423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-18 12:50:38 +08:00
Code Issues Packages Projects Releases Wiki Activity

[test] add npu test yaml and add ascend a3 docker file (#9547)

Browse Source 
Co-authored-by: jiaqiw09 <jiaqiw960714@gmail.com>
This commit is contained in:
Username_Full
2025-11-30 09:37:08 +08:00
committed by GitHub
parent 22be45c78c
commit e43a972b25
33 changed files with 322 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
tests/model/model_utils/test_checkpointing.py
View File

@@ -39,6 +39,7 @@ TRAIN_ARGS = {
}
@pytest.mark.runs_on(["cpu","npu"])
@pytest.mark.parametrize("disable_gradient_checkpointing", [False, True])
def test_vanilla_checkpointing(disable_gradient_checkpointing: bool):
model = load_train_model(disable_gradient_checkpointing=disable_gradient_checkpointing, **TRAIN_ARGS)
@@ -46,12 +47,14 @@ def test_vanilla_checkpointing(disable_gradient_checkpointing: bool):
assert getattr(module, "gradient_checkpointing") != disable_gradient_checkpointing
@pytest.mark.runs_on(["cpu","npu"])
def test_unsloth_gradient_checkpointing():
model = load_train_model(use_unsloth_gc=True, **TRAIN_ARGS)
for module in filter(lambda m: hasattr(m, "gradient_checkpointing"), model.modules()):
assert module._gradient_checkpointing_func.__self__.__name__ == "UnslothGradientCheckpointing"
@pytest.mark.runs_on(["cpu","npu"])
def test_upcast_layernorm():
model = load_train_model(upcast_layernorm=True, **TRAIN_ARGS)
for name, param in model.named_parameters():
@@ -59,6 +62,7 @@ def test_upcast_layernorm():
assert param.dtype == torch.float32
@pytest.mark.runs_on(["cpu","npu"])
def test_upcast_lmhead_output():
model = load_train_model(upcast_lmhead_output=True, **TRAIN_ARGS)
inputs = torch.randn((1, 16), dtype=torch.float16, device=get_current_device())