[Deprecation] Remove dead use_prefill_decode_attention flag (#52557)
Signed-off-by: Brian Woods <me@brian.biz> Co-authored-by: Claude Opus 5 (1M context) <noreply@anthropic.com> Co-authored-by: mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
This commit is contained in:
co-authored by
Claude Opus 5
mergify[bot] <37929162+mergify[bot]@users.noreply.github.com>
parent
2f55ef254c
commit
bbe8b23e1a
@@ -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
|
||||
|
||||
@@ -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",
|
||||
|
||||
@@ -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."""
|
||||
|
||||
|
||||
Reference in New Issue
Block a user