tp1 working multistream tp > 1 broken

Signed-off-by: Lucas Wilkinson <lwilkins@redhat.com>
This commit is contained in:
Lucas Wilkinson
2025-05-21 19:41:11 +00:00
parent 2259b47951
commit 9c60a6299d
5 changed files with 168 additions and 122 deletions

View File

@ -40,7 +40,7 @@ def main():
max_model_len=1024,
#load_format="dummy",
###############
tensor_parallel_size=2,
tensor_parallel_size=1,
#data_parallel_size=2,
enable_expert_parallel=False,
###############