423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-16 11:50:35 +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

3
tests/model/model_utils/test_visual.py
View File

@@ -23,6 +23,7 @@ from llamafactory.hparams import FinetuningArguments, ModelArguments
from llamafactory.model.adapter import init_adapter
@pytest.mark.runs_on(["cpu","npu"])
@pytest.mark.parametrize("freeze_vision_tower", (False, True))
@pytest.mark.parametrize("freeze_multi_modal_projector", (False, True))
@pytest.mark.parametrize("freeze_language_model", (False, True))
@@ -48,6 +49,7 @@ def test_visual_full(freeze_vision_tower: bool, freeze_multi_modal_projector: bo
assert param.requires_grad != freeze_language_model
@pytest.mark.runs_on(["cpu","npu"])
@pytest.mark.parametrize("freeze_vision_tower,freeze_language_model", ((False, False), (False, True), (True, False)))
def test_visual_lora(freeze_vision_tower: bool, freeze_language_model: bool):
model_args = ModelArguments(model_name_or_path="Qwen/Qwen2-VL-2B-Instruct")
@@ -80,6 +82,7 @@ def test_visual_lora(freeze_vision_tower: bool, freeze_language_model: bool):
assert (merger_param_name in trainable_params) is False
@pytest.mark.runs_on(["cpu","npu"])
def test_visual_model_save_load():
# check VLM's state dict: https://github.com/huggingface/transformers/pull/38385
model_args = ModelArguments(model_name_or_path="Qwen/Qwen2-VL-2B-Instruct")