convert exponential notation lr to floats (#771)
Browse files
src/axolotl/utils/config.py
CHANGED
|
@@ -119,6 +119,9 @@ def normalize_config(cfg):
|
|
| 119 |
or (cfg.model_type and "mistral" in cfg.model_type.lower())
|
| 120 |
)
|
| 121 |
|
|
|
|
|
|
|
|
|
|
| 122 |
log_gpu_memory_usage(LOG, "baseline", cfg.device)
|
| 123 |
|
| 124 |
|
|
|
|
| 119 |
or (cfg.model_type and "mistral" in cfg.model_type.lower())
|
| 120 |
)
|
| 121 |
|
| 122 |
+
if isinstance(cfg.learning_rate, str):
|
| 123 |
+
cfg.learning_rate = float(cfg.learning_rate)
|
| 124 |
+
|
| 125 |
log_gpu_memory_usage(LOG, "baseline", cfg.device)
|
| 126 |
|
| 127 |
|
tests/test_normalize_config.py
ADDED
|
@@ -0,0 +1,39 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
"""
|
| 2 |
+
Test classes for checking functionality of the cfg normalization
|
| 3 |
+
"""
|
| 4 |
+
import unittest
|
| 5 |
+
|
| 6 |
+
from axolotl.utils.config import normalize_config
|
| 7 |
+
from axolotl.utils.dict import DictDefault
|
| 8 |
+
|
| 9 |
+
|
| 10 |
+
class NormalizeConfigTestCase(unittest.TestCase):
|
| 11 |
+
"""
|
| 12 |
+
test class for normalize_config checks
|
| 13 |
+
"""
|
| 14 |
+
|
| 15 |
+
def _get_base_cfg(self):
|
| 16 |
+
return DictDefault(
|
| 17 |
+
{
|
| 18 |
+
"base_model": "JackFram/llama-68m",
|
| 19 |
+
"base_model_config": "JackFram/llama-68m",
|
| 20 |
+
"tokenizer_type": "LlamaTokenizer",
|
| 21 |
+
"num_epochs": 1,
|
| 22 |
+
"micro_batch_size": 1,
|
| 23 |
+
"gradient_accumulation_steps": 1,
|
| 24 |
+
}
|
| 25 |
+
)
|
| 26 |
+
|
| 27 |
+
def test_lr_as_float(self):
|
| 28 |
+
cfg = (
|
| 29 |
+
self._get_base_cfg()
|
| 30 |
+
| DictDefault( # pylint: disable=unsupported-binary-operation
|
| 31 |
+
{
|
| 32 |
+
"learning_rate": "5e-5",
|
| 33 |
+
}
|
| 34 |
+
)
|
| 35 |
+
)
|
| 36 |
+
|
| 37 |
+
normalize_config(cfg)
|
| 38 |
+
|
| 39 |
+
assert cfg.learning_rate == 0.00005
|