Julien Genoud retweeted
⚡️ We are excited to share updates from KDA-v0.5 (kernel design agents). Our latest Cute-DSL kernels can beat the human winners at FlashInfer Kernel Contest by 16%~69%! An amazing progress in just 3 months!
- GDN prefill: 1.69x Speedup over Kachua
- DSA attention: 1.41x Speedup over Dogacel
- FP8 MoE: 1.17x Speedup over Team Wombat
Humanize2 Flame Chase: docs.humanfia.ai/humanize2/f…
Results and Reproduction: github.com/mit-han-lab/mlsys…
MLSys 2026 Contest: mlsys26.flashinfer.ai/
KDA-v0.5 achieved this by integrating the Cute-DSL primitive, better workflows (humanize1 -> humanize2), updated kernel-wiki (self-evolved), and better profiling skills (IKET). The results are achieved by the flame chase flow from humanize2 – using gpt-5.6-sol and fable-5 to iteratively optimize . We have released the kernels for validation and more details will come soon!
By the KDA Team: Dongyun Zou, Yixin Dong, Junxian Guo, Changye Li, Yahui Cui, Zihao Ye, Junru Shao, Zijian Zhang, Sihao Liu, Song Bian and Ligeng Zhu.
Julien Genoud retweeted
Nous Research presents Hermes 4, our latest line of hybrid reasoning models.
hermes4.nousresearch.com
Hermes 4 builds on our legacy of user-aligned models with expanded test-time compute capabilities.
Special attention was given to making the models creative and interesting to interact with, unencumbered by censorship, and neutrally aligned while maintaining state of the art level math, coding, and reasoning performance for open weight models.
it would be awesome to support "Loras" direction, do you think it's doable with ledits ? @linoy_tsaban @multimodalart
Julien Genoud retweeted
I launched this week and made my first sales! 🥳
I cloned @levelsio Photo AI. Targeted customers in Sweden that never got exposed to Photo AI. My goal is $100,000/m⚡️
I spent 100+ hours on cloning it and 8 hours getting the first 5 customers.
Should I create a video and share with everybody how I did it? Comment if you want the video :) Or do you want the github repo?
Julien Genoud retweeted
BIG news. @Google just released official code for RB-Modulation! Want to condition your diffusion model on a subject or style image without using IP-Adapters, then try our method. (works on StableCascade and is adaptable to SDXL and Flux)
github.com/google/RB-Modulat…
that's all !! I used @fal_ai_data to make my first lora (its cost around 4$) and now i'm using my own computer with the nice ai-toolkit from @ostrisai I have a lot of ideas to train different lora styles, stay tuned !!
flux Lora of the day😍
Here's another FLUX finetune that blew me away.
I love the style from Return of the Obra Dinn's by @dukope.
Using @everartai, I trained a model with a few screenshots. Now I can create anything I want in that universe.
It's hard to stop.
Julien Genoud retweeted
We at have @fal just launched AuraSR v2!
AuraSR v1 could not upscale JPG compressed images, but v2 does a great job.
Blog: blog.fal.ai/aurasr-v2/
Weights: huggingface.co/fal/AuraSR-v2
API: fal.ai/models/fal-ai/aura-sr
Julien Genoud retweeted
This is great news, @ollama now supports function calling.
A local alternative of my Claude Engineer is now very possible. Probably updating the script later today.
