A runner and the objects it builds freeze the placement they describe (#40341)
This commit is contained in:
+2
-1
@@ -91,7 +91,8 @@ def _make_prefill_aware_swa_runner(
|
||||
kv_cache_dtype=torch.float16,
|
||||
kv_cache_dtype_str="auto",
|
||||
page_size=1,
|
||||
ps=SimpleNamespace(attn_cp_size=1, tp_size=1),
|
||||
attn_cp_size=1,
|
||||
tp_size=1,
|
||||
is_draft_worker=False,
|
||||
server_args=server_args,
|
||||
attention_chunk_size=None,
|
||||
|
||||
Reference in New Issue
Block a user