metis foundation: a series of innovations on the cutting edge of web3 & AI, by e. sinelnikova et al (2025)

how flashattention 4 works, by GPUMODE

deep learning determinism, by d. riach (2019)

the state of open models, by n. lambert (2025)

towards deterministic inference in sgLang and reproducible rl training, by the sglang team (2025)

determinism and reproducibility and floating-point non-associativity

parallax: efficient distributed llm inference on heterogeneous hardware, by tong et al. (2025)

memory fragmentation when storing KV cache of several different shapes: slab allocation to build unified KV caches for every possible shape (every KV cache region - VRAM or DRAM- is divided into fixed-size chunks/slabs)

aegaeon: effective gpu pooling for concurrent llm serving on the market, by xiang et al. (2025)

docs: just a bunch of bittensor projects for arxiv (with no preference, order, or endorsement)

a few more nice articles for this beautiful sunday research hours - now back to building the next gen of prediction

another delightful essay and a picture of a teacher in the future

add x explains z% of the variance in y, by l. lang as a delightful essay for this beautiful peaceful sunday morning

the new AI infrastructure stack part I, by devansh

AI, mars, and immortality: are we dreaming big enough?, with peter thiel