423A35C7/LLaMA-Factory
1
0
Fork 0
mirror of https://github.com/hiyouga/LLaMA-Factory.git synced 2025-12-17 12:20:37 +08:00
Code Issues Packages Projects Releases Wiki Activity

move configure_packing to llamafactory.model.patcher and fix constants

Browse Source 
Former-commit-id: 770f75dc83
This commit is contained in:
ancv
2024-06-21 00:45:06 +07:00
parent 988231026a
commit 5319447aa5
6 changed files with 16 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
src/llamafactory/model/model_utils/packing.py
View File

@@ -19,7 +19,7 @@ from transformers.modeling_attn_mask_utils import (
from transformers.utils import is_torch_bf16_gpu_available
from ...extras.logging import get_logger
from ...extras.constants import SUPPORTED_CLASS_FOR_MULTIPACK
from ...extras.constants import SUPPORTED_CLASS_EFFECIENT_PACKING
if TYPE_CHECKING:
from transformers import PretrainedConfig
@@ -303,7 +303,7 @@ def configure_packing(config: "PretrainedConfig", model_args: "ModelArguments")
else:
attn_implementation = getattr(config, "_attn_implementation", "")
if getattr(config, "model_type", None) in SUPPORTED_CLASS_FOR_MULTIPACK:
if getattr(config, "model_type", None) in SUPPORTED_CLASS_EFFECIENT_PACKING:
patch_for_multipack(getattr(config, "model_type", None), model_args.model_name_or_path, attn_implementation)
logger.info("Using packing sequences without cross-contamination attention for efficient training.")
else: