SpargeAttn: Accurate Sparse Attention Accelerating Any Model Inference
Paper
•
2502.18137
•
Published
•
59
Sparge-attention model zoo
Welcome to Sparge-attention model zoo, this repo contains list of hyperparameters pre-tuned for branch of models.
It was presented in the paper SpargeAttn: Accurate Sparse Attention Accelerating Any Model Inference.
Naming of ckpt
The tuned ckpt is often named by following format:${moddel name or type}_${l1}_${pv_l1}.pt, in some cases the pv_l1 will be omitted when not choose to tune pv.
The larger l1 and pv_l1 make model more sparse, but may sacrifice output quality.
Overview
| model name | tuned ckpt dir |
|---|---|
| CogVideoX-2b | cogvideox-2b |
| want2v-1.3b | want2v-1.3B |
Per model detail
Inference Providers
NEW
This model isn't deployed by any Inference Provider.
🙋
Ask for provider support