memory efficient 1/2

Gasoonjia · Gasoonjia · commit ccf1900ae31e · 2025-12-03T14:59:54.000-08:00
diff --git a/backends/cuda/runtime/shims/sdpa.h b/backends/cuda/runtime/shims/sdpa.h
@@ -90,7 +90,6 @@ aoti_torch_cuda__scaled_dot_product_flash_attention(
  *
  * @return AOTITorchError error code
  *
- * Note: Currently implemented using flash attention backend as a fallback.
  */
 AOTI_SHIM_EXPORT AOTITorchError
 aoti_torch_cuda__scaled_dot_product_efficient_attention(

Original file line number	Diff line number	Diff line change
`@@ -90,7 +90,6 @@ aoti_torch_cuda__scaled_dot_product_flash_attention(`
`90`	`90`	`*`
`91`	`91`	`* @return AOTITorchError error code`
`92`	`92`	`*`
`93`		`- * Note: Currently implemented using flash attention backend as a fallback.`
`94`	`93`	`*/`
`95`	`94`	`AOTI_SHIM_EXPORT AOTITorchError`
`96`	`95`	`aoti_torch_cuda__scaled_dot_product_efficient_attention(`