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 15, 07:15 UTC
[ROCm] Remove unnecessary assertion of max_model_len in ROCM_AITER_MLA attention backend. (#18938)
Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com>
vllmellm committed
May 30, 2025 at 13:33 UTC
77b6e74fe2b6259b2ad4dbca55eca50017a68926
2 files changed
-5
vllm/attention/backends/rocm_aiter_mla.py
-2
@@ -132,8 +132,6 @@ class AiterMLAMetadataBuilder(MLACommonMetadataBuilder[AiterMLAMetadata]):
132
133
def __init__(self, input_builder: "ModelInputForGPUBuilder"):
134
super().__init__(input_builder)
135
- assert self.runner.model_config.max_model_len == 32768,\
136
- "AITER MLA requires max model len to be set to 32768"
135
assert self.block_size == 1, "AITER MLA requires only block size 1."
136
137
def prepare(self):
vllm/v1/attention/backends/mla/rocm_aiter_mla.py
-3
@@ -66,9 +66,6 @@ class AiterMLAMetadataBuilder(MLACommonMetadataBuilder[AiterMLAMetadata]):
66
def __init__(self, runner, kv_cache_spec: AttentionSpec,
67
block_table: BlockTable):
68
super().__init__(runner, kv_cache_spec, block_table)
69
- max_model_len = self.runner.model_config.max_model_len
70
- assert max_model_len == 32768,\
71
- "AITER MLA requires max_model_len=32768"
69
assert self.kv_cache_spec.block_size == 1, "AITER MLA" \
70
"only supports block size 1."
71