Skip to content

Commit d6ab7c2

Browse files
committed
Generator: Reuse job reference when requeueing (req for async jobs)
1 parent f2b8349 commit d6ab7c2

File tree

1 file changed

+6
-2
lines changed

1 file changed

+6
-2
lines changed

exllamav3/generator/job.py

Lines changed: 6 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -800,7 +800,11 @@ def prepare_for_requeue(self):
800800
"rq_new_tokens": self.new_tokens - 1
801801
}
802802

803-
rq_job = Job(
803+
serial_number = self.serial_number
804+
generator = self.generator
805+
806+
rq_job = self
807+
self.__init__(
804808
input_ids = new_input,
805809
max_new_tokens = self.max_new_tokens - last_completed_tokens,
806810
min_new_tokens = max(self.min_new_tokens - last_completed_tokens, 0),
@@ -818,7 +822,7 @@ def prepare_for_requeue(self):
818822
rq_state = rq_state,
819823
)
820824

821-
rq_job.prepare_for_queue(self.generator, self.serial_number, rq = True)
825+
rq_job.prepare_for_queue(generator, serial_number, rq = True)
822826
return rq_job
823827

824828

0 commit comments

Comments
 (0)