Prefer FlashAttention MLA as default over FlashMLA #27363

MatthewBonanni · 2025-10-22T19:01:54Z

Purpose

Based on benchmarking in #26835, FlashAttention MLA is faster than FlashMLA across head counts, batch sizes, and query lengths on Hopper. This PR sets it as the preferred backend.

Based on #24794, merge that first

Note: Results shown below apply the fix of #27368, FlashMLA performance is worse prior to the bugfix

Test Plan

Test Result

Model Parameter Sweep Results:

num_q_heads = 16
              Attention Benchmark Results               
┏━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━┓
┃ Batch      ┃    famla ┃   famla ┃     fmla ┃    fmla ┃
┃ Spec       ┃ Time (s) ┃ vs Best ┃ Time (s) ┃ vs Best ┃
┡━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━┩
│ 128q128s1k │ 0.001261 │  100.0% │ 0.001912 │  151.7% │
│ 128q16s1k  │ 0.000186 │  100.0% │ 0.000245 │  132.2% │
│ 128q1s1k   │ 0.000071 │  100.0% │ 0.000084 │  118.3% │
│ 128q256s1k │ 0.002368 │  100.0% │ 0.003813 │  161.0% │
│ 128q2s1k   │ 0.000073 │  100.0% │ 0.000085 │  115.9% │
│ 128q32s1k  │ 0.000351 │  100.0% │ 0.000470 │  133.6% │
│ 128q4s1k   │ 0.000075 │  100.0% │ 0.000088 │  116.8% │
│ 128q64s1k  │ 0.000675 │  100.0% │ 0.000924 │  136.9% │
│ 128q8s1k   │ 0.000106 │  100.0% │ 0.000132 │  125.2% │
│ 16q128s1k  │ 0.000175 │  100.0% │ 0.000245 │  140.2% │
│ 16q16s1k   │ 0.000053 │  100.0% │ 0.000093 │  175.1% │
│ 16q1s1k    │ 0.000030 │  100.0% │ 0.000046 │  151.3% │
│ 16q256s1k  │ 0.000319 │  100.0% │ 0.000478 │  150.0% │
│ 16q2s1k    │ 0.000031 │  100.0% │ 0.000047 │  153.8% │
│ 16q32s1k   │ 0.000053 │  100.0% │ 0.000066 │  123.6% │
│ 16q4s1k    │ 0.000031 │  100.0% │ 0.000053 │  170.0% │
│ 16q64s1k   │ 0.000096 │  100.0% │ 0.000125 │  130.0% │
│ 16q8s1k    │ 0.000040 │  100.0% │ 0.000056 │  141.9% │
│ 1q128s1k   │ 0.000037 │  100.0% │ 0.000057 │  154.4% │
│ 1q16s1k    │ 0.000031 │  100.0% │ 0.000050 │  158.4% │
│ 1q1s1k     │ 0.000031 │  100.0% │ 0.000042 │  135.2% │
│ 1q256s1k   │ 0.000049 │  100.0% │ 0.000065 │  133.2% │
│ 1q2s1k     │ 0.000030 │  100.0% │ 0.000043 │  146.6% │
│ 1q32s1k    │ 0.000032 │  100.0% │ 0.000048 │  151.9% │
│ 1q4s1k     │ 0.000031 │  100.0% │ 0.000046 │  148.5% │
│ 1q512s1k   │ 0.000047 │  100.0% │ 0.000064 │  136.7% │
│ 1q64s1k    │ 0.000032 │  100.0% │ 0.000053 │  164.7% │
│ 1q8s1k     │ 0.000033 │  100.0% │ 0.000048 │  147.3% │
│ 2q128s1k   │ 0.000050 │  100.0% │ 0.000088 │  177.3% │
│ 2q16s1k    │ 0.000031 │  100.0% │ 0.000047 │  152.2% │
│ 2q1s1k     │ 0.000030 │  100.0% │ 0.000046 │  151.2% │
│ 2q256s1k   │ 0.000049 │  100.0% │ 0.000065 │  131.7% │
│ 2q2s1k     │ 0.000030 │  100.0% │ 0.000047 │  156.6% │
│ 2q32s1k    │ 0.000031 │  100.0% │ 0.000053 │  170.1% │
│ 2q4s1k     │ 0.000030 │  100.0% │ 0.000049 │  163.8% │
│ 2q512s1k   │ 0.000083 │  100.0% │ 0.000125 │  150.4% │
│ 2q64s1k    │ 0.000037 │  100.0% │ 0.000057 │  153.1% │
│ 2q8s1k     │ 0.000030 │  100.0% │ 0.000050 │  164.2% │
│ 32q128s1k  │ 0.000337 │  100.0% │ 0.000480 │  142.5% │
│ 32q16s1k   │ 0.000055 │  100.0% │ 0.000066 │  120.1% │
│ 32q1s1k    │ 0.000032 │  100.0% │ 0.000051 │  162.2% │
│ 32q256s1k  │ 0.000616 │  100.0% │ 0.000954 │  155.0% │
│ 32q2s1k    │ 0.000034 │  100.0% │ 0.000055 │  162.1% │
│ 32q32s1k   │ 0.000097 │  100.0% │ 0.000127 │  130.4% │
│ 32q4s1k    │ 0.000042 │  100.0% │ 0.000060 │  143.8% │
│ 32q64s1k   │ 0.000181 │  100.0% │ 0.000240 │  133.2% │
│ 32q8s1k    │ 0.000055 │  100.0% │ 0.000094 │  170.9% │
│ 4q128s1k   │ 0.000049 │  100.0% │ 0.000065 │  131.1% │
│ 4q16s1k    │ 0.000031 │  100.0% │ 0.000052 │  168.1% │
│ 4q1s1k     │ 0.000030 │  100.0% │ 0.000045 │  153.3% │
│ 4q256s1k   │ 0.000092 │  100.0% │ 0.000126 │  136.5% │
│ 4q2s1k     │ 0.000030 │  100.0% │ 0.000047 │  158.5% │
│ 4q32s1k    │ 0.000037 │  100.0% │ 0.000055 │  148.1% │
│ 4q4s1k     │ 0.000030 │  100.0% │ 0.000049 │  161.8% │
│ 4q512s1k   │ 0.000146 │  100.0% │ 0.000247 │  169.2% │
│ 4q64s1k    │ 0.000052 │  100.0% │ 0.000091 │  177.0% │
│ 4q8s1k     │ 0.000031 │  100.0% │ 0.000047 │  154.8% │
│ 64q128s1k  │ 0.000655 │  100.0% │ 0.000956 │  146.0% │
│ 64q16s1k   │ 0.000099 │  100.0% │ 0.000127 │  129.4% │
│ 64q1s1k    │ 0.000047 │  100.0% │ 0.000087 │  183.4% │
│ 64q256s1k  │ 0.001197 │  100.0% │ 0.001906 │  159.2% │
│ 64q2s1k    │ 0.000053 │  100.0% │ 0.000090 │  170.9% │
│ 64q32s1k   │ 0.000182 │  100.0% │ 0.000244 │  134.2% │
│ 64q4s1k    │ 0.000063 │  100.0% │ 0.000100 │  158.1% │
│ 64q64s1k   │ 0.000349 │  100.0% │ 0.000470 │  134.8% │
│ 64q8s1k    │ 0.000058 │  100.0% │ 0.000070 │  120.4% │
│ 8q128s1k   │ 0.000094 │  100.0% │ 0.000127 │  134.3% │
│ 8q16s1k    │ 0.000039 │  100.0% │ 0.000054 │  140.3% │
│ 8q1s1k     │ 0.000030 │  100.0% │ 0.000043 │  142.6% │
│ 8q256s1k   │ 0.000169 │  100.0% │ 0.000245 │  145.1% │
│ 8q2s1k     │ 0.000030 │  100.0% │ 0.000045 │  147.9% │
│ 8q32s1k    │ 0.000052 │  100.0% │ 0.000092 │  178.2% │
│ 8q4s1k     │ 0.000031 │  100.0% │ 0.000047 │  153.6% │
│ 8q512s1k   │ 0.000261 │  100.0% │ 0.000474 │  181.5% │
│ 8q64s1k    │ 0.000051 │  100.0% │ 0.000065 │  126.9% │
│ 8q8s1k     │ 0.000031 │  100.0% │ 0.000051 │  165.2% │
└────────────┴──────────┴─────────┴──────────┴─────────┘

num_q_heads = 32
              Attention Benchmark Results               
┏━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━┓
┃ Batch      ┃    famla ┃   famla ┃     fmla ┃    fmla ┃
┃ Spec       ┃ Time (s) ┃ vs Best ┃ Time (s) ┃ vs Best ┃
┡━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━┩
│ 128q128s1k │ 0.002494 │  100.0% │ 0.003787 │  151.8% │
│ 128q16s1k  │ 0.000351 │  100.0% │ 0.000478 │  136.0% │
│ 128q1s1k   │ 0.000075 │  100.0% │ 0.000085 │  113.2% │
│ 128q256s1k │ 0.004156 │  100.0% │ 0.007579 │  182.4% │
│ 128q2s1k   │ 0.000080 │  100.0% │ 0.000087 │  108.3% │
│ 128q32s1k  │ 0.000686 │  100.0% │ 0.000942 │  137.2% │
│ 128q4s1k   │ 0.000106 │  100.0% │ 0.000132 │  124.1% │
│ 128q64s1k  │ 0.001294 │  100.0% │ 0.001901 │  146.9% │
│ 128q8s1k   │ 0.000186 │  100.0% │ 0.000248 │  133.8% │
│ 16q128s1k  │ 0.000334 │  100.0% │ 0.000481 │  144.1% │
│ 16q16s1k   │ 0.000053 │  100.0% │ 0.000066 │  124.6% │
│ 16q1s1k    │ 0.000031 │  100.0% │ 0.000047 │  153.3% │
│ 16q256s1k  │ 0.000552 │  100.0% │ 0.000950 │  172.2% │
│ 16q2s1k    │ 0.000031 │  100.0% │ 0.000052 │  170.2% │
│ 16q32s1k   │ 0.000099 │  100.0% │ 0.000127 │  128.9% │
│ 16q4s1k    │ 0.000039 │  100.0% │ 0.000056 │  145.2% │
│ 16q64s1k   │ 0.000182 │  100.0% │ 0.000245 │  134.6% │
│ 16q8s1k    │ 0.000053 │  100.0% │ 0.000092 │  172.7% │
│ 1q128s1k   │ 0.000051 │  100.0% │ 0.000066 │  128.8% │
│ 1q16s1k    │ 0.000031 │  100.0% │ 0.000048 │  152.4% │
│ 1q1s1k     │ 0.000031 │  100.0% │ 0.000043 │  138.2% │
│ 1q256s1k   │ 0.000048 │  100.0% │ 0.000065 │  135.0% │
│ 1q2s1k     │ 0.000031 │  100.0% │ 0.000046 │  147.6% │
│ 1q32s1k    │ 0.000032 │  100.0% │ 0.000053 │  162.5% │
│ 1q4s1k     │ 0.000032 │  100.0% │ 0.000048 │  148.9% │
│ 1q512s1k   │ 0.000079 │  100.0% │ 0.000120 │  152.8% │
│ 1q64s1k    │ 0.000037 │  100.0% │ 0.000057 │  152.0% │
│ 1q8s1k     │ 0.000031 │  100.0% │ 0.000048 │  152.3% │
│ 2q128s1k   │ 0.000051 │  100.0% │ 0.000065 │  128.7% │
│ 2q16s1k    │ 0.000031 │  100.0% │ 0.000053 │  170.4% │
│ 2q1s1k     │ 0.000034 │  100.0% │ 0.000046 │  137.8% │
│ 2q256s1k   │ 0.000086 │  100.0% │ 0.000126 │  147.9% │
│ 2q2s1k     │ 0.000030 │  100.0% │ 0.000049 │  161.3% │
│ 2q32s1k    │ 0.000037 │  100.0% │ 0.000057 │  154.0% │
│ 2q4s1k     │ 0.000030 │  100.0% │ 0.000050 │  164.0% │
│ 2q512s1k   │ 0.000144 │  100.0% │ 0.000233 │  161.5% │
│ 2q64s1k    │ 0.000051 │  100.0% │ 0.000089 │  173.5% │
│ 2q8s1k     │ 0.000031 │  100.0% │ 0.000046 │  150.3% │
│ 32q128s1k  │ 0.000652 │  100.0% │ 0.000958 │  146.8% │
│ 32q16s1k   │ 0.000098 │  100.0% │ 0.000127 │  128.9% │
│ 32q1s1k    │ 0.000035 │  100.0% │ 0.000055 │  159.9% │
│ 32q256s1k  │ 0.001065 │  100.0% │ 0.001909 │  179.3% │
│ 32q2s1k    │ 0.000042 │  100.0% │ 0.000059 │  139.4% │
│ 32q32s1k   │ 0.000183 │  100.0% │ 0.000246 │  134.8% │
│ 32q4s1k    │ 0.000054 │  100.0% │ 0.000094 │  173.8% │
│ 32q64s1k   │ 0.000348 │  100.0% │ 0.000477 │  136.9% │
│ 32q8s1k    │ 0.000055 │  100.0% │ 0.000066 │  120.2% │
│ 4q128s1k   │ 0.000095 │  100.0% │ 0.000126 │  132.9% │
│ 4q16s1k    │ 0.000038 │  100.0% │ 0.000056 │  149.1% │
│ 4q1s1k     │ 0.000030 │  100.0% │ 0.000047 │  156.4% │
│ 4q256s1k   │ 0.000156 │  100.0% │ 0.000248 │  159.3% │
│ 4q2s1k     │ 0.000030 │  100.0% │ 0.000049 │  162.6% │
│ 4q32s1k    │ 0.000052 │  100.0% │ 0.000091 │  176.5% │
│ 4q4s1k     │ 0.000031 │  100.0% │ 0.000046 │  149.5% │
│ 4q512s1k   │ 0.000260 │  100.0% │ 0.000458 │  176.4% │
│ 4q64s1k    │ 0.000052 │  100.0% │ 0.000065 │  126.5% │
│ 4q8s1k     │ 0.000031 │  100.0% │ 0.000052 │  168.6% │
│ 64q128s1k  │ 0.001257 │  100.0% │ 0.001910 │  152.0% │
│ 64q16s1k   │ 0.000183 │  100.0% │ 0.000249 │  135.8% │
│ 64q1s1k    │ 0.000052 │  100.0% │ 0.000089 │  170.2% │
│ 64q256s1k  │ 0.002093 │  100.0% │ 0.003779 │  180.5% │
│ 64q2s1k    │ 0.000065 │  100.0% │ 0.000097 │  150.9% │
│ 64q32s1k   │ 0.000351 │  100.0% │ 0.000478 │  136.0% │
│ 64q4s1k    │ 0.000058 │  100.0% │ 0.000070 │  121.6% │
│ 64q64s1k   │ 0.000678 │  100.0% │ 0.000937 │  138.3% │
│ 64q8s1k    │ 0.000099 │  100.0% │ 0.000127 │  128.2% │
│ 8q128s1k   │ 0.000177 │  100.0% │ 0.000250 │  141.3% │
│ 8q16s1k    │ 0.000052 │  100.0% │ 0.000092 │  176.3% │
│ 8q1s1k     │ 0.000030 │  100.0% │ 0.000044 │  147.6% │
│ 8q256s1k   │ 0.000288 │  100.0% │ 0.000486 │  168.5% │
│ 8q2s1k     │ 0.000031 │  100.0% │ 0.000047 │  150.8% │
│ 8q32s1k    │ 0.000067 │  101.0% │ 0.000066 │  100.0% │
│ 8q4s1k     │ 0.000031 │  100.0% │ 0.000051 │  165.8% │
│ 8q512s1k   │ 0.000491 │  100.0% │ 0.000900 │  183.2% │
│ 8q64s1k    │ 0.000098 │  100.0% │ 0.000126 │  129.5% │
│ 8q8s1k     │ 0.000039 │  100.0% │ 0.000054 │  140.6% │
└────────────┴──────────┴─────────┴──────────┴─────────┘

num_q_heads = 64
              Attention Benchmark Results               
┏━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━┓
┃ Batch      ┃    famla ┃   famla ┃     fmla ┃    fmla ┃
┃ Spec       ┃ Time (s) ┃ vs Best ┃ Time (s) ┃ vs Best ┃
┡━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━┩
│ 128q128s1k │ 0.004345 │  100.0% │ 0.007506 │  172.7% │
│ 128q16s1k  │ 0.000686 │  100.0% │ 0.000950 │  138.4% │
│ 128q1s1k   │ 0.000082 │  100.0% │ 0.000087 │  105.3% │
│ 128q256s1k │ 0.008165 │  100.0% │ 0.014819 │  181.5% │
│ 128q2s1k   │ 0.000114 │  100.0% │ 0.000131 │  114.6% │
│ 128q32s1k  │ 0.001306 │  100.0% │ 0.001896 │  145.1% │
│ 128q4s1k   │ 0.000187 │  100.0% │ 0.000248 │  132.5% │
│ 128q64s1k  │ 0.002533 │  100.0% │ 0.003774 │  149.0% │
│ 128q8s1k   │ 0.000352 │  100.0% │ 0.000483 │  137.2% │
│ 16q128s1k  │ 0.000585 │  100.0% │ 0.000961 │  164.2% │
│ 16q16s1k   │ 0.000100 │  100.0% │ 0.000129 │  129.1% │
│ 16q1s1k    │ 0.000031 │  100.0% │ 0.000052 │  167.6% │
│ 16q256s1k  │ 0.001061 │  100.0% │ 0.001848 │  174.2% │
│ 16q2s1k    │ 0.000040 │  100.0% │ 0.000056 │  141.0% │
│ 16q32s1k   │ 0.000185 │  100.0% │ 0.000247 │  133.9% │
│ 16q4s1k    │ 0.000053 │  100.0% │ 0.000092 │  172.7% │
│ 16q64s1k   │ 0.000350 │  100.0% │ 0.000483 │  137.8% │
│ 16q8s1k    │ 0.000053 │  100.0% │ 0.000066 │  124.3% │
│ 1q128s1k   │ 0.000049 │  100.0% │ 0.000066 │  134.5% │
│ 1q16s1k    │ 0.000032 │  100.0% │ 0.000052 │  162.2% │
│ 1q1s1k     │ 0.000032 │  100.0% │ 0.000045 │  141.6% │
│ 1q256s1k   │ 0.000083 │  100.0% │ 0.000126 │  151.1% │
│ 1q2s1k     │ 0.000031 │  100.0% │ 0.000048 │  153.2% │
│ 1q32s1k    │ 0.000037 │  100.0% │ 0.000057 │  153.3% │
│ 1q4s1k     │ 0.000032 │  100.0% │ 0.000048 │  150.2% │
│ 1q512s1k   │ 0.000143 │  100.0% │ 0.000236 │  165.2% │
│ 1q64s1k    │ 0.000052 │  100.0% │ 0.000065 │  126.7% │
│ 1q8s1k     │ 0.000032 │  100.0% │ 0.000046 │  141.5% │
│ 2q128s1k   │ 0.000088 │  100.0% │ 0.000128 │  145.0% │
│ 2q16s1k    │ 0.000037 │  100.0% │ 0.000057 │  154.5% │
│ 2q1s1k     │ 0.000030 │  100.0% │ 0.000048 │  161.9% │
│ 2q256s1k   │ 0.000157 │  100.0% │ 0.000244 │  155.2% │
│ 2q2s1k     │ 0.000030 │  100.0% │ 0.000050 │  165.0% │
│ 2q32s1k    │ 0.000051 │  100.0% │ 0.000091 │  177.4% │
│ 2q4s1k     │ 0.000031 │  100.0% │ 0.000046 │  151.1% │
│ 2q512s1k   │ 0.000262 │  100.0% │ 0.000452 │  172.6% │
│ 2q64s1k    │ 0.000052 │  100.0% │ 0.000065 │  125.7% │
│ 2q8s1k     │ 0.000031 │  100.0% │ 0.000052 │  167.6% │
│ 32q128s1k  │ 0.001120 │  100.0% │ 0.001892 │  169.0% │
│ 32q16s1k   │ 0.000185 │  100.0% │ 0.000246 │  133.1% │
│ 32q1s1k    │ 0.000042 │  100.0% │ 0.000059 │  139.5% │
│ 32q256s1k  │ 0.002086 │  100.0% │ 0.003681 │  176.4% │
│ 32q2s1k    │ 0.000056 │  100.0% │ 0.000092 │  165.0% │
│ 32q32s1k   │ 0.000351 │  100.0% │ 0.000486 │  138.7% │
│ 32q4s1k    │ 0.000054 │  100.0% │ 0.000066 │  121.9% │
│ 32q64s1k   │ 0.000675 │  100.0% │ 0.000957 │  141.7% │
│ 32q8s1k    │ 0.000099 │  100.0% │ 0.000128 │  129.0% │
│ 4q128s1k   │ 0.000163 │  100.0% │ 0.000250 │  153.4% │
│ 4q16s1k    │ 0.000052 │  100.0% │ 0.000093 │  178.5% │
│ 4q1s1k     │ 0.000030 │  100.0% │ 0.000048 │  160.3% │
│ 4q256s1k   │ 0.000288 │  100.0% │ 0.000475 │  164.7% │
│ 4q2s1k     │ 0.000031 │  100.0% │ 0.000046 │  150.0% │
│ 4q32s1k    │ 0.000052 │  100.0% │ 0.000065 │  126.1% │
│ 4q4s1k     │ 0.000031 │  100.0% │ 0.000051 │  161.9% │
│ 4q512s1k   │ 0.000491 │  100.0% │ 0.000890 │  181.1% │
│ 4q64s1k    │ 0.000099 │  100.0% │ 0.000129 │  129.6% │
│ 4q8s1k     │ 0.000038 │  100.0% │ 0.000056 │  148.5% │
│ 64q128s1k  │ 0.002193 │  100.0% │ 0.003764 │  171.7% │
│ 64q16s1k   │ 0.000354 │  100.0% │ 0.000486 │  137.1% │
│ 64q1s1k    │ 0.000063 │  100.0% │ 0.000097 │  155.4% │
│ 64q256s1k  │ 0.004130 │  100.0% │ 0.007392 │  179.0% │
│ 64q2s1k    │ 0.000061 │  100.0% │ 0.000069 │  114.0% │
│ 64q32s1k   │ 0.000685 │  100.0% │ 0.000947 │  138.3% │
│ 64q4s1k    │ 0.000099 │  100.0% │ 0.000128 │  128.3% │
│ 64q64s1k   │ 0.001290 │  100.0% │ 0.001912 │  148.3% │
│ 64q8s1k    │ 0.000184 │  100.0% │ 0.000249 │  135.1% │
│ 8q128s1k   │ 0.000304 │  100.0% │ 0.000492 │  161.9% │
│ 8q16s1k    │ 0.000052 │  100.0% │ 0.000066 │  126.6% │
│ 8q1s1k     │ 0.000031 │  100.0% │ 0.000046 │  150.7% │
│ 8q256s1k   │ 0.000551 │  100.0% │ 0.000930 │  168.9% │
│ 8q2s1k     │ 0.000031 │  100.0% │ 0.000051 │  165.5% │
│ 8q32s1k    │ 0.000099 │  100.0% │ 0.000129 │  130.0% │
│ 8q4s1k     │ 0.000037 │  100.0% │ 0.000054 │  145.5% │
│ 8q512s1k   │ 0.000949 │  100.0% │ 0.001751 │  184.6% │
│ 8q64s1k    │ 0.000185 │  100.0% │ 0.000246 │  133.5% │
│ 8q8s1k     │ 0.000052 │  100.0% │ 0.000092 │  176.5% │
└────────────┴──────────┴─────────┴──────────┴─────────┘

num_q_heads = 128
              Attention Benchmark Results               
┏━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━┓
┃ Batch      ┃    famla ┃   famla ┃     fmla ┃    fmla ┃
┃ Spec       ┃ Time (s) ┃ vs Best ┃ Time (s) ┃ vs Best ┃
┡━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━┩
│ 128q128s1k │ 0.008551 │  100.0% │ 0.015035 │  175.8% │
│ 128q16s1k  │ 0.001307 │  100.0% │ 0.001907 │  145.9% │
│ 128q1s1k   │ 0.000113 │  100.0% │ 0.000129 │  114.7% │
│ 128q256s1k │ 0.016206 │  100.0% │ 0.029083 │  179.5% │
│ 128q2s1k   │ 0.000203 │  100.0% │ 0.000248 │  122.2% │
│ 128q32s1k  │ 0.002556 │  100.0% │ 0.003769 │  147.5% │
│ 128q4s1k   │ 0.000355 │  100.0% │ 0.000482 │  135.9% │
│ 128q64s1k  │ 0.004442 │  100.0% │ 0.007509 │  169.0% │
│ 128q8s1k   │ 0.000686 │  100.0% │ 0.000946 │  138.0% │
│ 16q128s1k  │ 0.001122 │  100.0% │ 0.001904 │  169.8% │
│ 16q16s1k   │ 0.000188 │  100.0% │ 0.000249 │  132.6% │
│ 16q1s1k    │ 0.000036 │  100.0% │ 0.000055 │  153.9% │
│ 16q256s1k  │ 0.002079 │  100.0% │ 0.003655 │  175.8% │
│ 16q2s1k    │ 0.000053 │  100.0% │ 0.000092 │  174.0% │
│ 16q32s1k   │ 0.000356 │  100.0% │ 0.000488 │  137.0% │
│ 16q4s1k    │ 0.000053 │  100.0% │ 0.000066 │  123.6% │
│ 16q64s1k   │ 0.000604 │  100.0% │ 0.000960 │  158.8% │
│ 16q8s1k    │ 0.000099 │  100.0% │ 0.000128 │  129.2% │
│ 1q128s1k   │ 0.000088 │  100.0% │ 0.000131 │  148.2% │
│ 1q16s1k    │ 0.000037 │  100.0% │ 0.000057 │  152.9% │
│ 1q1s1k     │ 0.000031 │  100.0% │ 0.000046 │  150.8% │
│ 1q256s1k   │ 0.000160 │  100.0% │ 0.000252 │  157.5% │
│ 1q2s1k     │ 0.000032 │  100.0% │ 0.000048 │  151.0% │
│ 1q32s1k    │ 0.000052 │  100.0% │ 0.000066 │  127.5% │
│ 1q4s1k     │ 0.000033 │  100.0% │ 0.000046 │  140.8% │
│ 1q512s1k   │ 0.000267 │  100.0% │ 0.000465 │  173.8% │
│ 1q64s1k    │ 0.000049 │  100.0% │ 0.000066 │  134.1% │
│ 1q8s1k     │ 0.000033 │  100.0% │ 0.000050 │  151.1% │
│ 2q128s1k   │ 0.000166 │  100.0% │ 0.000255 │  153.6% │
│ 2q16s1k    │ 0.000051 │  100.0% │ 0.000091 │  177.3% │
│ 2q1s1k     │ 0.000030 │  100.0% │ 0.000049 │  165.3% │
│ 2q256s1k   │ 0.000294 │  100.0% │ 0.000487 │  165.4% │
│ 2q2s1k     │ 0.000030 │  100.0% │ 0.000046 │  152.2% │
│ 2q32s1k    │ 0.000052 │  100.0% │ 0.000067 │  128.7% │
│ 2q4s1k     │ 0.000031 │  100.0% │ 0.000052 │  166.8% │
│ 2q512s1k   │ 0.000497 │  100.0% │ 0.000893 │  179.8% │
│ 2q64s1k    │ 0.000089 │  100.0% │ 0.000128 │  143.9% │
│ 2q8s1k     │ 0.000037 │  100.0% │ 0.000056 │  150.3% │
│ 32q128s1k  │ 0.002194 │  100.0% │ 0.003770 │  171.9% │
│ 32q16s1k   │ 0.000358 │  100.0% │ 0.000487 │  136.0% │
│ 32q1s1k    │ 0.000050 │  100.0% │ 0.000092 │  182.8% │
│ 32q256s1k  │ 0.004116 │  100.0% │ 0.007261 │  176.4% │
│ 32q2s1k    │ 0.000056 │  100.0% │ 0.000065 │  114.7% │
│ 32q32s1k   │ 0.000680 │  100.0% │ 0.000966 │  142.1% │
│ 32q4s1k    │ 0.000097 │  100.0% │ 0.000126 │  130.9% │
│ 32q64s1k   │ 0.001144 │  100.0% │ 0.001916 │  167.5% │
│ 32q8s1k    │ 0.000186 │  100.0% │ 0.000249 │  133.4% │
│ 4q128s1k   │ 0.000310 │  100.0% │ 0.000493 │  158.9% │
│ 4q16s1k    │ 0.000052 │  100.0% │ 0.000066 │  127.8% │
│ 4q1s1k     │ 0.000031 │  100.0% │ 0.000045 │  149.0% │
│ 4q256s1k   │ 0.000556 │  100.0% │ 0.000958 │  172.2% │
│ 4q2s1k     │ 0.000031 │  100.0% │ 0.000051 │  163.4% │
│ 4q32s1k    │ 0.000099 │  100.0% │ 0.000128 │  129.6% │
│ 4q4s1k     │ 0.000036 │  100.0% │ 0.000054 │  149.6% │
│ 4q512s1k   │ 0.000950 │  100.0% │ 0.001756 │  184.9% │
│ 4q64s1k    │ 0.000171 │  100.0% │ 0.000251 │  146.9% │
│ 4q8s1k     │ 0.000052 │  100.0% │ 0.000093 │  178.2% │
│ 64q128s1k  │ 0.004298 │  100.0% │ 0.007497 │  174.4% │
│ 64q16s1k   │ 0.000689 │  100.0% │ 0.000959 │  139.2% │
│ 64q1s1k    │ 0.000059 │  100.0% │ 0.000069 │  116.6% │
│ 64q256s1k  │ 0.008185 │  100.0% │ 0.014529 │  177.5% │
│ 64q2s1k    │ 0.000103 │  100.0% │ 0.000124 │  120.6% │
│ 64q32s1k   │ 0.001301 │  100.0% │ 0.001910 │  146.8% │
│ 64q4s1k    │ 0.000185 │  100.0% │ 0.000248 │  134.2% │
│ 64q64s1k   │ 0.002246 │  100.0% │ 0.003775 │  168.1% │
│ 64q8s1k    │ 0.000355 │  100.0% │ 0.000485 │  136.8% │
│ 8q128s1k   │ 0.000588 │  100.0% │ 0.000973 │  165.4% │
│ 8q16s1k    │ 0.000100 │  100.0% │ 0.000129 │  128.7% │
│ 8q1s1k     │ 0.000031 │  100.0% │ 0.000050 │  163.0% │
│ 8q256s1k   │ 0.001060 │  100.0% │ 0.001879 │  177.2% │
│ 8q2s1k     │ 0.000038 │  100.0% │ 0.000054 │  142.2% │
│ 8q32s1k    │ 0.000187 │  100.0% │ 0.000251 │  133.9% │
│ 8q4s1k     │ 0.000050 │  100.0% │ 0.000092 │  183.2% │
│ 8q512s1k   │ 0.001862 │  100.0% │ 0.003422 │  183.8% │
│ 8q64s1k    │ 0.000319 │  100.0% │ 0.000488 │  153.1% │
│ 8q8s1k     │ 0.000052 │  100.0% │ 0.000066 │  126.7% │
└────────────┴──────────┴─────────┴──────────┴─────────┘

num_q_heads = 256
              Attention Benchmark Results               
┏━━━━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━┳━━━━━━━━━━┳━━━━━━━━━┓
┃ Batch      ┃    famla ┃   famla ┃     fmla ┃    fmla ┃
┃ Spec       ┃ Time (s) ┃ vs Best ┃ Time (s) ┃ vs Best ┃
┡━━━━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━╇━━━━━━━━━━╇━━━━━━━━━┩
│ 128q128s1k │ 0.017040 │  100.0% │ 0.029658 │  174.0% │
│ 128q16s1k  │ 0.002558 │  100.0% │ 0.003774 │  147.6% │
│ 128q1s1k   │ 0.000198 │  100.0% │ 0.000242 │  122.2% │
│ 128q256s1k │ 0.032271 │  100.0% │ 0.057412 │  177.9% │
│ 128q2s1k   │ 0.000391 │  100.0% │ 0.000483 │  123.5% │
│ 128q32s1k  │ 0.004430 │  100.0% │ 0.007510 │  169.6% │
│ 128q4s1k   │ 0.000690 │  100.0% │ 0.000947 │  137.2% │
│ 128q64s1k  │ 0.008824 │  100.0% │ 0.014886 │  168.7% │
│ 128q8s1k   │ 0.001306 │  100.0% │ 0.001902 │  145.6% │
│ 16q128s1k  │ 0.002194 │  100.0% │ 0.003728 │  169.9% │
│ 16q16s1k   │ 0.000362 │  100.0% │ 0.000491 │  135.6% │
│ 16q1s1k    │ 0.000047 │  100.0% │ 0.000089 │  190.9% │
│ 16q256s1k  │ 0.004116 │  100.0% │ 0.007171 │  174.2% │
│ 16q2s1k    │ 0.000054 │  100.0% │ 0.000066 │  122.7% │
│ 16q32s1k   │ 0.000608 │  100.0% │ 0.000967 │  159.2% │
│ 16q4s1k    │ 0.000099 │  100.0% │ 0.000128 │  128.3% │
│ 16q64s1k   │ 0.001153 │  100.0% │ 0.001883 │  163.3% │
│ 16q8s1k    │ 0.000188 │  100.0% │ 0.000251 │  133.5% │
│ 1q128s1k   │ 0.000170 │  100.0% │ 0.000260 │  152.9% │
│ 1q16s1k    │ 0.000052 │  100.0% │ 0.000066 │  127.4% │
│ 1q1s1k     │ 0.000032 │  100.0% │ 0.000046 │  143.6% │
│ 1q256s1k   │ 0.000299 │  100.0% │ 0.000490 │  163.8% │
│ 1q2s1k     │ 0.000032 │  100.0% │ 0.000046 │  142.3% │
│ 1q32s1k    │ 0.000049 │  100.0% │ 0.000066 │  134.4% │
│ 1q4s1k     │ 0.000032 │  100.0% │ 0.000051 │  157.7% │
│ 1q512s1k   │ 0.000510 │  100.0% │ 0.000905 │  177.4% │
│ 1q64s1k    │ 0.000090 │  100.0% │ 0.000130 │  144.2% │
│ 1q8s1k     │ 0.000037 │  100.0% │ 0.000057 │  151.4% │
│ 2q128s1k   │ 0.000313 │  100.0% │ 0.000498 │  159.2% │
│ 2q16s1k    │ 0.000052 │  100.0% │ 0.000067 │  128.6% │
│ 2q1s1k     │ 0.000030 │  100.0% │ 0.000045 │  148.7% │
│ 2q256s1k   │ 0.000562 │  100.0% │ 0.000934 │  166.2% │
│ 2q2s1k     │ 0.000031 │  100.0% │ 0.000052 │  168.0% │
│ 2q32s1k    │ 0.000089 │  100.0% │ 0.000131 │  147.3% │
│ 2q4s1k     │ 0.000036 │  100.0% │ 0.000056 │  155.0% │
│ 2q512s1k   │ 0.000957 │  100.0% │ 0.001722 │  179.8% │
│ 2q64s1k    │ 0.000171 │  100.0% │ 0.000253 │  148.3% │
│ 2q8s1k     │ 0.000051 │  100.0% │ 0.000089 │  173.7% │
│ 32q128s1k  │ 0.004335 │  100.0% │ 0.007442 │  171.7% │
│ 32q16s1k   │ 0.000686 │  100.0% │ 0.000967 │  141.0% │
│ 32q1s1k    │ 0.000055 │  100.0% │ 0.000065 │  117.9% │
│ 32q256s1k  │ 0.008182 │  100.0% │ 0.014302 │  174.8% │
│ 32q2s1k    │ 0.000101 │  100.0% │ 0.000125 │  124.2% │
│ 32q32s1k   │ 0.001150 │  100.0% │ 0.001911 │  166.3% │
│ 32q4s1k    │ 0.000186 │  100.0% │ 0.000246 │  132.7% │
│ 32q64s1k   │ 0.002247 │  100.0% │ 0.003726 │  165.8% │
│ 32q8s1k    │ 0.000359 │  100.0% │ 0.000490 │  136.6% │
│ 4q128s1k   │ 0.000591 │  100.0% │ 0.000961 │  162.7% │
│ 4q16s1k    │ 0.000099 │  100.0% │ 0.000130 │  130.8% │
│ 4q1s1k     │ 0.000031 │  100.0% │ 0.000050 │  163.1% │
│ 4q256s1k   │ 0.001068 │  100.0% │ 0.001855 │  173.7% │
│ 4q2s1k     │ 0.000037 │  100.0% │ 0.000054 │  145.8% │
│ 4q32s1k    │ 0.000171 │  100.0% │ 0.000251 │  147.0% │
│ 4q4s1k     │ 0.000050 │  100.0% │ 0.000090 │  180.6% │
│ 4q512s1k   │ 0.001863 │  100.0% │ 0.003406 │  182.9% │
│ 4q64s1k    │ 0.000325 │  100.0% │ 0.000492 │  151.2% │
│ 4q8s1k     │ 0.000052 │  100.0% │ 0.000066 │  127.9% │
│ 64q128s1k  │ 0.008611 │  100.0% │ 0.014794 │  171.8% │
│ 64q16s1k   │ 0.001301 │  100.0% │ 0.001911 │  146.9% │
│ 64q1s1k    │ 0.000101 │  100.0% │ 0.000124 │  123.5% │
│ 64q256s1k  │ 0.016176 │  100.0% │ 0.028746 │  177.7% │
│ 64q2s1k    │ 0.000192 │  100.0% │ 0.000247 │  128.6% │
│ 64q32s1k   │ 0.002251 │  100.0% │ 0.003769 │  167.4% │
│ 64q4s1k    │ 0.000358 │  100.0% │ 0.000487 │  136.3% │
│ 64q64s1k   │ 0.004441 │  100.0% │ 0.007430 │  167.3% │
│ 64q8s1k    │ 0.000690 │  100.0% │ 0.000960 │  139.2% │
│ 8q128s1k   │ 0.001117 │  100.0% │ 0.001883 │  168.5% │
│ 8q16s1k    │ 0.000189 │  100.0% │ 0.000251 │  133.0% │
│ 8q1s1k     │ 0.000034 │  100.0% │ 0.000053 │  156.2% │
│ 8q256s1k   │ 0.002083 │  100.0% │ 0.003610 │  173.3% │
│ 8q2s1k     │ 0.000052 │  100.0% │ 0.000092 │  178.8% │
│ 8q32s1k    │ 0.000324 │  100.0% │ 0.000490 │  151.4% │
│ 8q4s1k     │ 0.000051 │  100.0% │ 0.000066 │  130.1% │
│ 8q512s1k   │ 0.003753 │  100.0% │ 0.006626 │  176.5% │
│ 8q64s1k    │ 0.000606 │  100.0% │ 0.000950 │  156.6% │
│ 8q8s1k     │ 0.000100 │  100.0% │ 0.000128 │  128.7% │
└────────────┴──────────┴─────────┴──────────┴─────────┘

Essential Elements of an Effective PR Description Checklist

The purpose of the PR, such as "Fix some issue (link existing issues this PR will resolve)".
The test plan, such as providing test command.
The test results, such as pasting the results comparison before and after, or e2e results
(Optional) The necessary documentation update, such as updating supported_models.md and examples for a new model.
(Optional) Release notes update. If your change is user facing, please update the release notes draft in the Google Doc.

chatgpt-codex-connector

💡 Codex Review

vllm/vllm/config/multimodal.py

Lines 160 to 176 in c8a0a46

    
           @field_validator("mm_encoder_attn_backend", mode="before") 
        
           @classmethod 
        
           def _validate_mm_encoder_attn_backend(cls, value: object) -> _Backend | None: 
        
               from vllm.attention.backends.registry import ( 
        
                   _Backend as BackendEnum, 
        
               ) 
        
               from vllm.attention.backends.registry import ( 
        
                   backend_name_to_enum, 
        
               ) 
        
               if value is None or isinstance(value, BackendEnum): 
        
                   return value 
        
               if isinstance(value, str): 
        
                   candidate = backend_name_to_enum(value.upper()) 
        
                   if candidate is not None: 
        
                       return candidate

Restore backend_name_to_enum used by multimodal config

The commit removes backend_name_to_enum from the attention backend registry, but MultiModalConfig._validate_mm_encoder_attn_backend still imports and calls it. Importing vllm.config.multimodal now raises ImportError: cannot import name 'backend_name_to_enum', preventing multimodal configurations from even loading. Either reintroduce the helper or update the validator to use _Backend[...] directly.

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

vllm/attention/selector.py

gemini-code-assist

Code Review

This pull request introduces a significant and well-executed refactoring of the attention backend selection logic. The new implementation is more structured, robust, and maintainable by centralizing configuration validation within the backend classes themselves and using a priority-based selection mechanism. This also correctly implements the preference for FlashAttention MLA over FlashMLA on Hopper GPUs as intended. The changes are extensive and touch many parts of the attention subsystem, but they represent a clear improvement in design. I've identified a critical issue in the implementation for FlashAttentionBackend and FlashAttnMLABackend that could lead to a runtime crash, and have provided suggestions for a fix.

vllm/v1/attention/backends/flash_attn.py

vllm/v1/attention/backends/mla/flashattn_mla.py

mergify · 2025-10-26T11:45:24Z

This pull request has merge conflicts that must be resolved before it can be
merged. Please rebase the PR, @MatthewBonanni.

https://docs.github.com/en/pull-requests/collaborating-with-pull-requests/working-with-forks/syncing-a-fork

MatthewBonanni · 2025-11-03T14:41:35Z

Related issue: #27787

Signed-off-by: Matthew Bonanni <[email protected]>

LucasWilkinson

Thanks!

mgoin · 2025-11-11T16:56:36Z

vllm/platforms/cuda.py

-                AttentionBackendEnum.FLASHMLA,
                AttentionBackendEnum.FLASH_ATTN_MLA,
+                AttentionBackendEnum.FLASHMLA,


Just with respect to Blackwell, do we actually know how flashmla and fa mla work on there? I'm curious if fa mla is even a valid option

I actually marked FA MLA as supporting only CC 9.x, so it'll get skipped on Blackwell regardless of its spot in the priority list. It could probably just be removed from the list, but I'll test it out anyway to check. FlashMLA supports Blackwell but I haven't benchmarked it

Signed-off-by: Matthew Bonanni <[email protected]>

Signed-off-by: Matthew Bonanni <[email protected]> Signed-off-by: Xingyu Liu <[email protected]>

mergify bot added rocm Related to AMD ROCm speculative-decoding v1 tpu Related to Google TPUs kv-connector labels Oct 22, 2025

chatgpt-codex-connector bot reviewed Oct 22, 2025

View reviewed changes

vllm/attention/selector.py Outdated Show resolved Hide resolved

gemini-code-assist bot reviewed Oct 22, 2025

View reviewed changes

vllm/v1/attention/backends/flash_attn.py Outdated Show resolved Hide resolved

vllm/v1/attention/backends/mla/flashattn_mla.py Outdated Show resolved Hide resolved

MatthewBonanni force-pushed the prefer_fa_mla branch from c8a0a46 to db96fd6 Compare October 22, 2025 19:20

Kay-Tian mentioned this pull request Oct 23, 2025

vLLM PR #27363 变更核心文件提醒 Kay-Tian/vllm#2

Closed

mergify bot added the needs-rebase label Oct 26, 2025

MatthewBonanni mentioned this pull request Nov 4, 2025

Add attention benchmarking tools #26835

Open

5 tasks

reorder priority list

a30f40e

Signed-off-by: Matthew Bonanni <[email protected]>

MatthewBonanni force-pushed the prefer_fa_mla branch from db96fd6 to a30f40e Compare November 11, 2025 14:47

mergify bot added nvidia and removed tpu Related to Google TPUs labels Nov 11, 2025

github-project-automation bot added this to NVIDIA Nov 11, 2025

mergify bot removed the needs-rebase label Nov 11, 2025

LucasWilkinson enabled auto-merge (squash) November 11, 2025 15:10

github-actions bot added the ready ONLY add when PR is ready to merge/full CI is needed label Nov 11, 2025

LucasWilkinson approved these changes Nov 11, 2025

View reviewed changes

github-project-automation bot moved this to Ready in NVIDIA Nov 11, 2025

mgoin approved these changes Nov 11, 2025

View reviewed changes

github-project-automation bot moved this from Ready to In review in NVIDIA Nov 11, 2025

LucasWilkinson merged commit 684f254 into vllm-project:main Nov 11, 2025
49 of 50 checks passed

github-project-automation bot moved this from In review to Done in NVIDIA Nov 11, 2025

MatthewBonanni deleted the prefer_fa_mla branch November 11, 2025 18:00

fangyuchu pushed a commit to fangyuchu/vllm that referenced this pull request Nov 12, 2025

Prefer FlashAttention MLA as default over FlashMLA (vllm-project#27363)

4079ef2

Signed-off-by: Matthew Bonanni <[email protected]>

LucasWilkinson mentioned this pull request Nov 14, 2025

[BugFix] Fix AssertionError: DCP not support reorder_batch_threshold > 1 now. #28751

Merged

devpatelio pushed a commit to SumanthRH/vllm that referenced this pull request Nov 29, 2025

Prefer FlashAttention MLA as default over FlashMLA (vllm-project#27363)

4189063

Signed-off-by: Matthew Bonanni <[email protected]>

charlotte12l pushed a commit to charlotte12l/vllm that referenced this pull request Dec 5, 2025

Prefer FlashAttention MLA as default over FlashMLA (vllm-project#27363)

f92c400

Signed-off-by: Matthew Bonanni <[email protected]> Signed-off-by: Xingyu Liu <[email protected]>

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Uh oh!

Prefer FlashAttention MLA as default over FlashMLA #27363

Prefer FlashAttention MLA as default over FlashMLA #27363

Uh oh!

MatthewBonanni commented Oct 22, 2025 •

edited by github-actions bot

Loading

Uh oh!

chatgpt-codex-connector bot left a comment

Uh oh!

Uh oh!

gemini-code-assist bot left a comment

Uh oh!

Uh oh!

Uh oh!

mergify bot commented Oct 26, 2025

Uh oh!

MatthewBonanni commented Nov 3, 2025

Uh oh!

LucasWilkinson left a comment

Uh oh!

mgoin Nov 11, 2025

Uh oh!

MatthewBonanni Nov 11, 2025

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

	@field_validator("mm_encoder_attn_backend", mode="before")
	@classmethod
	def _validate_mm_encoder_attn_backend(cls, value: object) -> _Backend \| None:
	from vllm.attention.backends.registry import (
	_Backend as BackendEnum,
	)
	from vllm.attention.backends.registry import (
	backend_name_to_enum,
	)

	if value is None or isinstance(value, BackendEnum):
	return value

	if isinstance(value, str):
	candidate = backend_name_to_enum(value.upper())
	if candidate is not None:
	return candidate

Uh oh!

Prefer FlashAttention MLA as default over FlashMLA #27363

Prefer FlashAttention MLA as default over FlashMLA #27363

Uh oh!

Conversation

MatthewBonanni commented Oct 22, 2025 • edited by github-actions bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Purpose

Test Plan

Test Result

Uh oh!

chatgpt-codex-connector bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

Uh oh!

gemini-code-assist bot left a comment

Choose a reason for hiding this comment

Code Review

Uh oh!

Uh oh!

Uh oh!

mergify bot commented Oct 26, 2025

Uh oh!

MatthewBonanni commented Nov 3, 2025

Uh oh!

LucasWilkinson left a comment

Choose a reason for hiding this comment

Uh oh!

mgoin Nov 11, 2025

Choose a reason for hiding this comment

Uh oh!

MatthewBonanni Nov 11, 2025

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

MatthewBonanni commented Oct 22, 2025 •

edited by github-actions bot

Loading