SUMIN/HuggingFace_transformer
1
0
Fork 0
Code Issues Pull Requests Actions 7 Packages Projects Releases Wiki Activity

Fix bnb training test failure (#34414)

Browse Source 
* Fix bnb training test: compatibility with OPTSdpaAttention
This commit is contained in:
Matthew Douglas
2024-10-25 10:23:20 -04:00
committed by GitHub
parent 186b8dc190
commit e447185b1f
2 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
tests/quantization/bnb/test_4bit.py
View File

@@ -29,6 +29,7 @@ from transformers import (
BitsAndBytesConfig, BitsAndBytesConfig,
pipeline, pipeline,
) )
from transformers.models.opt.modeling_opt import OPTAttention
from transformers.testing_utils import ( from transformers.testing_utils import (
apply_skip_if_not_implemented, apply_skip_if_not_implemented,
is_bitsandbytes_available, is_bitsandbytes_available,
@@ -565,7 +566,7 @@ class Bnb4BitTestTraining(Base4bitTest):
# Step 2: add adapters # Step 2: add adapters
for _, module in model.named_modules(): for _, module in model.named_modules():
if "OPTAttention" in repr(type(module)): if isinstance(module, OPTAttention):
module.q_proj = LoRALayer(module.q_proj, rank=16) module.q_proj = LoRALayer(module.q_proj, rank=16)
module.k_proj = LoRALayer(module.k_proj, rank=16) module.k_proj = LoRALayer(module.k_proj, rank=16)
module.v_proj = LoRALayer(module.v_proj, rank=16) module.v_proj = LoRALayer(module.v_proj, rank=16)

3
tests/quantization/bnb/test_mixed_int8.py
View File

@@ -29,6 +29,7 @@ from transformers import (
BitsAndBytesConfig, BitsAndBytesConfig,
pipeline, pipeline,
) )
from transformers.models.opt.modeling_opt import OPTAttention
from transformers.testing_utils import ( from transformers.testing_utils import (
apply_skip_if_not_implemented, apply_skip_if_not_implemented,
is_accelerate_available, is_accelerate_available,
@@ -868,7 +869,7 @@ class MixedInt8TestTraining(BaseMixedInt8Test):
# Step 2: add adapters # Step 2: add adapters
for _, module in model.named_modules(): for _, module in model.named_modules():
if "OPTAttention" in repr(type(module)): if isinstance(module, OPTAttention):
module.q_proj = LoRALayer(module.q_proj, rank=16) module.q_proj = LoRALayer(module.q_proj, rank=16)
module.k_proj = LoRALayer(module.k_proj, rank=16) module.k_proj = LoRALayer(module.k_proj, rank=16)
module.v_proj = LoRALayer(module.v_proj, rank=16) module.v_proj = LoRALayer(module.v_proj, rank=16)