Skip to content

Commit

Permalink
Solve the issue Tencent#122 by updating inference.py
Browse files Browse the repository at this point in the history
In issue Tencent#122 (Tencent#122), every for loop the parallelize_transformer will reset the pipeline causing the problem. If changing the parallelize_transfomer to __init__,it will solve the issue without affecting other functions.
  • Loading branch information
guankaisi authored Dec 16, 2024
1 parent 83ab6ec commit ca2ed2d
Showing 1 changed file with 2 additions and 6 deletions.
8 changes: 2 additions & 6 deletions hyvideo/inference.py
Original file line number Diff line number Diff line change
Expand Up @@ -402,6 +402,8 @@ def __init__(
)

self.default_negative_prompt = NEGATIVE_PROMPT
if self.parallel_args['ulysses_degree'] > 1 or self.parallel_args['ring_degree'] > 1:
parallelize_transformer(self.pipeline)

def load_diffusion_pipeline(
self,
Expand Down Expand Up @@ -521,12 +523,6 @@ def predict(
num_images_per_prompt (int): The number of images per prompt. Default is 1.
infer_steps (int): The number of inference steps. Default is 100.
"""
if self.parallel_args['ulysses_degree'] > 1 or self.parallel_args['ring_degree'] > 1:
assert seed is not None, \
"You have to set a seed in the distributed environment, please rerun with --seed <your-seed>."

parallelize_transformer(self.pipeline)

out_dict = dict()

# ========================================================================
Expand Down

0 comments on commit ca2ed2d

Please sign in to comment.