fix: wire benchmark cache selection
This commit is contained in:
@@ -68,6 +68,15 @@ class GenerationBenchmark:
|
||||
self.cuda_graph = cuda_graph
|
||||
|
||||
def _make_pool(self, batch_size: int, max_seq_len: int) -> PagePool:
|
||||
if self.cache_type == "contiguous":
|
||||
n_tokens = None
|
||||
elif self.cache_type == "paged":
|
||||
# Keep the total token capacity equal to contiguous mode while
|
||||
# routing allocation through the shared paged pool.
|
||||
n_tokens = batch_size * max_seq_len
|
||||
else:
|
||||
raise ValueError(f"unsupported cache type: {self.cache_type}")
|
||||
|
||||
return PagePool(
|
||||
n_layers=self.config.num_hidden_layers,
|
||||
n_kv_heads=self.config.num_key_value_heads,
|
||||
@@ -77,7 +86,7 @@ class GenerationBenchmark:
|
||||
device=self.device,
|
||||
dtype=self.dtype,
|
||||
page_size=1,
|
||||
n_tokens=None,
|
||||
n_tokens=n_tokens,
|
||||
)
|
||||
|
||||
@staticmethod
|
||||
|
||||
Reference in New Issue
Block a user