

TP=2, @Alibaba_Qwen's 3.8-Flash-Next 108 tok/s code, 260+ for 4 users 📈

Prefill ~2.3× vLLM, one 17K prompt, 5.8 vs 9.2s 💻

Worked w/ @jmacftw, >NVFP4 🧠

Try it:
sparkrun run @atlas/qwen3.8-flash-next-fp4-local-tp2

[Azeez @AtlasInference]

@PlsHoldMyHalo @Alibaba_Qwen 🤫

[Azeez @AtlasInference]

Our community is so awesome ♥️

@Alibaba_Qwen Flash-Next orchestrated the whole thing with support from Voxtral for voices, Minimax for video and music, and Krea 2 for images 🎬

Running e2e through Atlas ☺️
https://t.co/yQalnYlOCK

[Azeez @AtlasInference]

@PlsHoldMyHalo @Alibaba_Qwen Can't wait. Great time to go local 😉

[Azeez @AtlasInference]

@defipoodle @DevinAI @GeorgiaTech @dabit3 @cognition @HackerHouseCo Well they can because the Devin Max plan given to builders extends to November 24th 😉

[Azeez @AtlasInference]

A break from the usual... as a @DevinAI ambassador, I co-hosted a hackathon at @GeorgiaTech, met the most cracked builders on campus 🔥

100+ students, 7 hours, and >25 shipped projects 🚀

Thank you @dabit3, @cognition, @HackerHouseCo 🐝

[Azeez @AtlasInference]

@abheer @DevinAI @GeorgiaTech @dabit3 @cognition @HackerHouseCo I hope to do a hackathon one day with @nvidia too! As you know there’s a laundry list of NVIDIA-centric components to build with :)

[Azeez @AtlasInference]

@dabit3 @DevinAI @GeorgiaTech @cognition @HackerHouseCo I appreciate your involvement throughout this to make it possible. You made sure it was a seamless process from start to finish 😁🥳