[deepseek] replace attn (a workaround for bug in transformers)

2025-09-01 01:06:00 +00:00 · 2024-07-23 12:56:58 +00:00
parent 6c39f0b144
commit c3dc9b4dba
2 changed files with 30 additions and 5 deletions
--- a/tests/test_shardformer/test_model/test_shard_deepseek_ghz.py
+++ b/tests/test_shardformer/test_model/test_shard_deepseek_ghz.py
@@ -220,6 +220,7 @@ def check_deepseek(rank, world_size, port):
    run_deepseek_test()


+@pytest.mark.skip("redundant")
@pytest.mark.dist
@rerun_if_address_is_in_use()
@clear_cache_before_run()