fc4e37920b
* Prepare for transformers v5 upgrade * fix hf cli * update for hf hub changes * fix tokenizer apply_chat_template args * remap include_tokens_per_second * fix tps * handle migration for warmup * use latest hf hub * Fix scan -> ls * fix import * fix for renaming of mistral common tokenizer -> backend * update for fixed tokenziation for llama * Skip phi35 tests for now * remove mistral patch fixed upstream in huggingface/transformers#41439 * use namespacing for patch * don't rely on sdist for e2e tests for now * run modal ci without waiting too * Fix dep for ci * fix imports * Fix fp8 check * fsdp2 fixes * fix version handling * update fsdp version tests for new v5 behavior * Fail multigpu tests after 3 failures * skip known v5 broken tests for now and cleanup * bump deps * unmark skipped test * re-enable test_fsdp_qlora_prequant_packed test * increase multigpu ci timeout * skip broken gemma3 test * reduce timout back to original 120min now that the hanging test is skipped * fix for un-necessary collator for pretraining with bsz=1 * fix: safe_serialization deprecated in transformers v5 rc01 (#3318) * torch_dtype deprecated * load model in float32 for consistency with tests * revert some test fixtures back * use hf cache ls instead of scan * don't strip fsdp_version more fdsp_Version fixes for v5 fix version in fsdp_config fix aliasing fix fsdp_version check check fsdp_version is 2 in both places * Transformers v5 rc2 (#3347) * bump dep * use latest fbgemm, grab model config as part of fixture, un-skip test * import AutoConfig * don't need more problematic autoconfig when specifying config.json manually * add fixtures for argilla ultrafeedback datasets * download phi4-reasoning * fix arg * update tests for phi fast tokenizer changes * use explicit model types for gemma3 --------- Co-authored-by: Wing Lian <wing@axolotl.ai> * fix: AutoModelForVision2Seq -> AutoModelForImageTextToText * chore: remove duplicate * fix: attempt fix gemma3 text mode * chore: lint * ga release of v5 * need property setter for name_or_path for mistral tokenizer * vllm not compatible with transformers v5 * setter for chat_template w mistral too --------- Co-authored-by: NanoCode012 <nano@axolotl.ai> Co-authored-by: salman <salman.mohammadi@outlook.com>
285 lines
9.3 KiB
Python
285 lines
9.3 KiB
Python
"""
|
|
E2E tests for custom optimizers using Llama
|
|
"""
|
|
|
|
import unittest
|
|
|
|
from axolotl.common.datasets import load_datasets
|
|
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,
|
|
require_torch_2_5_1,
|
|
require_torch_2_6_0,
|
|
require_torch_2_7_0,
|
|
with_temp_dir,
|
|
)
|
|
|
|
|
|
class TestCustomOptimizers(unittest.TestCase):
|
|
"""
|
|
Test case for Llama models using LoRA
|
|
"""
|
|
|
|
@with_temp_dir
|
|
def test_optimi_adamw(self, temp_dir):
|
|
cfg = DictDefault(
|
|
{
|
|
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
|
"model_type": "AutoModelForCausalLM",
|
|
"tokenizer_type": "AutoTokenizer",
|
|
"sequence_len": 1024,
|
|
"load_in_8bit": True,
|
|
"adapter": "lora",
|
|
"lora_r": 8,
|
|
"lora_alpha": 16,
|
|
"lora_dropout": 0.05,
|
|
"lora_target_linear": True,
|
|
"val_set_size": 0.02,
|
|
"special_tokens": {
|
|
"pad_token": "<|endoftext|>",
|
|
},
|
|
"datasets": [
|
|
{
|
|
"path": "mhenrichsen/alpaca_2k_test",
|
|
"type": "alpaca",
|
|
},
|
|
],
|
|
"num_epochs": 1,
|
|
"micro_batch_size": 8,
|
|
"gradient_accumulation_steps": 1,
|
|
"output_dir": temp_dir,
|
|
"learning_rate": 0.00001,
|
|
"optimizer": "optimi_adamw",
|
|
"max_steps": 5,
|
|
"lr_scheduler": "cosine",
|
|
"save_first_step": False,
|
|
}
|
|
)
|
|
|
|
cfg = validate_config(cfg)
|
|
normalize_config(cfg)
|
|
dataset_meta = load_datasets(cfg=cfg)
|
|
|
|
_, _, trainer = train(cfg=cfg, dataset_meta=dataset_meta)
|
|
check_model_output_exists(temp_dir, cfg)
|
|
assert trainer.optimizer.optimizer.__class__.__name__ == "AdamW"
|
|
|
|
@with_temp_dir
|
|
@require_torch_2_5_1
|
|
def test_adopt_adamw(self, temp_dir):
|
|
cfg = DictDefault(
|
|
{
|
|
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
|
"model_type": "AutoModelForCausalLM",
|
|
"tokenizer_type": "AutoTokenizer",
|
|
"sequence_len": 1024,
|
|
"load_in_8bit": True,
|
|
"adapter": "lora",
|
|
"lora_r": 8,
|
|
"lora_alpha": 16,
|
|
"lora_dropout": 0.05,
|
|
"lora_target_linear": True,
|
|
"val_set_size": 0.02,
|
|
"special_tokens": {
|
|
"pad_token": "<|endoftext|>",
|
|
},
|
|
"datasets": [
|
|
{
|
|
"path": "mhenrichsen/alpaca_2k_test",
|
|
"type": "alpaca",
|
|
},
|
|
],
|
|
"num_epochs": 1,
|
|
"max_steps": 5,
|
|
"micro_batch_size": 8,
|
|
"gradient_accumulation_steps": 1,
|
|
"output_dir": temp_dir,
|
|
"learning_rate": 0.00001,
|
|
"optimizer": "adopt_adamw",
|
|
"lr_scheduler": "cosine",
|
|
"save_first_step": False,
|
|
}
|
|
)
|
|
|
|
cfg = validate_config(cfg)
|
|
normalize_config(cfg)
|
|
dataset_meta = load_datasets(cfg=cfg)
|
|
|
|
_, _, trainer = train(cfg=cfg, dataset_meta=dataset_meta)
|
|
check_model_output_exists(temp_dir, cfg)
|
|
assert "ADOPT" in trainer.optimizer.optimizer.__class__.__name__
|
|
|
|
@with_temp_dir
|
|
@require_torch_2_5_1
|
|
def test_muon(self, temp_dir):
|
|
cfg = DictDefault(
|
|
{
|
|
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
|
"model_type": "AutoModelForCausalLM",
|
|
"tokenizer_type": "AutoTokenizer",
|
|
"sequence_len": 1024,
|
|
"load_in_8bit": True,
|
|
"adapter": "lora",
|
|
"lora_r": 8,
|
|
"lora_alpha": 16,
|
|
"lora_dropout": 0.05,
|
|
"lora_target_linear": True,
|
|
"val_set_size": 0.02,
|
|
"special_tokens": {
|
|
"pad_token": "<|endoftext|>",
|
|
},
|
|
"datasets": [
|
|
{
|
|
"path": "mhenrichsen/alpaca_2k_test",
|
|
"type": "alpaca",
|
|
},
|
|
],
|
|
"num_epochs": 1,
|
|
"max_steps": 5,
|
|
"micro_batch_size": 8,
|
|
"gradient_accumulation_steps": 1,
|
|
"output_dir": temp_dir,
|
|
"learning_rate": 0.00001,
|
|
"optimizer": "muon",
|
|
"lr_scheduler": "cosine",
|
|
"weight_decay": 0.01,
|
|
"save_first_step": False,
|
|
}
|
|
)
|
|
|
|
cfg = validate_config(cfg)
|
|
normalize_config(cfg)
|
|
dataset_meta = load_datasets(cfg=cfg)
|
|
|
|
_, _, trainer = train(cfg=cfg, dataset_meta=dataset_meta)
|
|
check_model_output_exists(temp_dir, cfg)
|
|
assert "Muon" in trainer.optimizer.optimizer.__class__.__name__
|
|
|
|
@with_temp_dir
|
|
@require_torch_2_7_0
|
|
def test_dion(self, temp_dir):
|
|
cfg = DictDefault(
|
|
{
|
|
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
|
"model_type": "AutoModelForCausalLM",
|
|
"tokenizer_type": "AutoTokenizer",
|
|
"sequence_len": 1024,
|
|
"val_set_size": 0.0,
|
|
"special_tokens": {
|
|
"pad_token": "<|endoftext|>",
|
|
},
|
|
"datasets": [
|
|
{
|
|
"path": "mhenrichsen/alpaca_2k_test",
|
|
"type": "alpaca",
|
|
},
|
|
],
|
|
"num_epochs": 1,
|
|
"max_steps": 5,
|
|
"micro_batch_size": 8,
|
|
"gradient_accumulation_steps": 1,
|
|
"output_dir": temp_dir,
|
|
"learning_rate": 0.00001,
|
|
"optimizer": "dion",
|
|
"dion_lr": 0.01,
|
|
"dion_momentum": 0.95,
|
|
"lr_scheduler": "cosine",
|
|
"weight_decay": 0.01,
|
|
"save_first_step": False,
|
|
}
|
|
)
|
|
|
|
cfg = validate_config(cfg)
|
|
normalize_config(cfg)
|
|
dataset_meta = load_datasets(cfg=cfg)
|
|
|
|
_, _, trainer = train(cfg=cfg, dataset_meta=dataset_meta)
|
|
check_model_output_exists(temp_dir, cfg)
|
|
assert "Dion" in trainer.optimizer.optimizer.__class__.__name__
|
|
|
|
@with_temp_dir
|
|
def test_fft_schedule_free_adamw(self, temp_dir):
|
|
cfg = DictDefault(
|
|
{
|
|
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
|
"model_type": "AutoModelForCausalLM",
|
|
"sequence_len": 1024,
|
|
"val_set_size": 0.01,
|
|
"special_tokens": {
|
|
"pad_token": "<|endoftext|>",
|
|
},
|
|
"datasets": [
|
|
{
|
|
"path": "mhenrichsen/alpaca_2k_test",
|
|
"type": "alpaca",
|
|
},
|
|
],
|
|
"num_epochs": 1,
|
|
"micro_batch_size": 2,
|
|
"gradient_accumulation_steps": 2,
|
|
"output_dir": temp_dir,
|
|
"learning_rate": 0.00001,
|
|
"optimizer": "schedule_free_adamw",
|
|
"lr_scheduler": "constant",
|
|
"max_steps": 10,
|
|
"save_first_step": False,
|
|
}
|
|
)
|
|
|
|
cfg = validate_config(cfg)
|
|
normalize_config(cfg)
|
|
dataset_meta = load_datasets(cfg=cfg)
|
|
|
|
train(cfg=cfg, dataset_meta=dataset_meta)
|
|
check_model_output_exists(temp_dir, cfg)
|
|
|
|
@with_temp_dir
|
|
@require_torch_2_6_0
|
|
def test_came_pytorch(self, temp_dir):
|
|
cfg = DictDefault(
|
|
{
|
|
"base_model": "JackFram/llama-68m",
|
|
"tokenizer_type": "LlamaTokenizer",
|
|
"sequence_len": 1024,
|
|
"load_in_8bit": True,
|
|
"adapter": "lora",
|
|
"lora_r": 8,
|
|
"lora_alpha": 16,
|
|
"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": 1,
|
|
"micro_batch_size": 8,
|
|
"gradient_accumulation_steps": 1,
|
|
"output_dir": temp_dir,
|
|
"learning_rate": 0.00001,
|
|
"optimizer": "came_pytorch",
|
|
"adam_beta3": 0.9999,
|
|
"adam_epsilon2": 1e-16,
|
|
"max_steps": 5,
|
|
"lr_scheduler": "cosine",
|
|
"save_first_step": False,
|
|
}
|
|
)
|
|
|
|
cfg = validate_config(cfg)
|
|
normalize_config(cfg)
|
|
dataset_meta = load_datasets(cfg=cfg)
|
|
|
|
train(cfg=cfg, dataset_meta=dataset_meta)
|
|
check_model_output_exists(temp_dir, cfg)
|