Submissions from hanlab.mit.edu

		How attention sinks keep language models stable (hanlab.mit.edu)
		219 points by pr337h4m 29 days ago \| past \| 36 comments
		Radial Attention: O(nlogn) Attention for Long Video Generation with 2-4× Speedup (hanlab.mit.edu)
		6 points by lmxyy 61 days ago \| past \| 1 comment
		SVDQuant+NVFP4: 4× Smaller, 3× Faster FLUX with 16-bit Quality on Blackwell GPUs (hanlab.mit.edu)
		52 points by lmxyy 6 months ago \| past \| 10 comments
		RTX 5090 Workstation Configuration Journey (hanlab.mit.edu)
		5 points by lmxyy 6 months ago \| past \| 1 comment
		SVDQuant: 4-Bit Quantization Powers 12B Flux on a 16GB 4090 GPU with 3x Speedup (hanlab.mit.edu)
		179 points by lmxyy 10 months ago \| past \| 65 comments
		HART: Efficient Visual Generation with Hybrid Autoregressive Transformer (hanlab.mit.edu)
		2 points by lnyan 10 months ago \| past
		TinyChat: Large Language Model on the Edge (hanlab.mit.edu)
		2 points by enduku on Dec 8, 2023 \| past \| 1 comment