423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-16 20:00:36 +08:00
Code Issues Packages Projects Releases Wiki Activity

[v1] support automatic discovery of registered kernels. (#9509)

Browse Source 
Co-authored-by: frozenleaves <frozen@Mac.local>
This commit is contained in:
浮梦
2025-11-27 01:47:22 +08:00
committed by GitHub
parent 591fc9ed02
commit f17efde693
5 changed files with 228 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

20
tests_v1/plugins/model_plugins/test_kernel_plugin.py
View File

@@ -42,3 +42,23 @@ class TestKernelPlugin(unittest.TestCase):
model = apply_kernel(model, npu_swiglu.NpuSwiGluKernel)
assert model.model.layers[0].mlp.forward is not original_swiglu_forward
class Test_Use_V1_Kernels(unittest.TestCase):
@patch("torch.accelerator.current_accelerator")
def test_use_v1_kernels(self, mock_get_accelerator):
mock_device = MagicMock()
mock_device.type = "npu"
mock_get_accelerator.return_value = mock_device
model = AutoModelForCausalLM.from_pretrained("llamafactory/tiny-random-qwen2.5")
original_rmsnorm_forward = model.model.layers[0].input_layernorm.forward
original_swiglu_forward = model.model.layers[0].mlp.forward
from llamafactory.v1.plugins.model_plugins.kernels.registry import apply_available_kernels
model = apply_available_kernels(model)
assert model.model.layers[0].input_layernorm is not original_rmsnorm_forward
assert model.model.layers[0].mlp.forward is not original_swiglu_forward