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

misc fixes to add gptq tests (#621)

Browse Source 
* misc fixes to add gptq tests

* set bf16 needed for fa2
This commit is contained in:
Wing Lian
2023-09-21 21:52:12 -04:00
committed by GitHub
parent 97d3776ce6
commit 03e59077a0
5 changed files with 93 additions and 21 deletions
Download Patch File Download Diff File Expand all files Collapse all files

58
tests/e2e/test_lora_llama.py
View File

@@ -6,6 +6,7 @@ import logging
import os
import tempfile
import unittest
from pathlib import Path
from axolotl.cli import load_datasets
from axolotl.common.cli import TrainerCliArgs
@@ -24,6 +25,7 @@ class TestLoraLlama(unittest.TestCase):
def test_lora(self):
# pylint: disable=duplicate-code
output_dir = tempfile.mkdtemp()
cfg = DictDefault(
{
"base_model": "JackFram/llama-68m",
@@ -51,7 +53,7 @@ class TestLoraLlama(unittest.TestCase):
"num_epochs": 2,
"micro_batch_size": 8,
"gradient_accumulation_steps": 1,
"output_dir": tempfile.mkdtemp(),
"output_dir": output_dir,
"learning_rate": 0.00001,
"optimizer": "adamw_torch",
"lr_scheduler": "cosine",
@@ -62,9 +64,11 @@ class TestLoraLlama(unittest.TestCase):
dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
assert (Path(output_dir) / "adapter_model.bin").exists()
def test_lora_packing(self):
# pylint: disable=duplicate-code
output_dir = tempfile.mkdtemp()
cfg = DictDefault(
{
"base_model": "JackFram/llama-68m",
@@ -94,7 +98,7 @@ class TestLoraLlama(unittest.TestCase):
"num_epochs": 2,
"micro_batch_size": 8,
"gradient_accumulation_steps": 1,
"output_dir": tempfile.mkdtemp(),
"output_dir": output_dir,
"learning_rate": 0.00001,
"optimizer": "adamw_torch",
"lr_scheduler": "cosine",
@@ -105,3 +109,53 @@ class TestLoraLlama(unittest.TestCase):
dataset_meta = load_datasets(cfg=cfg, cli_args=cli_args)
train(cfg=cfg, cli_args=cli_args, dataset_meta=dataset_meta)
assert (Path(output_dir) / "adapter_model.bin").exists()
def test_lora_gptq(self):
# pylint: disable=duplicate-code
output_dir = tempfile.mkdtemp()
cfg = DictDefault(
{
"base_model": "TheBlokeAI/jackfram_llama-68m-GPTQ",
"base_model_config": "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": 0.1,
"special_tokens": {
"unk_token": "<unk>",
"bos_token": "<s>",
"eos_token": "</s>",
},
"datasets": [
{
"path": "mhenrichsen/alpaca_2k_test",
"type": "alpaca",
},
],
"num_epochs": 2,
"save_steps": 0.5,
"micro_batch_size": 8,
"gradient_accumulation_steps": 1,
"output_dir": output_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(output_dir) / "adapter_model.bin").exists()

14
tests/e2e/test_phi.py
View File

@@ -31,9 +31,9 @@ class TestPhi(unittest.TestCase):
"trust_remote_code": True,
"model_type": "MixFormerSequentialForCausalLM",
"tokenizer_type": "AutoTokenizer",
"sequence_len": 2048,
"sequence_len": 512,
"sample_packing": False,
"load_in_8bit": True,
"load_in_8bit": False,
"adapter": None,
"val_set_size": 0.1,
"special_tokens": {
@@ -55,8 +55,9 @@ class TestPhi(unittest.TestCase):
"gradient_accumulation_steps": 1,
"output_dir": tempfile.mkdtemp(),
"learning_rate": 0.00001,
"optimizer": "adamw_torch",
"optimizer": "adamw_bnb_8bit",
"lr_scheduler": "cosine",
"bf16": True,
}
)
normalize_config(cfg)
@@ -74,9 +75,9 @@ class TestPhi(unittest.TestCase):
"trust_remote_code": True,
"model_type": "MixFormerSequentialForCausalLM",
"tokenizer_type": "AutoTokenizer",
"sequence_len": 2048,
"sequence_len": 512,
"sample_packing": True,
"load_in_8bit": True,
"load_in_8bit": False,
"adapter": None,
"val_set_size": 0.1,
"special_tokens": {
@@ -98,8 +99,9 @@ class TestPhi(unittest.TestCase):
"gradient_accumulation_steps": 1,
"output_dir": tempfile.mkdtemp(),
"learning_rate": 0.00001,
"optimizer": "adamw_torch",
"optimizer": "adamw_bnb_8bit",
"lr_scheduler": "cosine",
"bf16": True,
}
)
normalize_config(cfg)