diff --git a/tests/engine/test_arg_utils.py b/tests/engine/test_arg_utils.py index 6a799ee912..db64c5bb6f 100644 --- a/tests/engine/test_arg_utils.py +++ b/tests/engine/test_arg_utils.py @@ -383,8 +383,6 @@ def test_attention_config(): "FLASH_ATTN", "--attention-config.flash_attn_version", "3", - "--attention-config.use_prefill_decode_attention", - "true", "--attention-config.flash_attn_max_num_splits_for_cuda_graph", "16", "--attention-config.use_trtllm_attention", @@ -398,7 +396,6 @@ def test_attention_config(): assert engine_args.attention_config.backend is not None assert engine_args.attention_config.backend.name == "FLASH_ATTN" assert engine_args.attention_config.flash_attn_version == 3 - assert engine_args.attention_config.use_prefill_decode_attention is True assert engine_args.attention_config.flash_attn_max_num_splits_for_cuda_graph == 16 assert engine_args.attention_config.use_trtllm_attention is True assert engine_args.attention_config.disable_flashinfer_q_quantization is True @@ -408,7 +405,6 @@ def test_attention_config(): [ "--attention-config=" '{"backend": "FLASHINFER", "flash_attn_version": 2, ' - '"use_prefill_decode_attention": false, ' '"flash_attn_max_num_splits_for_cuda_graph": 8, ' '"use_trtllm_attention": false, ' '"disable_flashinfer_q_quantization": false}', @@ -419,7 +415,6 @@ def test_attention_config(): assert engine_args.attention_config.backend is not None assert engine_args.attention_config.backend.name == "FLASHINFER" assert engine_args.attention_config.flash_attn_version == 2 - assert engine_args.attention_config.use_prefill_decode_attention is False assert engine_args.attention_config.flash_attn_max_num_splits_for_cuda_graph == 8 assert engine_args.attention_config.use_trtllm_attention is False assert engine_args.attention_config.disable_flashinfer_q_quantization is False diff --git a/tests/v1/attention/utils.py b/tests/v1/attention/utils.py index 6aeeb12b4b..ee702bdbdf 100644 --- a/tests/v1/attention/utils.py +++ b/tests/v1/attention/utils.py @@ -387,7 +387,6 @@ full_cg_backend_configs = { name="RocmAttn", attention_config={ "backend": "ROCM_ATTN", - "use_prefill_decode_attention": True, }, comp_config={ "cudagraph_mode": "FULL", diff --git a/vllm/config/attention.py b/vllm/config/attention.py index 907917d091..ea92be32cd 100644 --- a/vllm/config/attention.py +++ b/vllm/config/attention.py @@ -42,10 +42,6 @@ class AttentionConfig: """Force vllm to use a specific flash-attention version (2, 3, or 4). Only valid when using the flash-attention backend.""" - use_prefill_decode_attention: bool = False - """Use separate prefill and decode kernels for attention instead of - the unified triton kernel.""" - flash_attn_max_num_splits_for_cuda_graph: int = 32 """Flash Attention max number splits for cuda graph decode."""