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

Falcon embeddings (#1149) [skip docker]

Browse Source 
* 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
This commit is contained in:
Wing Lian
2024-01-22 21:01:42 -05:00
committed by GitHub
parent 0f77b8d798
commit e799e08d3c
10 changed files with 326 additions and 19 deletions
Download Patch File Download Diff File Expand all files Collapse all files

112
tests/e2e/patched/test_falcon_samplepack.py Normal file
View File

@@ -0,0 +1,112 @@
"""
E2E tests for falcon
"""
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 TestFalconPatched(unittest.TestCase):
"""
Test case for Falcon models
"""
@with_temp_dir
def test_qlora(self, temp_dir):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "illuin/tiny-random-FalconForCausalLM",
"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,
"lora_modules_to_save": ["word_embeddings", "lm_head"],
"val_set_size": 0.1,
"special_tokens": {
"bos_token": "<|endoftext|>",
"pad_token": "<|endoftext|>",
},
"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": "illuin/tiny-random-FalconForCausalLM",
"flash_attention": True,
"sample_packing": True,
"sequence_len": 2048,
"val_set_size": 0.1,
"special_tokens": {
"bos_token": "<|endoftext|>",
"pad_token": "<|endoftext|>",
},
"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) / "pytorch_model.bin").exists()

4
tests/e2e/patched/test_mixtral_samplepack.py
View File

@@ -32,6 +32,7 @@ class TestMixtral(unittest.TestCase):
"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",
@@ -57,7 +58,6 @@ class TestMixtral(unittest.TestCase):
"max_steps": 20,
"save_steps": 10,
"eval_steps": 10,
"sample_packing": True,
"bf16": "auto",
}
)
@@ -76,6 +76,7 @@ class TestMixtral(unittest.TestCase):
"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": {},
@@ -95,7 +96,6 @@ class TestMixtral(unittest.TestCase):
"max_steps": 20,
"save_steps": 10,
"eval_steps": 10,
"sample_packing": True,
"bf16": "auto",
}
)

166
tests/e2e/test_falcon.py Normal file
View File

@@ -0,0 +1,166 @@
"""
E2E tests for falcon
"""
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 TestFalcon(unittest.TestCase):
"""
Test case for falcon
"""
@with_temp_dir
def test_lora(self, temp_dir):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "illuin/tiny-random-FalconForCausalLM",
"flash_attention": True,
"sequence_len": 1024,
"load_in_8bit": True,
"adapter": "lora",
"lora_r": 32,
"lora_alpha": 64,
"lora_dropout": 0.05,
"lora_target_linear": True,
"lora_modules_to_save": [
"word_embeddings",
"lm_head",
],
"val_set_size": 0.1,
"special_tokens": {
"bos_token": "<|endoftext|>",
"pad_token": "<|endoftext|>",
},
"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_torch",
"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_lora_added_vocab(self, temp_dir):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "illuin/tiny-random-FalconForCausalLM",
"flash_attention": True,
"sequence_len": 1024,
"load_in_8bit": True,
"adapter": "lora",
"lora_r": 32,
"lora_alpha": 64,
"lora_dropout": 0.05,
"lora_target_linear": True,
"lora_modules_to_save": [
"word_embeddings",
"lm_head",
],
"val_set_size": 0.1,
"special_tokens": {
"bos_token": "<|endoftext|>",
"pad_token": "<|endoftext|>",
},
"tokens": [
"<|im_start|>",
"<|im_end|>",
],
"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_torch",
"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": "illuin/tiny-random-FalconForCausalLM",
"flash_attention": True,
"sequence_len": 1024,
"val_set_size": 0.1,
"special_tokens": {
"bos_token": "<|endoftext|>",
"pad_token": "<|endoftext|>",
},
"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_torch",
"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) / "pytorch_model.bin").exists()