A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 9, 06:15 UTC
[ROCM] Enable aiter attn backend for qwen3-next model (#32492)
Signed-off-by: jennyyyyzhen <yzhen@hmc.edu>
jennyyyyzhen committed
Jan 31, 2026 at 01:03 UTC
527bcd14d46595b2bde9e48f3d2f10879623b891
2 files changed
+2
-2
docs/design/attention_backends.md
+1
-1
@@ -168,7 +168,7 @@ Priority is **1 = highest** (tried first).
168
| `FLASH_ATTN` | FA3* | fp16, bf16 | `auto`, `bfloat16`, `fp8`, `fp8_e4m3`, `fp8_e5m2` | %16 | Any | ✅ | ❌ | All | 9.x |
169
| `FLASH_ATTN_DIFFKV` | | fp16, bf16 | `auto` | Any | Any | ❌ | ❌ | Decoder | Any |
170
| `FLEX_ATTENTION` | | fp16, bf16, fp32 | `auto`, `bfloat16` | Any | Any | ❌ | ✅ | Decoder, Encoder Only | Any |
171
-| `ROCM_AITER_FA` | | fp16, bf16 | `auto` | %16 | 64, 128, 256 | ❌ | ❌ | Decoder | N/A |
171
+| `ROCM_AITER_FA` | | fp16, bf16 | `auto` | 16, 32 | 64, 128, 256 | ❌ | ❌ | Decoder | N/A |
172
| `ROCM_AITER_UNIFIED_ATTN` | | fp16, bf16 | `auto` | Any | Any | ❌ | ❌ | Decoder | N/A |
173
| `ROCM_ATTN` | | fp16, bf16, fp32 | `auto` | 16, 32, 544 | 32, 64, 96, 128, 160, 192, 224, 256 | ❌ | ❌ | Decoder | N/A |
174
| `TREE_ATTN` | | fp16, bf16 | `auto` | %16 | 32, 64, 96, 128, 160, 192, 224, 256 | ❌ | ❌ | Decoder | Any |
vllm/v1/attention/backends/rocm_aiter_fa.py
+1
-1
@@ -683,7 +683,7 @@ class AiterFlashAttentionBackend(AttentionBackend):
683
684
@staticmethod
685
def get_supported_kernel_block_sizes() -> list[int | MultipleOf]:
686
- return [MultipleOf(16)]
686
+ return [16, 32]
687
688
@classmethod
689
def get_supported_head_sizes(cls) -> list[int]: