Ahmad Farhan

How much are you willing to bleed, to suffer for what you believe?

[Ahmad Farhan @_afarhan_]

Chris motionless optimising my life’s loss function one scream at a time.
#MIW

[Mahir 🇹🇷🇬🇧 @ScrewderiaF1]

Paraguay after scoring a goal

[Alex Weers @a_weers]

Today’s reads are about why self-distillation sometimes might not work and how to fix it

[Simon Weber @SimWeberTUM]

What if attention wasn't about matching tokens, but operating in function space?

Glad to share our #ICML2026 paper:
📄 Functional Attention: From Pairwise Affinities to Functional Correspondences

w/ @Jiefang_Xiao @GaoMaolin @stevenygd Daniel Cremers
📄 https://t.co/rhn9NtwrBm

[Red Hat AI @RedHat_AI]

Gemma 4 Diffusion landed in vLLM last week. Day 0.

First diffusion LLM natively supported in vLLM. Instead of one token at a time, it predicts 256 tokens at once and iteratively denoises them in parallel.

Result: 1,000+ tokens per second at batch size 1 on a single H100.

Built on Model Runner V2. @googlegemma

[Interesting things @awkwardgoogle]

Packing thousands of straws together basically creates a low-tech pixel screen.

Each straw acts as an independent light pathway, perfectly mimicking how data channels work.

[cargo short dad @cargoshortdad64]

BREAKING: Mistral reveals compute cluster for the upcoming Le Chaton Obése at 900T parameters. 50 Billion Blackwell equivalent, directly powered by the sun

[Ahmad @TheAhmadOsman]

Prediction

Fable 5 equivalent in Opensource is ~8 months away

[Grace Li @grx_xce]

BREAKING: Le Chaton Fat has fully saturated our benchmark.

We are at a loss for words.

In response, we are retiring Design Arena.

Congratulations to the @MistralAI team, and thanks for putting us on vacation.

[xlr8harder @xlr8harder]

he comes

[fabian @fabianstelzer]

Rumour mill going crazy on this new mistral model

- Napoleon class model with >10T params
- smokes Mythos on VoltaireBench
- for safety reasons only outputs French language code

[Roger Gilmour @RogerGilmour13]

@clashreport Sure

[Lisan al Gaib @scaling01]

the european mind is truly special

Mistral is very competitive with Gemini and GPT, wait (checking notes)
...
GPT-5.4-nano and Gemma 4 31B

[eric @ericzxchen]

how to win hackathons 101

[alphaXiv @askalphaxiv]

"MiniMax Sparse Attention"

This paper from Minimax adds a tiny Index Branch to GQA that picks top k KV blocks per group, then runs exact softmax only on those blocks, making sparsity GPU native, with exp free TopK and KV outer sparse kernels.

On a 109B multimodal MoE, it keeps dense GQA quality while cutting 1M context attention compute by 28.4x, with 14.2x prefill and 7.6x decode speedups.

[Alpin @AlpinDale]

In perplexity measurements, FP8 is pretty much the same as FP16, but NVFP4 is noticeably worse. Although not by that much.

[NVIDIA AI @NVIDIAAI]

Congrats to @GoogleDeepMind on the launch of DiffusionGemma.

The model generates 256 tokens in parallel per step, delivering 150+ TPS on DGX Spark, and 1,000+ TPS on a single H100.

We're supporting it from day one with:
• BF16 and NVFP4 checkpoints on @huggingface🤗
• Free GPU-accelerated endpoints on https://t.co/6T0R9P7EXS
• @vllm_project support with FP8 precision

Get started with DiffusionGemma on NVIDIA: https://t.co/vurk7GCQUs

[Lisan al Gaib @scaling01]

Fable 5 refused 200 out of 200 ProgramBench tasks lmao

[Google @Google]

Meet DiffusionGemma ⚡ Our latest experimental open model (Apache 2.0) that generates text up to 4x faster.

Instead of predicting and typing just one word at a time like most language models, it drafts and refines entire blocks of text simultaneously.

Here’s how it works 🧵 ↓

[vLLM @vllm_project]

🎉 Meet vLLM-Omni v0.22.0, a major upgrade for omnimodal world models and production-grade multimodal serving.

🌍 Day-0 @NVIDIAAI Cosmos 3 world models: text, image, audio, video, and action, in and out.
🤖 Robot serving: DreamZero + OpenPI realtime API.
🎙️ Production TTS: Qwen3-TTS, Qwen3-Omni, VoxCPM2 and more.
🎨 Faster image/video/diffusion: Wan 2.2, HunyuanVideo 1.5, LTX-2.3.
⚡ Broader quantization (FP8/INT8, MXFP4/MXFP8, W4A16, ModelOpt) and hardware coverage.

339 commits, 124 contributors, 52 of them new. Thank you all. 🙌

🔗 https://t.co/76ttSM6FHs