tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

use temp_dir kwarg instead

Browse Source
This commit is contained in:
Wing Lian
2023-11-06 07:31:46 -05:00
parent 7de6a5639c
commit 6dc68a653f
6 changed files with 31 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

18
tests/e2e/test_lora_llama.py
View File

@@ -25,7 +25,7 @@ class TestLoraLlama(unittest.TestCase):
"""
@with_temp_dir
def test_lora(self, output_dir):
def test_lora(self, temp_dir):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
@@ -53,7 +53,7 @@ class TestLoraLlama(unittest.TestCase):
"num_epochs": 2,
"micro_batch_size": 8,
"gradient_accumulation_steps": 1,
"output_dir": output_dir,
"output_dir": temp_dir,
"learning_rate": 0.00001,
"optimizer": "adamw_torch",
"lr_scheduler": "cosine",
@@ -64,10 +64,10 @@ class TestLoraLlama(unittest.TestCase):
dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
assert (Path(output_dir) / "adapter_model.bin").exists()
assert (Path(temp_dir) / "adapter_model.bin").exists()
@with_temp_dir
def test_lora_packing(self, output_dir):
def test_lora_packing(self, temp_dir):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
@@ -97,7 +97,7 @@ class TestLoraLlama(unittest.TestCase):
"num_epochs": 2,
"micro_batch_size": 8,
"gradient_accumulation_steps": 1,
"output_dir": output_dir,
"output_dir": temp_dir,
"learning_rate": 0.00001,
"optimizer": "adamw_torch",
"lr_scheduler": "cosine",
@@ -108,10 +108,10 @@ class TestLoraLlama(unittest.TestCase):
dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
assert (Path(output_dir) / "adapter_model.bin").exists()
assert (Path(temp_dir) / "adapter_model.bin").exists()
@with_temp_dir
def test_lora_gptq(self, output_dir):
def test_lora_gptq(self, temp_dir):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
@@ -145,7 +145,7 @@ class TestLoraLlama(unittest.TestCase):
"save_steps": 0.5,
"micro_batch_size": 8,
"gradient_accumulation_steps": 1,
"output_dir": output_dir,
"output_dir": temp_dir,
"learning_rate": 0.00001,
"optimizer": "adamw_torch",
"lr_scheduler": "cosine",
@@ -156,4 +156,4 @@ class TestLoraLlama(unittest.TestCase):
dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
assert (Path(output_dir) / "adapter_model.bin").exists()
assert (Path(temp_dir) / "adapter_model.bin").exists()