Tanmay Patil

@TanmayPatil79

Building ML and AI things

tanmaypatil.com

Joined August 2021

505 Following

110 Followers

Replies Reposts Media

Pinned
Tanmay Patil
@TanmayPatil79
Jul 3
🚀 Just released: Krea-2 Depth ControlNet-LoRA Keeps near-perfect 3D structure while letting you completely change the image with any prompt. Works great with Krea-2 Turbo too. huggingface.co/Patil/Krea-2-d… Big thanks to @edwixxxx @Shauray7 who helped test and refine it along
4
Tanmay Patil
@TanmayPatil79
Oct 17
I distilled the FLUX 3 Klein VAE into a 2.7M-param decoder. 30× smaller than the 84M original, 11× faster at 1024², and 251× faster at 4K (78 s → 0.31 s). Write-up with drag-to-compare samples, editing tests and what didn’t work: tanmaypatil.com/posts/2026/tin… Thank you @bfl_ai for
1
Tanmay Patil
@TanmayPatil79
Oct 14
I gave 5 models (DeepSeek V4.1 Flash, Qwen 3.8 Max, Kimi K3, GPT 6.1, GLM 5.3) a terminal via @NousResearch 's Hermes Agent and asked them to write real GPU kernels for all 6 KernelBench-Hard problems. Results + all 30 kernels:
KernelBench-Hard × Hermes Agent | Tanmay Patil's Blog
From tanmaypatil.com
4
Tanmay Patil
@TanmayPatil79
Aug 11
My PR adding online FP8 support for Krea-2 and Krea-2 Turbo has been merged into SGLang
[Diffusion] Add online FP8 support for Krea-2 by Tanmaypatil123 · Pull Request #34136 · sgl-proje...
From github.com
1
Tanmay Patil
@TanmayPatil79
Jul 24
Nice weekend read
Anurag Kanade
@edwixxxx
Jul 24
New blog!! anuragkanade.com/blog/shrinking… I’ve been working on a creating a smaller version of a full duplex moshi in a mere 600M parameters with some distillation trick which is very underrated.