File size: 3,147 Bytes
bcc78d8 788649f bcc78d8 6910e6a bcc78d8 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 |
"""
E2E smoke tests to check that the monkeypatches are in place for certain configurations
"""
import unittest
from axolotl.common.cli import TrainerCliArgs
from axolotl.utils.config import normalize_config
from axolotl.utils.dict import DictDefault
from axolotl.utils.models import load_model, load_tokenizer
from ..utils import with_temp_dir
class TestModelPatches(unittest.TestCase):
"""
TestCases for the multipack monkey patches
"""
@with_temp_dir
def test_mixtral_multipack(self, temp_dir):
cfg = DictDefault(
{
"base_model": "hf-internal-testing/Mixtral-tiny",
"tokenizer_config": "mistralai/Mixtral-8x7B-v0.1",
"flash_attention": True,
"sample_packing": True,
"sequence_len": 2048,
"val_set_size": 0.1,
"special_tokens": {},
"datasets": [
{
"path": "mhenrichsen/alpaca_2k_test",
"type": "alpaca",
},
],
"num_epochs": 2,
"micro_batch_size": 2,
"gradient_accumulation_steps": 1,
"output_dir": temp_dir,
"learning_rate": 0.00001,
"optimizer": "adamw_bnb_8bit",
"lr_scheduler": "cosine",
"max_steps": 20,
"save_steps": 10,
"eval_steps": 10,
}
)
normalize_config(cfg)
cli_args = TrainerCliArgs()
tokenizer = load_tokenizer(cfg)
model, _ = load_model(cfg, tokenizer, inference=cli_args.inference)
assert (
"MixtralFlashAttention2"
in model.model.layers[0].self_attn.__class__.__name__
)
@with_temp_dir
def test_mistral_multipack(self, temp_dir):
cfg = DictDefault(
{
"base_model": "openaccess-ai-collective/tiny-mistral",
"flash_attention": True,
"sample_packing": True,
"sequence_len": 2048,
"val_set_size": 0.1,
"special_tokens": {},
"datasets": [
{
"path": "mhenrichsen/alpaca_2k_test",
"type": "alpaca",
},
],
"num_epochs": 2,
"micro_batch_size": 2,
"gradient_accumulation_steps": 1,
"output_dir": temp_dir,
"learning_rate": 0.00001,
"optimizer": "adamw_bnb_8bit",
"lr_scheduler": "cosine",
"max_steps": 20,
"save_steps": 10,
"eval_steps": 10,
}
)
normalize_config(cfg)
cli_args = TrainerCliArgs()
tokenizer = load_tokenizer(cfg)
model, _ = load_model(cfg, tokenizer, inference=cli_args.inference)
assert (
"axolotl.monkeypatch.mistral_attn_hijack_flash"
in model.model.layers[0].self_attn.forward.__module__
)
|