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

swap tinymodels that have safetensors for some ci tests (#2641)

Browse Source
This commit is contained in:
Wing Lian
2025-05-07 15:06:07 -04:00
committed by GitHub
parent 25e6c5f9bd
commit 0f3587174d
14 changed files with 137 additions and 20 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
tests/e2e/multigpu/test_llama.py
View File

@@ -479,7 +479,7 @@ class TestMultiGPULlama:
"sample_packing": True,
"pad_to_sequence_len": True,
"sequence_len": 2048,
"val_set_size": 0.05,
"val_set_size": 0.1,
"special_tokens": {
"pad_token": "<|endoftext|>",
},

10
tests/e2e/patched/lora_kernels/test_lora_kernel_patching.py
View File

@@ -29,12 +29,12 @@ from axolotl.utils.dict import DictDefault
MODEL_CONFIGS = [
{
"name": "openaccess-ai-collective/tiny-mistral",
"name": "trl-internal-testing/tiny-MistralForCausalLM-0.2",
"expected_activation": apply_lora_mlp_swiglu,
"dtype": torch.float16,
},
{
"name": "Qwen/Qwen2-7B",
"name": "trl-internal-testing/tiny-Qwen2ForCausalLM-2.5",
"expected_activation": apply_lora_mlp_swiglu,
"dtype": torch.float16,
},
@@ -44,7 +44,7 @@ MODEL_CONFIGS = [
"dtype": torch.float32,
},
{
"name": "mhenrichsen/gemma-2b",
"name": "trl-internal-testing/tiny-Gemma2ForCausalLM",
"expected_activation": apply_lora_mlp_geglu,
"dtype": torch.float16,
},
@@ -156,7 +156,9 @@ def test_swiglu_mlp_integration(small_llama_model):
def test_geglu_model_integration():
"""Test GeGLU activation with Gemma model."""
model = AutoModelForCausalLM.from_pretrained(
"mhenrichsen/gemma-2b", torch_dtype=torch.float16, device_map="cuda:0"
"trl-internal-testing/tiny-Gemma2ForCausalLM",
torch_dtype=torch.float16,
device_map="cuda:0",
)
peft_config = get_peft_config(
{

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

@@ -6,6 +6,8 @@ import logging
import os
import unittest
import pytest
from axolotl.cli.args import TrainerCliArgs
from axolotl.common.datasets import load_datasets
from axolotl.train import train
@@ -23,6 +25,7 @@ class TestFalconPatched(unittest.TestCase):
Test case for Falcon models
"""
@pytest.mark.skip(reason="no tiny models for testing with safetensors")
@with_temp_dir
def test_qlora(self, temp_dir):
# pylint: disable=duplicate-code
@@ -71,6 +74,7 @@ class TestFalconPatched(unittest.TestCase):
train(cfg=cfg, dataset_meta=dataset_meta)
check_model_output_exists(temp_dir, cfg)
@pytest.mark.skip(reason="no tiny models for testing with safetensors")
@with_temp_dir
def test_ft(self, temp_dir):
# pylint: disable=duplicate-code

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

@@ -28,7 +28,7 @@ class TestMistral(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "openaccess-ai-collective/tiny-mistral",
"base_model": "trl-internal-testing/tiny-MistralForCausalLM-0.2",
"flash_attention": True,
"sample_packing": True,
"sequence_len": 1024,
@@ -76,7 +76,7 @@ class TestMistral(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "openaccess-ai-collective/tiny-mistral",
"base_model": "trl-internal-testing/tiny-MistralForCausalLM-0.2",
"flash_attention": True,
"sample_packing": True,
"sequence_len": 1024,

2
tests/e2e/patched/test_model_patches.py
View File

@@ -56,7 +56,7 @@ class TestModelPatches(unittest.TestCase):
def test_mistral_multipack(self, temp_dir):
cfg = DictDefault(
{
"base_model": "openaccess-ai-collective/tiny-mistral",
"base_model": "trl-internal-testing/tiny-MistralForCausalLM-0.2",
"flash_attention": True,
"sample_packing": True,
"sequence_len": 2048,

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

@@ -15,7 +15,7 @@ from axolotl.train import train
from axolotl.utils.config import normalize_config, validate_config
from axolotl.utils.dict import DictDefault
from ..utils import check_model_output_exists, most_recent_subdir
from ..utils import check_model_output_exists, most_recent_subdir, require_torch_2_6_0
LOG = logging.getLogger("axolotl.tests.e2e")
os.environ["WANDB_DISABLED"] = "true"
@@ -26,6 +26,7 @@ class TestResumeLlama:
Test case for resuming training of llama models
"""
@require_torch_2_6_0
def test_resume_lora_packed(self, temp_dir):
# pylint: disable=duplicate-code
cfg = DictDefault(
@@ -62,6 +63,7 @@ class TestResumeLlama:
"save_total_limit": 5,
"max_steps": 15,
"use_tensorboard": True,
"save_safetensors": True,
}
)
if is_torch_bf16_gpu_available():

7
tests/e2e/test_evaluate.py
View File

@@ -19,14 +19,11 @@ class TestE2eEvaluate:
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "JackFram/llama-68m",
"tokenizer_type": "LlamaTokenizer",
"base_model": "HuggingFaceTB/SmolLM2-135M",
"sequence_len": 1024,
"val_set_size": 0.02,
"special_tokens": {
"unk_token": "<unk>",
"bos_token": "<s>",
"eos_token": "</s>",
"pad_token": "<|endoftext|>",
},
"datasets": [
{

5
tests/e2e/test_falcon.py
View File

@@ -6,6 +6,8 @@ import logging
import os
import unittest
import pytest
from axolotl.cli.args import TrainerCliArgs
from axolotl.common.datasets import load_datasets
from axolotl.train import train
@@ -23,6 +25,7 @@ class TestFalcon(unittest.TestCase):
Test case for falcon
"""
@pytest.mark.skip(reason="no tiny models for testing with safetensors")
@with_temp_dir
def test_lora(self, temp_dir):
# pylint: disable=duplicate-code
@@ -74,6 +77,7 @@ class TestFalcon(unittest.TestCase):
train(cfg=cfg, dataset_meta=dataset_meta)
check_model_output_exists(temp_dir, cfg)
@pytest.mark.skip(reason="no tiny models for testing with safetensors")
@with_temp_dir
def test_lora_added_vocab(self, temp_dir):
# pylint: disable=duplicate-code
@@ -129,6 +133,7 @@ class TestFalcon(unittest.TestCase):
train(cfg=cfg, dataset_meta=dataset_meta)
check_model_output_exists(temp_dir, cfg)
@pytest.mark.skip(reason="no tiny models for testing with safetensors")
@with_temp_dir
def test_ft(self, temp_dir):
# pylint: disable=duplicate-code

4
tests/e2e/test_mistral.py
View File

@@ -30,7 +30,7 @@ class TestMistral(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "openaccess-ai-collective/tiny-mistral",
"base_model": "trl-internal-testing/tiny-MistralForCausalLM-0.2",
"flash_attention": True,
"sequence_len": 1024,
"load_in_8bit": True,
@@ -77,7 +77,7 @@ class TestMistral(unittest.TestCase):
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "openaccess-ai-collective/tiny-mistral",
"base_model": "trl-internal-testing/tiny-MistralForCausalLM-0.2",
"flash_attention": True,
"sequence_len": 1024,
"val_set_size": 0.02,

1
tests/test_datasets.py
View File

@@ -414,7 +414,6 @@ class TestDatasetPreparation:
snapshot_path = snapshot_download(
repo_id="mhenrichsen/alpaca_2k_test",
repo_type="dataset",
local_dir=tmp_ds_path,
)
shutil.copytree(snapshot_path, tmp_ds_path, dirs_exist_ok=True)