Talk to Sayak about
How Sayak thinks
Speed over brute force
Optimizes diffusion pipelines by adding KV‑caching, achieving up to 2.5× faster inference with modest memory increase.
For you: Try applying KV‑caching to your own model to cut runtime.
Open‑source first
Releases MIT‑licensed video‑audio models and contributes portable kernels, ensuring broad accessibility for the
For you: Consider using the MIT‑licensed Kandinsky model for rapid prototyping.
Mentor through code
Volunteers in Google‑related programs, guiding startups and students in machine‑learning and TensorFlow.
For you: If you’re starting with ML, ask for resources or mentorship.
Ask this first
- How does KV‑caching improve diffusion model performance?
- What challenges did you face releasing an MIT‑licensed video‑audio model?
- Can you share tips for enabling TPU execution in Diffusers?
Sayak's brain
Recently updated
In September 2026 I posted a series of X updates about recent work with 🤗 Diffusers, HF Jobs, and 🤗 Kernels, covering a one‑shot video…
In early October 2026 I shared a series of X (formerly Twitter) posts about recent advances I was working on: KV‑caching for flow‑type…
In September 2026 I shared three LinkedIn posts highlighting recent AI‑system contributions: enabling TPU execution for Diffusers via…
In October 2026 I posted on LinkedIn about three key developments: KV‑caching for flow‑based image generation, the MIT‑licensed Kandinsky…
I have received a variety of honors and awards for my contributions to open‑source machine‑learning, research competitions, and community…
I have been actively volunteering in Google‑related programs and local community initiatives since 2017, focusing on machine‑learning…