Fix TURBOQUANT backend selection in cuda.py (#40060)

Signed-off-by: Michael Goin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin
2026-04-17 07:31:41 -07:00
committed by GitHub
parent 6b2b7bd0eb
commit 1174723eba
2 changed files with 4 additions and 5 deletions
+2
View File
@@ -106,6 +106,7 @@ Priority is **1 = highest** (tried first).
| 2 | `FLASH_ATTN` |
| 3 | `TRITON_ATTN` |
| 4 | `FLEX_ATTENTION` |
| 5 | `TURBOQUANT` |
**Ampere/Hopper (SM 8.x-9.x):**
@@ -115,6 +116,7 @@ Priority is **1 = highest** (tried first).
| 2 | `FLASHINFER` |
| 3 | `TRITON_ATTN` |
| 4 | `FLEX_ATTENTION` |
| 5 | `TURBOQUANT` |
### MLA Attention (DeepSeek-style)