huggingface/text-generation-inference
5
0
Fork 0
mirror of https://github.com/huggingface/text-generation-inference.git synced 2025-04-29 22:12:07 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix: only keep stop sequence buffer if we have some

Browse Source
This commit is contained in:
OlivierDehaene 2023-12-14 17:04:58 +01:00
parent 80a69204c1
commit 9b78a6eee3
1 changed files with 10 additions and 9 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

3
server/text_generation_server/utils/tokens.py
View File

@ -112,7 +112,7 @@ class StoppingCriteria:
self.stop_sequence_criterias = stop_sequence_criterias self.stop_sequence_criterias = stop_sequence_criterias
self.max_new_tokens = max_new_tokens self.max_new_tokens = max_new_tokens
self.current_tokens = 0 self.current_tokens = 0
self.current_output = "test" self.current_output = ""
self.ignore_eos_token = ignore_eos_token self.ignore_eos_token = ignore_eos_token
def __call__(self, last_token: int, last_output: str) -> Tuple[bool, Optional[str]]: def __call__(self, last_token: int, last_output: str) -> Tuple[bool, Optional[str]]:
@ -123,6 +123,7 @@ class StoppingCriteria:
if not self.ignore_eos_token and last_token == self.eos_token_id: if not self.ignore_eos_token and last_token == self.eos_token_id:
return True, FinishReason.FINISH_REASON_EOS_TOKEN return True, FinishReason.FINISH_REASON_EOS_TOKEN
if self.stop_sequence_criterias:
self.current_output += last_output self.current_output += last_output
# There is no need to keep an output that is too long # There is no need to keep an output that is too long
if len(self.current_output) > 300: if len(self.current_output) > 300: