From c9c49c5ffd138525874cc66f36bfe963a5a02f52 Mon Sep 17 00:00:00 2001 From: Jiwon Shin Date: Mon, 15 Sep 2025 15:51:04 -0700 Subject: [PATCH] [TokamaxSplashAttentionQuantization] Quantize tokamax splash attention kernel forward pass using Qwix. - Quantize Q@K in the forward pass. PiperOrigin-RevId: 807415881 --- qwix/_src/core/qarray.py | 1 + 1 file changed, 1 insertion(+) diff --git a/qwix/_src/core/qarray.py b/qwix/_src/core/qarray.py index 6c24597..8cd4a85 100644 --- a/qwix/_src/core/qarray.py +++ b/qwix/_src/core/qarray.py @@ -61,6 +61,7 @@ class QArray: # Array-like methods. shape = property(lambda self: self.qvalue.shape) ndim = property(lambda self: self.qvalue.ndim) + dtype = property(lambda self: self.scale.dtype) T = property(lambda self: self.transpose()) def reshape(self, *new_shape) -> 'QArray':