Feat: Update validate_config and add tests
Browse files- src/axolotl/utils/validation.py +24 -9
- tests/test_validation.py +95 -0
src/axolotl/utils/validation.py
CHANGED
|
@@ -3,24 +3,39 @@ import logging
|
|
| 3 |
|
| 4 |
def validate_config(cfg):
|
| 5 |
if cfg.load_4bit:
|
| 6 |
-
raise ValueError(
|
|
|
|
|
|
|
| 7 |
|
| 8 |
if cfg.adapter == "qlora":
|
| 9 |
if cfg.merge_lora:
|
| 10 |
# can't merge qlora if loaded in 8bit or 4bit
|
| 11 |
-
|
| 12 |
-
|
| 13 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 14 |
else:
|
| 15 |
-
|
| 16 |
-
|
| 17 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
|
| 19 |
if not cfg.load_in_8bit and cfg.adapter == "lora":
|
| 20 |
logging.warning("We recommend setting `load_in_8bit: true` for LORA finetuning")
|
| 21 |
-
|
| 22 |
if cfg.trust_remote_code:
|
| 23 |
-
logging.warning(
|
|
|
|
|
|
|
| 24 |
|
| 25 |
# TODO
|
| 26 |
# MPT 7b
|
|
|
|
| 3 |
|
| 4 |
def validate_config(cfg):
|
| 5 |
if cfg.load_4bit:
|
| 6 |
+
raise ValueError(
|
| 7 |
+
"cfg.load_4bit parameter has been deprecated and replaced by cfg.gptq"
|
| 8 |
+
)
|
| 9 |
|
| 10 |
if cfg.adapter == "qlora":
|
| 11 |
if cfg.merge_lora:
|
| 12 |
# can't merge qlora if loaded in 8bit or 4bit
|
| 13 |
+
if cfg.load_in_8bit:
|
| 14 |
+
raise ValueError("Can't merge qlora if loaded in 8bit")
|
| 15 |
+
|
| 16 |
+
if cfg.gptq:
|
| 17 |
+
raise ValueError("Can't merge qlora if gptq")
|
| 18 |
+
|
| 19 |
+
if cfg.load_in_4bit:
|
| 20 |
+
raise ValueError("Can't merge qlora if loaded in 4bit")
|
| 21 |
+
|
| 22 |
else:
|
| 23 |
+
if cfg.load_in_8bit:
|
| 24 |
+
raise ValueError("Can't load qlora in 8bit")
|
| 25 |
+
|
| 26 |
+
if cfg.gptq:
|
| 27 |
+
raise ValueError("Can't load qlora if gptq")
|
| 28 |
+
|
| 29 |
+
if not cfg.load_in_4bit:
|
| 30 |
+
raise ValueError("Require cfg.load_in_4bit to be True for qlora")
|
| 31 |
|
| 32 |
if not cfg.load_in_8bit and cfg.adapter == "lora":
|
| 33 |
logging.warning("We recommend setting `load_in_8bit: true` for LORA finetuning")
|
| 34 |
+
|
| 35 |
if cfg.trust_remote_code:
|
| 36 |
+
logging.warning(
|
| 37 |
+
"`trust_remote_code` is set to true. Please make sure that you reviewed the remote code/model."
|
| 38 |
+
)
|
| 39 |
|
| 40 |
# TODO
|
| 41 |
# MPT 7b
|
tests/test_validation.py
ADDED
|
@@ -0,0 +1,95 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
import unittest
|
| 2 |
+
|
| 3 |
+
import pytest
|
| 4 |
+
|
| 5 |
+
from axolotl.utils.validation import validate_config
|
| 6 |
+
from axolotl.utils.dict import DictDefault
|
| 7 |
+
|
| 8 |
+
|
| 9 |
+
class ValidationTest(unittest.TestCase):
|
| 10 |
+
def test_load_4bit_deprecate(self):
|
| 11 |
+
cfg = DictDefault(
|
| 12 |
+
{
|
| 13 |
+
"load_4bit": True,
|
| 14 |
+
}
|
| 15 |
+
)
|
| 16 |
+
|
| 17 |
+
with pytest.raises(ValueError):
|
| 18 |
+
validate_config(cfg)
|
| 19 |
+
|
| 20 |
+
def test_qlora(self):
|
| 21 |
+
base_cfg = DictDefault(
|
| 22 |
+
{
|
| 23 |
+
"adapter": "qlora",
|
| 24 |
+
}
|
| 25 |
+
)
|
| 26 |
+
|
| 27 |
+
cfg = base_cfg | DictDefault(
|
| 28 |
+
{
|
| 29 |
+
"load_in_8bit": True,
|
| 30 |
+
}
|
| 31 |
+
)
|
| 32 |
+
|
| 33 |
+
with pytest.raises(ValueError, match=r".*8bit.*"):
|
| 34 |
+
validate_config(cfg)
|
| 35 |
+
|
| 36 |
+
cfg = base_cfg | DictDefault(
|
| 37 |
+
{
|
| 38 |
+
"gptq": True,
|
| 39 |
+
}
|
| 40 |
+
)
|
| 41 |
+
|
| 42 |
+
with pytest.raises(ValueError, match=r".*gptq.*"):
|
| 43 |
+
validate_config(cfg)
|
| 44 |
+
|
| 45 |
+
cfg = base_cfg | DictDefault(
|
| 46 |
+
{
|
| 47 |
+
"load_in_4bit": False,
|
| 48 |
+
}
|
| 49 |
+
)
|
| 50 |
+
|
| 51 |
+
with pytest.raises(ValueError, match=r".*4bit.*"):
|
| 52 |
+
validate_config(cfg)
|
| 53 |
+
|
| 54 |
+
cfg = base_cfg | DictDefault(
|
| 55 |
+
{
|
| 56 |
+
"load_in_4bit": True,
|
| 57 |
+
}
|
| 58 |
+
)
|
| 59 |
+
|
| 60 |
+
validate_config(cfg)
|
| 61 |
+
|
| 62 |
+
def test_qlora_merge(self):
|
| 63 |
+
base_cfg = DictDefault(
|
| 64 |
+
{
|
| 65 |
+
"adapter": "qlora",
|
| 66 |
+
"merge_lora": True,
|
| 67 |
+
}
|
| 68 |
+
)
|
| 69 |
+
|
| 70 |
+
cfg = base_cfg | DictDefault(
|
| 71 |
+
{
|
| 72 |
+
"load_in_8bit": True,
|
| 73 |
+
}
|
| 74 |
+
)
|
| 75 |
+
|
| 76 |
+
with pytest.raises(ValueError, match=r".*8bit.*"):
|
| 77 |
+
validate_config(cfg)
|
| 78 |
+
|
| 79 |
+
cfg = base_cfg | DictDefault(
|
| 80 |
+
{
|
| 81 |
+
"gptq": True,
|
| 82 |
+
}
|
| 83 |
+
)
|
| 84 |
+
|
| 85 |
+
with pytest.raises(ValueError, match=r".*gptq.*"):
|
| 86 |
+
validate_config(cfg)
|
| 87 |
+
|
| 88 |
+
cfg = base_cfg | DictDefault(
|
| 89 |
+
{
|
| 90 |
+
"load_in_4bit": True,
|
| 91 |
+
}
|
| 92 |
+
)
|
| 93 |
+
|
| 94 |
+
with pytest.raises(ValueError, match=r".*4bit.*"):
|
| 95 |
+
validate_config(cfg)
|