perform flakey patched tests in individual runner
This commit is contained in:
Wing Lian
0
tests/e2e/each/__init__.py
Normal file
0
tests/e2e/each/__init__.py
Normal file
89
tests/e2e/each/test_fa_xentropy.py
Normal file
89
tests/e2e/each/test_fa_xentropy.py
Normal file
@@ -0,0 +1,89 @@
|
||||
"""
|
||||
E2E tests for lora llama
|
||||
"""
|
||||
|
||||
import logging
|
||||
import os
|
||||
from pathlib import Path
|
||||
|
||||
import pytest
|
||||
from transformers.utils import is_torch_bf16_gpu_available
|
||||
|
||||
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 check_tensorboard
|
||||
|
||||
LOG = logging.getLogger("axolotl.tests.e2e")
|
||||
os.environ["WANDB_DISABLED"] = "true"
|
||||
|
||||
|
||||
class TestFAXentropyLlama:
|
||||
"""
|
||||
Test case for Llama models using LoRA w multipack
|
||||
"""
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
"gradient_accumulation_steps",
|
||||
[1, 4],
|
||||
)
|
||||
def test_lora_packing_fa_cross_entropy(self, temp_dir, gradient_accumulation_steps):
|
||||
# pylint: disable=duplicate-code
|
||||
cfg = DictDefault(
|
||||
{
|
||||
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
||||
"sequence_len": 1024,
|
||||
"sample_packing": True,
|
||||
"flash_attention": True,
|
||||
"flash_attn_cross_entropy": True,
|
||||
"load_in_8bit": True,
|
||||
"adapter": "lora",
|
||||
"lora_r": 8,
|
||||
"lora_alpha": 16,
|
||||
"lora_dropout": 0.05,
|
||||
"lora_target_linear": True,
|
||||
"val_set_size": 0.05,
|
||||
"special_tokens": {
|
||||
"pad_token": "<|endoftext|>",
|
||||
},
|
||||
"chat_template": "chatml",
|
||||
"datasets": [
|
||||
{
|
||||
"path": "mlabonne/FineTome-100k",
|
||||
"field_messages": "conversations",
|
||||
"message_field_content": "value",
|
||||
"message_field_role": "from",
|
||||
"type": "chat_template",
|
||||
"split": "train[:2%]",
|
||||
},
|
||||
],
|
||||
"num_epochs": 1,
|
||||
"max_steps": 5,
|
||||
"save_steps": 5,
|
||||
"micro_batch_size": 2,
|
||||
"gradient_accumulation_steps": gradient_accumulation_steps,
|
||||
"output_dir": temp_dir,
|
||||
"learning_rate": 0.00001,
|
||||
"optimizer": "adamw_8bit",
|
||||
"lr_scheduler": "cosine",
|
||||
"use_tensorboard": True,
|
||||
}
|
||||
)
|
||||
if is_torch_bf16_gpu_available():
|
||||
cfg.bf16 = True
|
||||
else:
|
||||
cfg.fp16 = True
|
||||
|
||||
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()
|
||||
|
||||
check_tensorboard(
|
||||
temp_dir + "/runs", "train/train_loss", 1.5, "Train Loss is too high"
|
||||
)
|
||||
129
tests/e2e/each/test_lora_llama_multipack.py
Normal file
129
tests/e2e/each/test_lora_llama_multipack.py
Normal file
@@ -0,0 +1,129 @@
|
||||
"""
|
||||
E2E tests for lora llama
|
||||
"""
|
||||
|
||||
import logging
|
||||
import os
|
||||
import unittest
|
||||
from pathlib import Path
|
||||
|
||||
import pytest
|
||||
from transformers.utils import is_auto_gptq_available, is_torch_bf16_gpu_available
|
||||
|
||||
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 TestLoraLlama(unittest.TestCase):
|
||||
"""
|
||||
Test case for Llama models using LoRA w multipack
|
||||
"""
|
||||
|
||||
@with_temp_dir
|
||||
def test_lora_packing(self, temp_dir):
|
||||
# pylint: disable=duplicate-code
|
||||
cfg = DictDefault(
|
||||
{
|
||||
"base_model": "JackFram/llama-68m",
|
||||
"tokenizer_type": "LlamaTokenizer",
|
||||
"sequence_len": 1024,
|
||||
"sample_packing": True,
|
||||
"flash_attention": True,
|
||||
"load_in_8bit": True,
|
||||
"adapter": "lora",
|
||||
"lora_r": 32,
|
||||
"lora_alpha": 64,
|
||||
"lora_dropout": 0.05,
|
||||
"lora_target_linear": True,
|
||||
"val_set_size": 0.2,
|
||||
"special_tokens": {
|
||||
"unk_token": "<unk>",
|
||||
"bos_token": "<s>",
|
||||
"eos_token": "</s>",
|
||||
},
|
||||
"datasets": [
|
||||
{
|
||||
"path": "mhenrichsen/alpaca_2k_test",
|
||||
"type": "alpaca",
|
||||
},
|
||||
],
|
||||
"num_epochs": 2,
|
||||
"max_steps": 20,
|
||||
"save_steps": 10,
|
||||
"micro_batch_size": 8,
|
||||
"gradient_accumulation_steps": 1,
|
||||
"output_dir": temp_dir,
|
||||
"learning_rate": 0.00001,
|
||||
"optimizer": "adamw_torch",
|
||||
"lr_scheduler": "cosine",
|
||||
}
|
||||
)
|
||||
if is_torch_bf16_gpu_available():
|
||||
cfg.bf16 = True
|
||||
else:
|
||||
cfg.fp16 = True
|
||||
|
||||
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()
|
||||
|
||||
@pytest.mark.skipif(not is_auto_gptq_available(), reason="auto-gptq not available")
|
||||
@with_temp_dir
|
||||
def test_lora_gptq_packed(self, temp_dir):
|
||||
# pylint: disable=duplicate-code
|
||||
cfg = DictDefault(
|
||||
{
|
||||
"base_model": "TheBlokeAI/jackfram_llama-68m-GPTQ",
|
||||
"model_type": "AutoModelForCausalLM",
|
||||
"tokenizer_type": "LlamaTokenizer",
|
||||
"sequence_len": 1024,
|
||||
"sample_packing": True,
|
||||
"flash_attention": True,
|
||||
"load_in_8bit": True,
|
||||
"adapter": "lora",
|
||||
"gptq": True,
|
||||
"gptq_disable_exllama": True,
|
||||
"lora_r": 32,
|
||||
"lora_alpha": 64,
|
||||
"lora_dropout": 0.05,
|
||||
"lora_target_linear": True,
|
||||
"val_set_size": 0.02,
|
||||
"special_tokens": {
|
||||
"unk_token": "<unk>",
|
||||
"bos_token": "<s>",
|
||||
"eos_token": "</s>",
|
||||
},
|
||||
"datasets": [
|
||||
{
|
||||
"path": "mhenrichsen/alpaca_2k_test",
|
||||
"type": "alpaca",
|
||||
},
|
||||
],
|
||||
"num_epochs": 2,
|
||||
"max_steps": 20,
|
||||
"save_steps": 0.5,
|
||||
"micro_batch_size": 8,
|
||||
"gradient_accumulation_steps": 1,
|
||||
"output_dir": temp_dir,
|
||||
"learning_rate": 0.00001,
|
||||
"optimizer": "adamw_torch",
|
||||
"lr_scheduler": "cosine",
|
||||
}
|
||||
)
|
||||
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()
|
||||
95
tests/e2e/each/test_resume.py
Normal file
95
tests/e2e/each/test_resume.py
Normal file
@@ -0,0 +1,95 @@
|
||||
"""
|
||||
E2E tests for resuming training
|
||||
"""
|
||||
|
||||
import logging
|
||||
import os
|
||||
import re
|
||||
import subprocess
|
||||
from pathlib import Path
|
||||
|
||||
from transformers.utils import is_torch_bf16_gpu_available
|
||||
|
||||
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 most_recent_subdir
|
||||
|
||||
LOG = logging.getLogger("axolotl.tests.e2e")
|
||||
os.environ["WANDB_DISABLED"] = "true"
|
||||
|
||||
|
||||
class TestResumeLlama:
|
||||
"""
|
||||
Test case for resuming training of llama models
|
||||
"""
|
||||
|
||||
def test_resume_lora_packed(self, temp_dir):
|
||||
# pylint: disable=duplicate-code
|
||||
cfg = DictDefault(
|
||||
{
|
||||
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
||||
"sequence_len": 1024,
|
||||
"sample_packing": True,
|
||||
"flash_attention": True,
|
||||
"load_in_8bit": True,
|
||||
"adapter": "lora",
|
||||
"lora_r": 8,
|
||||
"lora_alpha": 16,
|
||||
"lora_dropout": 0.05,
|
||||
"lora_target_linear": True,
|
||||
"val_set_size": 0.001,
|
||||
"special_tokens": {
|
||||
"pad_token": "<|endoftext|>",
|
||||
},
|
||||
"datasets": [
|
||||
{
|
||||
"path": "vicgalle/alpaca-gpt4",
|
||||
"type": "alpaca",
|
||||
},
|
||||
],
|
||||
"num_epochs": 2,
|
||||
"micro_batch_size": 1,
|
||||
"gradient_accumulation_steps": 1,
|
||||
"output_dir": temp_dir,
|
||||
"learning_rate": 0.00001,
|
||||
"optimizer": "adamw_8bit",
|
||||
"lr_scheduler": "cosine",
|
||||
"save_steps": 3,
|
||||
"save_total_limit": 5,
|
||||
"max_steps": 15,
|
||||
"use_tensorboard": True,
|
||||
}
|
||||
)
|
||||
if is_torch_bf16_gpu_available():
|
||||
cfg.bf16 = True
|
||||
else:
|
||||
cfg.fp16 = True
|
||||
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)
|
||||
|
||||
resume_cfg = cfg | DictDefault(
|
||||
{
|
||||
"resume_from_checkpoint": f"{temp_dir}/checkpoint-9/",
|
||||
}
|
||||
)
|
||||
normalize_config(resume_cfg)
|
||||
cli_args = TrainerCliArgs()
|
||||
|
||||
train(cfg=resume_cfg, cli_args=cli_args, dataset_meta=dataset_meta)
|
||||
assert (Path(temp_dir) / "adapter_model.bin").exists()
|
||||
|
||||
tb_log_path_1 = most_recent_subdir(temp_dir + "/runs")
|
||||
cmd = f"tensorboard --inspect --logdir {tb_log_path_1}"
|
||||
res = subprocess.run(
|
||||
cmd, shell=True, text=True, capture_output=True, check=True
|
||||
)
|
||||
pattern = r"first_step\s+(\d+)"
|
||||
first_steps = int(re.findall(pattern, res.stdout)[0])
|
||||
assert first_steps == 10
|
||||
186
tests/e2e/each/test_unsloth_qlora.py
Normal file
186
tests/e2e/each/test_unsloth_qlora.py
Normal file
@@ -0,0 +1,186 @@
|
||||
"""
|
||||
e2e tests for unsloth qlora
|
||||
"""
|
||||
import logging
|
||||
import os
|
||||
from pathlib import Path
|
||||
|
||||
import pytest
|
||||
|
||||
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 check_tensorboard
|
||||
|
||||
LOG = logging.getLogger("axolotl.tests.e2e")
|
||||
os.environ["WANDB_DISABLED"] = "true"
|
||||
|
||||
|
||||
# pylint: disable=duplicate-code
|
||||
class TestUnslothQLoRA:
|
||||
"""
|
||||
Test class for Unsloth QLoRA Llama models
|
||||
"""
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
"sample_packing",
|
||||
[True, False],
|
||||
)
|
||||
def test_unsloth_llama_qlora_fa2(self, temp_dir, sample_packing):
|
||||
cfg = DictDefault(
|
||||
{
|
||||
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
||||
"sequence_len": 1024,
|
||||
"sample_packing": sample_packing,
|
||||
"flash_attention": True,
|
||||
"unsloth_lora_mlp": True,
|
||||
"unsloth_lora_qkv": True,
|
||||
"unsloth_lora_o": True,
|
||||
"load_in_4bit": True,
|
||||
"adapter": "qlora",
|
||||
"lora_r": 16,
|
||||
"lora_alpha": 16,
|
||||
"lora_dropout": 0.05,
|
||||
"lora_target_linear": True,
|
||||
"val_set_size": 0.05,
|
||||
"special_tokens": {
|
||||
"pad_token": "<|endoftext|>",
|
||||
},
|
||||
"datasets": [
|
||||
{
|
||||
"path": "mhenrichsen/alpaca_2k_test",
|
||||
"type": "alpaca",
|
||||
},
|
||||
],
|
||||
"num_epochs": 1,
|
||||
"max_steps": 5,
|
||||
"save_steps": 10,
|
||||
"micro_batch_size": 4,
|
||||
"gradient_accumulation_steps": 2,
|
||||
"output_dir": temp_dir,
|
||||
"learning_rate": 0.00001,
|
||||
"optimizer": "adamw_8bit",
|
||||
"lr_scheduler": "cosine",
|
||||
"use_tensorboard": True,
|
||||
"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()
|
||||
|
||||
check_tensorboard(
|
||||
temp_dir + "/runs", "train/train_loss", 2.0, "Train Loss is too high"
|
||||
)
|
||||
|
||||
def test_unsloth_llama_qlora_unpacked(self, temp_dir):
|
||||
cfg = DictDefault(
|
||||
{
|
||||
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
||||
"sequence_len": 1024,
|
||||
"unsloth_lora_mlp": True,
|
||||
"unsloth_lora_qkv": True,
|
||||
"unsloth_lora_o": True,
|
||||
"sample_packing": False,
|
||||
"load_in_4bit": True,
|
||||
"adapter": "qlora",
|
||||
"lora_r": 16,
|
||||
"lora_alpha": 16,
|
||||
"lora_dropout": 0.05,
|
||||
"lora_target_linear": True,
|
||||
"val_set_size": 0.05,
|
||||
"special_tokens": {
|
||||
"pad_token": "<|endoftext|>",
|
||||
},
|
||||
"datasets": [
|
||||
{
|
||||
"path": "mhenrichsen/alpaca_2k_test",
|
||||
"type": "alpaca",
|
||||
},
|
||||
],
|
||||
"num_epochs": 1,
|
||||
"max_steps": 5,
|
||||
"save_steps": 10,
|
||||
"micro_batch_size": 4,
|
||||
"gradient_accumulation_steps": 2,
|
||||
"output_dir": temp_dir,
|
||||
"learning_rate": 0.00001,
|
||||
"optimizer": "adamw_8bit",
|
||||
"lr_scheduler": "cosine",
|
||||
"use_tensorboard": True,
|
||||
"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()
|
||||
|
||||
check_tensorboard(
|
||||
temp_dir + "/runs", "train/train_loss", 2.0, "Train Loss is too high"
|
||||
)
|
||||
|
||||
@pytest.mark.parametrize(
|
||||
"sdp_attention",
|
||||
[True, False],
|
||||
)
|
||||
def test_unsloth_llama_qlora_unpacked_no_fa2_fp16(self, temp_dir, sdp_attention):
|
||||
cfg = DictDefault(
|
||||
{
|
||||
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
||||
"sequence_len": 1024,
|
||||
"unsloth_lora_mlp": True,
|
||||
"unsloth_lora_qkv": True,
|
||||
"unsloth_lora_o": True,
|
||||
"sample_packing": False,
|
||||
"load_in_4bit": True,
|
||||
"adapter": "qlora",
|
||||
"lora_r": 16,
|
||||
"lora_alpha": 16,
|
||||
"lora_dropout": 0.05,
|
||||
"lora_target_linear": True,
|
||||
"val_set_size": 0.05,
|
||||
"special_tokens": {
|
||||
"pad_token": "<|endoftext|>",
|
||||
},
|
||||
"datasets": [
|
||||
{
|
||||
"path": "mhenrichsen/alpaca_2k_test",
|
||||
"type": "alpaca",
|
||||
},
|
||||
],
|
||||
"num_epochs": 1,
|
||||
"max_steps": 5,
|
||||
"save_steps": 10,
|
||||
"micro_batch_size": 4,
|
||||
"gradient_accumulation_steps": 2,
|
||||
"sdp_attention": sdp_attention,
|
||||
"output_dir": temp_dir,
|
||||
"learning_rate": 0.00001,
|
||||
"optimizer": "adamw_8bit",
|
||||
"lr_scheduler": "cosine",
|
||||
"use_tensorboard": True,
|
||||
"fp16": True,
|
||||
}
|
||||
)
|
||||
|
||||
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()
|
||||
|
||||
check_tensorboard(
|
||||
temp_dir + "/runs", "train/train_loss", 2.0, "Train Loss is too high"
|
||||
)
|
||||
Reference in New Issue
Block a user