e2f01de0e8
* Fix Axolotl ReLoRA optimizer reset scope * fix: make relora reset method honor relora_prune_ratio When relora_prune_method='reset' and relora_prune_ratio is explicitly set, the ratio was silently ignored and replaced with the hardcoded _FULL_RESET_RATIO (0.999). Fix by moving the default-ratio logic to ReLoRACallback.on_step_begin: None maps to _FULL_RESET_RATIO for reset and 0.9 for other methods. reset_optimizer now uses the same random pruning path for both 'random' and 'reset'. Also consolidate three-layer default mismatch: schema default for relora_prune_method is now 'magnitude' (single canonical source); dataclass defaults for both fields changed to None to eliminate the conflicting fallback layer. Tests updated: removed the test case that verified the old broken behavior (reset ignoring ratio), added two cases proving reset honors the passed ratio. E2E reset fixture now uses ratio=0.5 to make it unambiguous that the ratio is honored. * Fix ReLoRA uint8 pruning regression --------- Signed-off-by: Wing Lian <wing@axolotl.ai> Co-authored-by: Axolotl Swarm <no-reply@axolotl.ai>
149 lines
5.3 KiB
Python
149 lines
5.3 KiB
Python
"""
|
|
E2E tests for relora llama
|
|
"""
|
|
|
|
import unittest
|
|
from pathlib import Path
|
|
|
|
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, check_tensorboard, with_temp_dir
|
|
|
|
|
|
class TestReLoraLlama(unittest.TestCase):
|
|
"""
|
|
Test case for Llama models using LoRA
|
|
"""
|
|
|
|
@with_temp_dir
|
|
def test_relora(self, temp_dir):
|
|
cfg = DictDefault(
|
|
{
|
|
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
|
"sequence_len": 2048,
|
|
"sample_packing": True,
|
|
"pad_to_sequence_len": True,
|
|
"flash_attention": True,
|
|
"load_in_8bit": True,
|
|
"adapter": "lora",
|
|
"lora_r": 8,
|
|
"lora_alpha": 16,
|
|
"lora_dropout": 0.05,
|
|
"lora_target_modules": ["q_proj", "v_proj"],
|
|
"relora": True,
|
|
"jagged_restart_steps": 50,
|
|
"jagged_restart_warmup_steps": 10,
|
|
"jagged_restart_anneal_steps": 10,
|
|
"relora_prune_ratio": 0.9,
|
|
"relora_cpu_offload": True,
|
|
"val_set_size": 0.0,
|
|
"special_tokens": {
|
|
"pad_token": "<|endoftext|>",
|
|
},
|
|
"chat_template": "chatml",
|
|
"datasets": [
|
|
{
|
|
"path": "mlabonne/FineTome-100k",
|
|
"type": "chat_template",
|
|
"split": "train[:10%]",
|
|
"field_messages": "conversations",
|
|
"message_field_role": "from",
|
|
"message_field_content": "value",
|
|
},
|
|
],
|
|
"warmup_steps": 10,
|
|
"num_epochs": 2,
|
|
"max_steps": 105, # at least 2x restart cadence
|
|
"micro_batch_size": 2,
|
|
"gradient_accumulation_steps": 1,
|
|
"output_dir": temp_dir,
|
|
"learning_rate": 0.00001,
|
|
"optimizer": "adamw_8bit",
|
|
"lr_scheduler": "cosine",
|
|
"use_tensorboard": True,
|
|
"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(Path(temp_dir) / "checkpoint-100/adapter", cfg)
|
|
assert (Path(temp_dir) / "checkpoint-100/relora/model.safetensors").exists(), (
|
|
"Relora model checkpoint not found"
|
|
)
|
|
|
|
check_tensorboard(
|
|
temp_dir + "/runs", "train/grad_norm", 0.2, "grad_norm is too high"
|
|
)
|
|
|
|
@with_temp_dir
|
|
def test_relora_reset_method(self, temp_dir):
|
|
cfg = DictDefault(
|
|
{
|
|
"base_model": "HuggingFaceTB/SmolLM2-135M",
|
|
"sequence_len": 2048,
|
|
"sample_packing": True,
|
|
"pad_to_sequence_len": True,
|
|
"flash_attention": True,
|
|
"load_in_8bit": True,
|
|
"adapter": "lora",
|
|
"lora_r": 8,
|
|
"lora_alpha": 16,
|
|
"lora_dropout": 0.05,
|
|
"lora_target_modules": ["q_proj", "v_proj"],
|
|
"relora": True,
|
|
"jagged_restart_steps": 50,
|
|
"jagged_restart_warmup_steps": 10,
|
|
"jagged_restart_anneal_steps": 10,
|
|
"relora_prune_ratio": 0.5, # explicitly honored by reset (not ignored)
|
|
"relora_prune_method": "reset",
|
|
"relora_cpu_offload": True,
|
|
"val_set_size": 0.0,
|
|
"special_tokens": {
|
|
"pad_token": "<|endoftext|>",
|
|
},
|
|
"chat_template": "chatml",
|
|
"datasets": [
|
|
{
|
|
"path": "mlabonne/FineTome-100k",
|
|
"type": "chat_template",
|
|
"split": "train[:10%]",
|
|
"field_messages": "conversations",
|
|
"message_field_role": "from",
|
|
"message_field_content": "value",
|
|
},
|
|
],
|
|
"warmup_steps": 10,
|
|
"num_epochs": 2,
|
|
"max_steps": 105,
|
|
"micro_batch_size": 2,
|
|
"gradient_accumulation_steps": 1,
|
|
"output_dir": temp_dir,
|
|
"learning_rate": 0.00001,
|
|
"optimizer": "adamw_8bit",
|
|
"lr_scheduler": "cosine",
|
|
"use_tensorboard": True,
|
|
"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(Path(temp_dir) / "checkpoint-100/adapter", cfg)
|
|
assert (Path(temp_dir) / "checkpoint-100/relora/model.safetensors").exists(), (
|
|
"Relora model checkpoint not found"
|
|
)
|
|
|
|
check_tensorboard(
|
|
temp_dir + "/runs", "train/grad_norm", 0.2, "grad_norm is too high"
|
|
)
|