From 3a542e8f1b8fdc6d338c68e8521209bc9205b011 Mon Sep 17 00:00:00 2001 From: shcherbakov22 Date: Thu, 1 Oct 2026 08:44:16 +0200 Subject: [PATCH 1/2] Add attribution for FlashPrefill v2 Added credit for research on FlashPrefill v2 and linked paper. --- README.md | 3 +++ 1 file changed, 3 insertions(+) diff --git a/README.md b/README.md index 48fcd021..f713a8bc 100644 --- a/README.md +++ b/README.md @@ -56,6 +56,9 @@ At 16K, the merged build reached **632.1 prompt tok/s** versus 492.9 dense. At 32K, the earlier matched pair reached **604.9 prompt tok/s** versus 378.1 dense. The 64-token greedy output matched; perplexity has not been measured. [Configuration and measurements](docs/experimental/sparse-attention.md). +Credit to [shcherbakov22](https://github.com/shcherbakov22/) for providing research on FlashPrefill v2 +[FlashPrefill v2 paper](https://arxiv.org/html/2608.19758v1) + ## Kernel cache The CLI and HTTP server create `~/.lse/cache/` automatically and reuse compiled From 1cd15eb4eaf2110c6ee470f7466be3a81326ea62 Mon Sep 17 00:00:00 2001 From: shcherbakov22 Date: Thu, 1 Oct 2026 08:44:48 +0200 Subject: [PATCH 2/2] Minor formatting fix --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index f713a8bc..bee04703 100644 --- a/README.md +++ b/README.md @@ -56,7 +56,7 @@ At 16K, the merged build reached **632.1 prompt tok/s** versus 492.9 dense. At 32K, the earlier matched pair reached **604.9 prompt tok/s** versus 378.1 dense. The 64-token greedy output matched; perplexity has not been measured. [Configuration and measurements](docs/experimental/sparse-attention.md). -Credit to [shcherbakov22](https://github.com/shcherbakov22/) for providing research on FlashPrefill v2 +Credit to [shcherbakov22](https://github.com/shcherbakov22/) for providing research on FlashPrefill v2 [FlashPrefill v2 paper](https://arxiv.org/html/2608.19758v1) ## Kernel cache