tocmo0nlord/axolotl
1
0
Fork 0
Code Issues Pull Requests Actions 3 Packages Projects Releases Wiki Activity

Add CPU offload

Browse Source
This commit is contained in:
Charles Goddard
2023-07-24 21:07:36 -07:00
committed by Wing Lian
parent b57238ecec
commit c8f7213bc6
1 changed files with 6 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
src/axolotl/monkeypatch/relora.py
View File

@@ -42,8 +42,9 @@ def reset_optimizer(optimizer: torch.optim.Optimizer):
class ReLoRACallback(TrainerCallback):
def __init__(self, cfg: DictDefault):
self.relora_steps = cfg.relora_steps
self.last_full_model = cfg.base_model
self.cpu_offload = cfg.relora_cpu_offload
self.quantised = cfg.load_in_4bit or cfg.load_in_8bit
self.last_full_model = cfg.base_model
assert os.path.exists(
self.last_full_model
@@ -185,6 +186,7 @@ def merge_and_save(
model_dst: str,
reinit: bool = False,
quantized: bool = False,
cpu_offload: bool = False,
):
key_list = [key for key, _ in model.model.named_modules() if "lora" not in key]
@@ -235,9 +237,10 @@ def merge_and_save(
if isinstance(target, peft.tuners.lora.LoraLayer):
orig_weight = in_tensors[key + ".weight"]
old_dev = target.weight.device
math_dev = "cpu" if cpu_offload else old_dev
update = lora_delta_weight(target).detach()
new_weight = (orig_weight.to(old_dev) + update.to(old_dev)).cpu()
update = lora_delta_weight(target).detach().to(math_dev)
new_weight = orig_weight.to(math_dev) + update
out_tensors[key + ".weight"] = new_weight
if reinit: