378 49 193

Sayak Paul

sayakpaul

https://sayak.dev

AI & ML interests

Diffusion models, representation learning

Recent Activity

commented on a paper about 6 hours ago

Radial Attention: $O(n\log n)$ Sparse Attention with Energy Decay for Long Video Generation

upvoted a paper about 6 hours ago

Radial Attention: O(nlog n) Sparse Attention with Energy Decay for Long Video Generation

reacted to burtenshaw's post with ❤️ about 9 hours ago

Inference for generative ai models looks like a mine field, but there’s a simple protocol for picking the best inference: 🌍 95% of users >> If you’re using open (large) models and need fast online inference, then use Inference providers on auto mode, and let it choose the best provider for the model. https://huggingface.co/docs/inference-providers/index 👷 fine-tuners/ bespoke >> If you’ve got custom setups, use Inference Endpoints to define a configuration from AWS, Azure, GCP. https://endpoints.huggingface.co/ 🦫 Locals >> If you’re trying to stretch everything you can out of a server or local machine, use Llama.cpp, Jan, LMStudio or vLLM. https://huggingface.co/settings/local-apps#local-apps 🪟 Browsers >> If you need open models running right here in the browser, use transformers.js. https://github.com/huggingface/transformers.js Let me know what you’re using, and if you think it’s more complex than this.

View all activity

Organizations

commented a paper about 6 hours ago

Radial Attention: $O(n\log n)$ Sparse Attention with Energy Decay for Long Video Generation

Paper • 2506.19852 • Published 8 days ago • 18 •

upvoted a paper about 6 hours ago

Radial Attention: O(nlog n) Sparse Attention with Energy Decay for Long Video Generation

Paper • 2506.19852 • Published 8 days ago • 18

reacted to burtenshaw's post with ❤️ about 9 hours ago

Post

1091

Inference for generative ai models looks like a mine field, but there’s a simple protocol for picking the best inference:

🌍 95% of users >> If you’re using open (large) models and need fast online inference, then use Inference providers on auto mode, and let it choose the best provider for the model. https://huggingface.co/docs/inference-providers/index

👷 fine-tuners/ bespoke >> If you’ve got custom setups, use Inference Endpoints to define a configuration from AWS, Azure, GCP. https://endpoints.huggingface.co/

🦫 Locals >> If you’re trying to stretch everything you can out of a server or local machine, use Llama.cpp, Jan, LMStudio or vLLM. https://huggingface.co/settings/local-apps#local-apps

🪟 Browsers >> If you need open models running right here in the browser, use transformers.js. https://github.com/huggingface/transformers.js

Let me know what you’re using, and if you think it’s more complex than this.

New activity in black-forest-labs/FLUX.1-Kontext-dev 3 days ago

Fine-tune / Training

#24 opened 3 days ago by

MAK007

updated a dataset 4 days ago

sayakpaul/sample-datasets

Viewer • Updated 4 days ago • 6 • 28.1k • 1

updated 2 Spaces 5 days ago

Optimized Diffusers Code

🔥

Optimize Diffusers Code on your hardware.

Flux Quantized or Original?

🔎

Generate and compare quantized images

published a model 6 days ago

hf-internal-testing/tiny-flux-kontext-pipe

Updated 6 days ago • 1.07k

liked a model 6 days ago

black-forest-labs/FLUX.1-Kontext-dev

Image-to-Image • Updated 5 days ago • 122k • • 1.15k

published a Space 6 days ago

Optimized Diffusers Code

🔥

Optimize Diffusers Code on your hardware.

updated a model 6 days ago

hf-internal-testing/tiny-flux-kontext-pipe

Updated 6 days ago • 1.07k

liked a model 7 days ago

vrgamedevgirl84/Wan14BT2VFusioniX

Text-to-Video • Updated 11 days ago • 345

commented on (LoRA) Fine-Tuning FLUX.1-dev on Consumer Hardware 10 days ago

I would say it should work for Schnell but you might have to change the loss function a bit to account for the fact it's a timestep-distilled model.
Yes.
I don't think LoRA merge should increase the final state dict size.
The numbers shouldn't change much.