support liger kernel

Former-commit-id: 0f4e54abf6c5feb2329855a4047597ad5147720a
This commit is contained in:
hiyouga
2024-08-27 11:20:14 +08:00
parent 47877acc2a
commit 206a8364d4
8 changed files with 69 additions and 7 deletions

View File

@@ -0,0 +1,48 @@
# Copyright 2024 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.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from typing import TYPE_CHECKING
from ...extras.logging import get_logger
if TYPE_CHECKING:
from transformers import PretrainedConfig
from ...hparams import ModelArguments
logger = get_logger(__name__)
def configure_liger_kernel(config: "PretrainedConfig", model_args: "ModelArguments", is_trainable: bool) -> None:
if not is_trainable or not model_args.use_liger_kernel:
return
if getattr(config, "model_type", None) == "gemma":
from liger_kernel.transformers import apply_liger_kernel_to_gemma as apply_liger_kernel
elif getattr(config, "model_type", None) == "llama":
from liger_kernel.transformers import apply_liger_kernel_to_llama as apply_liger_kernel
elif getattr(config, "model_type", None) == "mistral":
from liger_kernel.transformers import apply_liger_kernel_to_mistral as apply_liger_kernel
elif getattr(config, "model_type", None) == "mixtral":
from liger_kernel.transformers import apply_liger_kernel_to_mixtral as apply_liger_kernel
elif getattr(config, "model_type", None) == "qwen2":
from liger_kernel.transformers import apply_liger_kernel_to_qwen2 as apply_liger_kernel
else:
logger.warning("Current model does not support liger kernel.")
return
apply_liger_kernel()
logger.info("Liger kernel has been applied to the model.")

View File

@@ -27,6 +27,7 @@ from ..extras.misc import infer_optim_dtype
from .model_utils.attention import configure_attn_implementation, print_attn_implementation
from .model_utils.checkpointing import prepare_model_for_training
from .model_utils.embedding import resize_embedding_layer
from .model_utils.liger_kernel import configure_liger_kernel
from .model_utils.longlora import configure_longlora
from .model_utils.moe import add_z3_leaf_module, configure_moe
from .model_utils.packing import configure_packing
@@ -70,6 +71,7 @@ def patch_config(
configure_attn_implementation(config, model_args, is_trainable)
configure_rope(config, model_args, is_trainable)
configure_liger_kernel(config, model_args, is_trainable)
configure_longlora(config, model_args, is_trainable)
configure_quantization(config, tokenizer, model_args, init_kwargs)
configure_moe(config, model_args, is_trainable)