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

checkpoint model on first step callback (#2906)

Browse Source 
* checkpoint model on first step callback

* remove debug

* add test cases; update existing tests not to save on first step

* move test out of solo

* delete

* default to False

* typo
This commit is contained in:
Dan Saunders
2025-07-15 15:00:48 -04:00
committed by GitHub
parent d320ef6199
commit 10ba1622f7
146 changed files with 419 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

102
tests/e2e/test_save_first_step.py Normal file
View File

@@ -0,0 +1,102 @@
"""
E2E tests for relora llama
"""
import unittest
from pathlib import Path
import pytest
from axolotl.common.datasets import load_datasets
from axolotl.train import train
from axolotl.utils.config import normalize_config, validate_config
from axolotl.utils.dict import DictDefault
from .utils import check_model_output_exists, with_temp_dir
class TestSaveFirstStepCallback(unittest.TestCase):
"""Test cases for save_first_step callback config."""
@with_temp_dir
def test_save_first_step(self, temp_dir):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "HuggingFaceTB/SmolLM2-135M",
"tokenizer_type": "AutoTokenizer",
"sequence_len": 512,
"val_set_size": 0.02,
"special_tokens": {
"pad_token": "<|endoftext|>",
},
"datasets": [
{
"path": "mhenrichsen/alpaca_2k_test",
"type": "alpaca",
},
],
"num_epochs": 1,
"max_steps": 3,
"micro_batch_size": 2,
"gradient_accumulation_steps": 1,
"output_dir": temp_dir,
"learning_rate": 0.00001,
"optimizer": "adamw_bnb_8bit",
"lr_scheduler": "cosine",
"flash_attention": True,
"sample_packing": True,
"bf16": True,
"save_safetensors": True,
"save_first_step": True,
}
)
cfg = validate_config(cfg)
normalize_config(cfg)
dataset_meta = load_datasets(cfg=cfg)
train(cfg=cfg, dataset_meta=dataset_meta)
check_model_output_exists(str(Path(temp_dir) / "checkpoint-1"), cfg)
@with_temp_dir
def test_no_save_first_step(self, temp_dir):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "HuggingFaceTB/SmolLM2-135M",
"tokenizer_type": "AutoTokenizer",
"sequence_len": 512,
"val_set_size": 0.02,
"special_tokens": {
"pad_token": "<|endoftext|>",
},
"datasets": [
{
"path": "mhenrichsen/alpaca_2k_test",
"type": "alpaca",
},
],
"num_epochs": 1,
"max_steps": 3,
"micro_batch_size": 2,
"gradient_accumulation_steps": 1,
"output_dir": temp_dir,
"learning_rate": 0.00001,
"optimizer": "adamw_bnb_8bit",
"lr_scheduler": "cosine",
"flash_attention": True,
"sample_packing": True,
"bf16": True,
"save_safetensors": True,
"save_first_step": False,
}
)
cfg = validate_config(cfg)
normalize_config(cfg)
dataset_meta = load_datasets(cfg=cfg)
train(cfg=cfg, dataset_meta=dataset_meta)
with pytest.raises(AssertionError):
check_model_output_exists(str(Path(temp_dir) / "checkpoint-1"), cfg)