[Bugfix] Make MLA+SWA check the layer's backend, not the model config (#48520)
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
@@ -628,8 +628,8 @@ class Attention(nn.Module, AttentionLayerBase):
|
||||
assert self.attn_type == AttentionType.DECODER
|
||||
quant_mode = get_kv_quant_mode(self.kv_cache_dtype)
|
||||
if self.sliding_window is not None:
|
||||
assert not vllm_config.model_config.use_mla, (
|
||||
"MLA is not supported for slidingwindow"
|
||||
assert not self.attn_backend.is_mla(), (
|
||||
"MLA is not supported for sliding window"
|
||||
)
|
||||
# SW chooses its own block_size, decoupled from the user's
|
||||
# ``--block-size`` (which only constrains primary attention).
|
||||
|
||||
Reference in New Issue
Block a user