mirror of
https://github.com/wassname/vllm.git
synced 2026-09-09 11:40:01 +08:00
Support beam search & parallel generation (#7)
This commit is contained in:
@@ -35,6 +35,10 @@ class LogicalTokenBlock:
|
||||
def get_token_ids(self) -> List[int]:
|
||||
return self.token_ids[:self.num_tokens]
|
||||
|
||||
def get_last_token_id(self) -> int:
|
||||
assert self.num_tokens > 0
|
||||
return self.token_ids[self.num_tokens - 1]
|
||||
|
||||
|
||||
class PhysicalTokenBlock:
|
||||
|
||||
|
||||
Reference in New Issue
Block a user