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

optim e2e tests to run a bit faster (#2069) [skip ci]

Browse Source 
* optim e2e tests to run a bit faster

* run prequant w/o lora_modules_to_save

* use smollm2
This commit is contained in:
Wing Lian
2024-11-18 12:35:31 -05:00
committed by GitHub
parent 70cf79ef52
commit 9871fa060b
6 changed files with 43 additions and 23 deletions
Download Patch File Download Diff File Expand all files Collapse all files

38
tests/e2e/multigpu/test_llama.py
View File

@@ -25,7 +25,7 @@ AXOLOTL_ROOT = Path(__file__).parent.parent.parent.parent
@pytest.fixture(scope="session", autouse=True)
def download_model():
# download the model
snapshot_download("TinyLlama/TinyLlama_v1.1")
snapshot_download("HuggingFaceTB/SmolLM2-135M")
class TestMultiGPULlama:
@@ -37,7 +37,7 @@ class TestMultiGPULlama:
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "HuggingFaceTB/SmolLM-135M",
"base_model": "HuggingFaceTB/SmolLM2-135M",
"sequence_len": 2048,
"adapter": "lora",
"lora_r": 8,
@@ -93,7 +93,7 @@ class TestMultiGPULlama:
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "HuggingFaceTB/SmolLM-135M",
"base_model": "HuggingFaceTB/SmolLM2-135M",
"sequence_len": 2048,
"sample_packing": True,
"eval_sample_packing": False,
@@ -149,8 +149,7 @@ class TestMultiGPULlama:
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "TinyLlama/TinyLlama_v1.1",
"tokenizer_type": "LlamaTokenizer",
"base_model": "HuggingFaceTB/SmolLM2-135M",
"sequence_len": 2048,
"sample_packing": False,
"eval_sample_packing": False,
@@ -163,12 +162,10 @@ class TestMultiGPULlama:
"lora_target_linear": True,
"val_set_size": 0.05,
"special_tokens": {
"unk_token": "<unk>",
"bos_token": "<s>",
"eos_token": "</s>",
"pad_token": "<|endoftext|>",
},
"rl": "dpo",
"chat_template": "llama3",
"chat_template": "chatml",
"datasets": [
{
"path": "fozziethebeat/alpaca_messages_2k_dpo_test",
@@ -221,7 +218,7 @@ class TestMultiGPULlama:
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "HuggingFaceTB/SmolLM-135M",
"base_model": "HuggingFaceTB/SmolLM2-135M",
"sequence_len": 2048,
"sample_packing": False,
"eval_sample_packing": False,
@@ -294,7 +291,7 @@ class TestMultiGPULlama:
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "HuggingFaceTB/SmolLM-135M",
"base_model": "HuggingFaceTB/SmolLM2-135M",
"sequence_len": 2048,
"val_set_size": 0.01,
"special_tokens": {
@@ -359,7 +356,7 @@ class TestMultiGPULlama:
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "HuggingFaceTB/SmolLM-135M",
"base_model": "HuggingFaceTB/SmolLM2-135M",
"sample_packing": True,
"pad_to_sequence_len": True,
"sequence_len": 2048,
@@ -422,8 +419,7 @@ class TestMultiGPULlama:
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "axolotl-ai-co/TinyLlama_v1.1-bnb-nf4-bf16",
"tokenizer_type": "AutoTokenizer",
"base_model": "axolotl-ai-co/SmolLM2-135M-bnb-nf4-bf16",
"adapter": "qlora",
"mean_resizing_embeddings": True,
"load_in_4bit": True,
@@ -431,17 +427,17 @@ class TestMultiGPULlama:
"lora_alpha": 16,
"lora_dropout": 0.05,
"lora_target_linear": True,
"lora_modules_to_save": [
"embed_tokens",
"lm_head",
],
# "lora_modules_to_save": [
# "embed_tokens",
# "lm_head",
# ],
"sample_packing": True,
"eval_sample_packing": False,
"pad_to_sequence_len": True,
"sequence_len": 2048,
"val_set_size": 0.05,
"special_tokens": {
"pad_token": "</s>",
"pad_token": "<|endoftext|>",
},
"datasets": [
{
@@ -503,7 +499,7 @@ class TestMultiGPULlama:
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "HuggingFaceTB/SmolLM-135M",
"base_model": "HuggingFaceTB/SmolLM2-135M",
"sample_packing": True,
"pad_to_sequence_len": True,
"sequence_len": 2048,
@@ -553,7 +549,7 @@ class TestMultiGPULlama:
# pylint: disable=duplicate-code
cfg = DictDefault(
{
"base_model": "HuggingFaceTB/SmolLM-135M",
"base_model": "HuggingFaceTB/SmolLM2-135M",
"load_in_4bit": True,
"adapter": "qlora",
"lora_r": 8,