423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-29 18:20:35 +08:00
Code Issues Packages Projects Releases Wiki Activity

[train] KTransformers SFT as backend engine for LLaMA-Factory (#9400)

Browse Source 
Co-authored-by: jimmy128 <jimmy128@noreply.gitcode.com>
Co-authored-by: Yaowei Zheng <hiyouga@buaa.edu.cn>
This commit is contained in:
Peilin Li
2025-11-04 15:54:12 +08:00
committed by GitHub
parent 3ae15da9c0
commit 934b3084ee
37 changed files with 2006 additions and 16 deletions
Download Patch File Download Diff File Expand all files Collapse all files

7
src/llamafactory/model/loader.py
View File

@@ -31,6 +31,7 @@ from trl import AutoModelForCausalLMWithValueHead
from ..extras import logging
from ..extras.misc import count_parameters, skip_check_imports, try_download_model_from_other_hub
from .adapter import init_adapter
from .model_utils.ktransformers import load_kt_pretrained_model
from .model_utils.liger_kernel import apply_liger_kernel
from .model_utils.misc import register_autoclass
from .model_utils.mod import convert_pretrained_model_to_mod, load_mod_pretrained_model
@@ -143,7 +144,11 @@ def load_model(
model = None
lazy_load = False
if model_args.use_unsloth:
if model_args.use_kt:
from ktransformers.sft.monkey_patch_torch_module import install_patch
install_patch()
model = load_kt_pretrained_model(config, model_args)
elif model_args.use_unsloth:
if model_args.adapter_name_or_path is not None:
lazy_load = True
elif is_trainable: