e799e08d3c
* also fix multipack for falcon and add smoke tests * make sure to handle special tokens and added tokens for lora * fix reference to model_type * fix tests for falcon * fix stray typo * fixes for smoke tests
112 lines
3.5 KiB
Python
112 lines
3.5 KiB
Python
"""
|
|
E2E tests for mixtral
|
|
"""
|
|
|
|
import logging
|
|
import os
|
|
import unittest
|
|
from pathlib import Path
|
|
|
|
from axolotl.cli import load_datasets
|
|
from axolotl.common.cli import TrainerCliArgs
|
|
from axolotl.train import train
|
|
from axolotl.utils.config import normalize_config
|
|
from axolotl.utils.dict import DictDefault
|
|
|
|
from ..utils import with_temp_dir
|
|
|
|
LOG = logging.getLogger("axolotl.tests.e2e")
|
|
os.environ["WANDB_DISABLED"] = "true"
|
|
|
|
|
|
class TestMixtral(unittest.TestCase):
|
|
"""
|
|
Test case for Llama models using LoRA
|
|
"""
|
|
|
|
@with_temp_dir
|
|
def test_qlora(self, temp_dir):
|
|
# pylint: disable=duplicate-code
|
|
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,
|
|
"load_in_4bit": True,
|
|
"adapter": "qlora",
|
|
"lora_r": 16,
|
|
"lora_alpha": 32,
|
|
"lora_dropout": 0.1,
|
|
"lora_target_linear": True,
|
|
"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,
|
|
"bf16": "auto",
|
|
}
|
|
)
|
|
normalize_config(cfg)
|
|
cli_args = TrainerCliArgs()
|
|
dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
|
|
|
|
train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
|
|
assert (Path(temp_dir) / "adapter_model.bin").exists()
|
|
|
|
@with_temp_dir
|
|
def test_ft(self, temp_dir):
|
|
# pylint: disable=duplicate-code
|
|
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,
|
|
"bf16": "auto",
|
|
}
|
|
)
|
|
normalize_config(cfg)
|
|
cli_args = TrainerCliArgs()
|
|
dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
|
|
|
|
model, _ = train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
|
|
assert (
|
|
"MixtralFlashAttention2"
|
|
in model.model.layers[0].self_attn.__class__.__name__
|
|
)
|
|
assert (Path(temp_dir) / "pytorch_model.bin").exists()
|