[ROCm] Remove unnecessary assertion of max_model_len in ROCM_AITER_MLA attention backend. (#18938)

Signed-off-by: vllmellm <vllm.ellm@embeddedllm.com>

vllm/attention/backends/rocm_aiter_mla.py

@@ -132,8 +132,6 @@ class AiterMLAMetadataBuilder(MLACommonMetadataBuilder[AiterMLAMetadata]):
 
     def __init__(self, input_builder: "ModelInputForGPUBuilder"):
         super().__init__(input_builder)
-        assert self.runner.model_config.max_model_len == 32768,\
-                "AITER MLA requires max model len to be set to 32768"
         assert self.block_size == 1, "AITER MLA requires only block size 1."
 
     def prepare(self):

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

@@ -66,9 +66,6 @@ class AiterMLAMetadataBuilder(MLACommonMetadataBuilder[AiterMLAMetadata]):
     def __init__(self, runner, kv_cache_spec: AttentionSpec,
                  block_table: BlockTable):
         super().__init__(runner, kv_cache_spec, block_table)
-        max_model_len = self.runner.model_config.max_model_len
-        assert max_model_len == 32768,\
-            "AITER MLA requires max_model_len=32768"
         assert self.kv_cache_spec.block_size == 1, "AITER MLA" \
             "only supports block size 1."