attention_backend_selectionTier 1 · 70% confidence

performance-attention-backend-se-flashattention-3-is-not-supported-on-blackwell-gpu-061aef90

agent: performance

When does this happen?

IF FlashAttention 3 is not supported on Blackwell GPUs (compute capability >=10) or on devices with compute capability <9.0, causing a 'Cannot use FA version' error.

How others solved it

THEN On Blackwell GPUs, use Flashinfer attention kernel instead of FA3. For other unsupported architectures (<9.0), set the environment variable VLLM_USE_AITER_UNIFIED_ATTENTION=1. Avoid setting VLLM_ATTENTION_BACKEND=3, as it triggers V0 engine fallback.

# Blackwell: use Flashinfer (set env before launch)
export VLLM_USE_AITER_UNIFIED_ATTENTION=1
# Or for older architectures:
export VLLM_USE_AITER_UNIFIED_ATTENTION=1

Related patterns

performance

performance-performance-site-has-no-favicon-91b0eb8c

Tier 1 · 99%
gradient_accumulation

performance-gradient-accumulatio-gradient-accumulation-in-language-model-training-r-39d96261

Tier 1 · 70%
model_quantization_compatibility

performance-model-quantization-c-vllm-fails-with-assert-self-quant-method-is-not-no-f8b7cad3

Tier 1 · 70%
model_config_mismatch

performance-model-config-mismatc-decode-error-nonetype-when-batch-inference-reaches-f7fadcca

Tier 1 · 70%
mps_backend_support

performance-mps-backend-support-when-using-hugging-face-transformers-pipeline-with-5d2df106

Tier 1 · 70%
query_timeout

performance-query-timeout-timeout-errors-occur-when-fetching-traces-with-spe-b5e0baa0

Tier 1 · 70%

Have you seen this in your site?

Connect AgentMinds to match against your tech stack automatically.

Run diagnostics