Transformer Inference

topic on 1 show · 3 statements across 2 episodes

Latent Space

3 statements about Transformer Inference, every show

Sohmers: Matrix-vector multiplication in transformer inference is fundamentally uncacheable
“So the second level of this is that matrix vector multiplication is basically uncacheable. When you're doing transformer inference, matrix A is the weights of your model. And so if you're talking about model weights that are tens of gigabytes, hundreds of giga…”
Thomas Sohmers Aug 18, 2025 ▶ 13:33 ⚡️Accelerators @ 3x NVIDIA H200 perf, Made in the USA - Thomas Sohmers + Mitesh Agrawal, Positron AI
Sohmers: AI hardware over-indexes on raw FLOPS instead of memory bandwidth
“Everyone else was focusing on the wrong things. They were just trying to have more and more flops when memory bandwidth, memory capacity were the real, real bottlenecks.”
Thomas Sohmers Aug 18, 2025 ▶ 6:26 ⚡️Accelerators @ 3x NVIDIA H200 perf, Made in the USA - Thomas Sohmers + Mitesh Agrawal, Positron AI
Jamil: KV Cache Prefilling Scales Quadratically While Token Generation Is Linear
“Token generation, which means generating one token using whatever is in the kvcash, is linear with respect to whatever is in the side of the kvcash. Prefilling the kvcash is quadratic, and mostly because it's quadratic, it's very expensive, so we are talking a…”
Umar Jamil Sep 19, 2024 ▶ 42:03 [Paper Club] Writing in the Margins: Chunked Prefill KV Caching for Long Context Retrieval

← every entity, every show

Made with StarZero

Turn any episode into a week of clips.

This entire site, thousands of episodes across every show transcribed, diarized, checked and made playable, runs on the StarZero media pipeline. Drop in your own episode and the podcast clipper finds the moments worth sharing, cuts them, captions them, and reframes them for every feed.