Pinned Tweet
Had the honor to present diffusion transformers at CS25, Stanford. The place is truly magical.
Slides: bit.ly/dit-cs25
Recording: youtu.be/vXtapCFctTI?si=dlcE…
Thanks to @stevenyfeng for making it happen!
Sayak Paul retweeted
Meet Fibo 2.0 from @bria_ai_ 🔥
The #1 *open weights* model in image editing is supported in Diffusers (day-0, of course)!
1. JSON-native control
2. Pretty good text rendering
3. Comes in two variants: Turbo and Dev
Check out the model here:
huggingface.co/docs/diffuser…
We just merged the first protein sequence design model RFD4-Proteina in Diffusers 🔥
I hope this pushes the state of open BioML further.
Kudos to @NVIDIAAI and Baker Lab for shipping this!
Check out the docs to know more:
huggingface.co/docs/diffuser…
Cheers to @DidiKieran for leading the PR integration and bearing with our feedback. It's been a real treat!
Meet Fibo 2.0 from @bria_ai_ 🔥
The #1 *open weights* model in image editing is supported in Diffusers (day-0, of course)!
1. JSON-native control
2. Pretty good text rendering
3. Comes in two variants: Turbo and Dev
Check out the model here:
huggingface.co/docs/diffuser…
Some extended comments.
My main purpose for the trip was my talk on Kernels at the PyTorch Conference.
Reaching the US by a day's delay (which wouldn't anyway allow me to do the talk) only to leave after 4 days would have been too much.
So, I thought it's better I stay home and support the upcoming releases. Might as well enjoy the rest of the Durga Puja while it lasts.
Thanks to @qatarairways my US trip now stands canceled.
Somehow the Kolkata flight got delayed and that led them to choose an option for me which would land in SF at 1:15 PM 21st Oct (delayed by a day).
So, it forced me to ultimately cancel it.
Absolutely gutted for this. But maybe this is what Maa Durga wanted as I was leaving her amidst the festivities.
Thanks to @qatarairways my US trip now stands canceled.
Somehow the Kolkata flight got delayed and that led them to choose an option for me which would land in SF at 1:15 PM 21st Oct (delayed by a day).
So, it forced me to ultimately cancel it.
Absolutely gutted for this. But maybe this is what Maa Durga wanted as I was leaving her amidst the festivities.
Sayak Paul retweeted
It's time. It's fucking time!
Flux.3 Image Klein is out and we bring you full support for it in Diffusers.
Catch all the details in our post:
huggingface.co/blog/flux-3
It's time. It's fucking time!
Flux.3 Image Klein is out and we bring you full support for it in Diffusers.
Catch all the details in our post:
huggingface.co/blog/flux-3
Both the distilled and non-distilled pipelines are implemented as modular diffusers pipelines for maximum flexibility!
We have also worked hard to support loading the quantized checkpoints (FP8r and NVFP4) and performing inference with them.
We have even given them a backend of their own!
A release is incomplete without a LoRA training script.
We're providing Diffusers-native LoRA training scripts for both T2I and I2I. Thanks to @linoy_tsaban for tackling this!
We also support loading LoRAs trained with @ostrisai's ai-toolkit!
Enjoy :)
I am giving a talk next week at the PyTorch Conference on the 21st. It's about the 🤗 Kernels project. If you're attending, please come and say hi :)
Leaving home in the middle of Durga Puja isn't fun, but I will be in SF next week. Happy to meet up 🤗
Sayak Paul retweeted
Was meant to be a Valentine's gift, but I'm releasing it early ❤️
Wrote a long piece on KV caching in flow models for image generation.
Includes intuition, commentary, implementation details, benchmarks, & some quirky experiments.
sayak.dev/posts/kv-flow.html
Was meant to be a Valentine's gift, but I'm releasing it early ❤️
Wrote a long piece on KV caching in flow models for image generation.
Includes intuition, commentary, implementation details, benchmarks, & some quirky experiments.
sayak.dev/posts/kv-flow.html
Then I checked what happens when we combine a more traditional caching technique in diffusion (like TaylorSeer) with KV caching.
That seems to lead to some noticeable quality degradation.
If you read till here, thanks!
I have also disclosed how much I used AI while writing the piece for full transparency.
Here's a link to the post:
sayak.dev/posts/kv-flow.html