423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2026-06-19 13:48:55 +08:00
Code Issues Packages Projects Releases Wiki Activity

[refactor] Add KTransformers AMX MoE SFT support via Accelerate (#10430)

Browse Source 
Co-authored-by: mrhaoxx <mr.haoxx@gmail.com>
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
Peilin Li
2026-05-01 01:47:58 +08:00
committed by GitHub
parent 6b08b948c9
commit 887ee2b121
39 changed files with 287 additions and 1968 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
src/llamafactory/model/loader.py
View File

@@ -31,7 +31,6 @@ from ..extras import logging
from ..extras.misc import count_parameters, skip_check_imports, try_download_model_from_other_hub
from ..extras.packages import is_torch_version_greater_than
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
@@ -144,12 +143,7 @@ def load_model(
model = None
lazy_load = False
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.use_unsloth:
if model_args.adapter_name_or_path is not None:
lazy_load = True
elif is_trainable: