[PD] use int32 for kv indices & get num_reserved_decode_tokens from server_args (#7214)
This commit is contained in:
@@ -576,7 +576,6 @@ class SchedulerDisaggregationPrefillMixin:
|
||||
self.req_to_token_pool.req_to_token[req.req_pool_idx, start_idx:end_idx]
|
||||
.cpu()
|
||||
.numpy()
|
||||
.astype(np.int64)
|
||||
)
|
||||
req.start_send_idx = end_idx
|
||||
if last_chunk:
|
||||
|
||||
Reference in New Issue
Block a user