From b153765fedf40817d801ea40093b7eb79801c4bf Mon Sep 17 00:00:00 2001 From: "Lin, Fanli" Date: Fri, 17 Jan 2025 04:28:26 -0800 Subject: [PATCH 1/2] fix quality issue --- tests/test_quantization.py | 9 +++------ 1 file changed, 3 insertions(+), 6 deletions(-) diff --git a/tests/test_quantization.py b/tests/test_quantization.py index 85b97b7f3a8..66fbc13496d 100644 --- a/tests/test_quantization.py +++ b/tests/test_quantization.py @@ -377,9 +377,8 @@ def test_int8_serialization(self): r""" Test whether it is possible to serialize a model in 8-bit. """ - from transformers import AutoConfig, AutoModelForCausalLM - from bitsandbytes.nn import Int8Params + from transformers import AutoConfig, AutoModelForCausalLM with tempfile.TemporaryDirectory() as tmpdirname: # saving state dict for now but will save config and other in the future @@ -411,10 +410,9 @@ def test_int8_serialization_offload(self): r""" Test whether it is possible to serialize a model in 8-bit and offload weights to cpu/disk """ - + from bitsandbytes.nn import Int8Params from transformers import AutoConfig, AutoModelForCausalLM - from bitsandbytes.nn import Int8Params with tempfile.TemporaryDirectory() as tmpdirname: # saving state dict for now but will save config and other in the future @@ -473,9 +471,8 @@ def test_int8_serialization_shard(self): r""" Test whether it is possible to serialize a model in 8-bit. """ - from transformers import AutoConfig, AutoModelForCausalLM - from bitsandbytes.nn import Int8Params + from transformers import AutoConfig, AutoModelForCausalLM with tempfile.TemporaryDirectory() as tmpdirname: # saving state dict for now but will save config and other in the future From 4690ec30b8115251904822505bfe4c2906d2f1b2 Mon Sep 17 00:00:00 2001 From: "Lin, Fanli" Date: Fri, 17 Jan 2025 04:30:31 -0800 Subject: [PATCH 2/2] furter fix quality --- tests/test_quantization.py | 11 ++++------- 1 file changed, 4 insertions(+), 7 deletions(-) diff --git a/tests/test_quantization.py b/tests/test_quantization.py index 66fbc13496d..fb3adfd2e0c 100644 --- a/tests/test_quantization.py +++ b/tests/test_quantization.py @@ -134,9 +134,8 @@ def test_llm_skip(self): r""" A simple test to check if `llm_int8_skip_modules` works as expected """ - from transformers import AutoConfig, AutoModelForCausalLM - import bitsandbytes as bnb + from transformers import AutoConfig, AutoModelForCausalLM bnb_quantization_config = BnbQuantizationConfig( load_in_8bit=True, skip_modules=["lm_head", "transformer.word_embeddings"] @@ -201,9 +200,9 @@ def test_fp32_8bit_conversion(self): @require_multi_device def test_cpu_gpu_loading_custom_device_map(self): + from bitsandbytes.nn import Int8Params from transformers import AutoConfig, AutoModelForCausalLM - from bitsandbytes.nn import Int8Params r""" A test to check is dispatching a model on cpu & gpu works correctly using a custom `device_map`. @@ -257,9 +256,8 @@ def test_cpu_gpu_loading_custom_device_map(self): @require_multi_device def test_cpu_gpu_loading_custom_device_map_offload_state_dict(self): - from transformers import AutoConfig, AutoModelForCausalLM - from bitsandbytes.nn import Int8Params + from transformers import AutoConfig, AutoModelForCausalLM r""" A test to check is dispatching a model on cpu & gpu works correctly using a custom `device_map` and offload_state_dict=True. @@ -315,9 +313,8 @@ def test_cpu_gpu_loading_custom_device_map_offload_state_dict(self): @require_multi_device def test_cpu_gpu_disk_loading_custom_device_map_kwargs(self): - from transformers import AutoConfig, AutoModelForCausalLM - from bitsandbytes.nn import Int8Params + from transformers import AutoConfig, AutoModelForCausalLM r""" A test to check is dispatching a model on cpu & gpu works correctly using a custom `device_map`.