423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-29 10:10:35 +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

10
src/llamafactory/train/tuner.py
View File

@@ -1,4 +1,4 @@
# Copyright 2025 the LlamaFactory team.
# Copyright 2025 the KVCache.AI team, Approaching AI, and the LlamaFactory team.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
@@ -24,7 +24,7 @@ from ..data import get_template_and_fix_tokenizer
from ..extras import logging
from ..extras.constants import V_HEAD_SAFE_WEIGHTS_NAME, V_HEAD_WEIGHTS_NAME
from ..extras.misc import infer_optim_dtype
from ..extras.packages import is_mcore_adapter_available, is_ray_available
from ..extras.packages import is_kt_available, is_mcore_adapter_available, is_ray_available
from ..hparams import get_infer_args, get_ray_args, get_train_args, read_args
from ..model import load_model, load_tokenizer
from .callbacks import LogCallback, PissaConvertCallback, ReporterCallback
@@ -85,6 +85,12 @@ def _training_function(config: dict[str, Any]) -> None:
elif finetuning_args.stage == "pt":
run_pt(model_args, data_args, training_args, finetuning_args, callbacks)
elif finetuning_args.stage == "sft":
if model_args.use_kt:
if not is_kt_available():
raise ImportError("KTransformers is not installed. Please install it with `pip install ktransformers`.")
from .ksft.workflow import run_sft as run_sft_kt
run_sft_kt(model_args, data_args, training_args, finetuning_args, generating_args, callbacks)
run_sft(model_args, data_args, training_args, finetuning_args, generating_args, callbacks)
elif finetuning_args.stage == "rm":
run_rm(model_args, data_args, training_args, finetuning_args, callbacks)