Enable memory saver for hybrid model (#11974)
This commit is contained in:
@@ -42,6 +42,7 @@ class TestMamba(unittest.TestCase):
|
||||
full_attention_layer_ids=full_attention_layer_ids,
|
||||
enable_kvcache_transpose=False,
|
||||
device=device,
|
||||
enable_memory_saver=False,
|
||||
mamba_pool=None,
|
||||
)
|
||||
assert pool._transfer_full_attention_id(global_interval - 1) == 0
|
||||
@@ -174,6 +175,7 @@ class TestMamba(unittest.TestCase):
|
||||
full_attention_layer_ids=full_attention_layer_ids,
|
||||
enable_kvcache_transpose=False,
|
||||
device=device,
|
||||
enable_memory_saver=False,
|
||||
mamba_pool=req_to_token_pool.mamba_pool,
|
||||
)
|
||||
|
||||
|
||||
Reference in New Issue
Block a user