Skip to content

Commit 13b05a9

Browse files
romitjainromitjain
authored
fix: Enabled GPU based unit tests (#649)
* Updated tox and pyproject Signed-off-by: romitjain <romit@ibm.com> * Updates Signed-off-by: romitjain <romit@ibm.com> * Updated ci/cd Signed-off-by: romit <romit@ibm.com> * Added monkey patching Signed-off-by: romit <romit@ibm.com> * Added copy for global args Signed-off-by: romit <romit@ibm.com> * Linting fixes Signed-off-by: romit <romit@ibm.com> * Fixed errors Signed-off-by: romit <romit@ibm.com> --------- Signed-off-by: romitjain <romit@ibm.com> Signed-off-by: romit <romit@ibm.com>
1 parent 890725c commit 13b05a9

16 files changed

+392
-196
lines changed

‎pyproject.toml‎

Lines changed: 8 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -52,6 +52,14 @@ fms-accel = ["fms-acceleration>=0.6.2"]
5252
gptq-dev = ["auto_gptq>0.4.2", "optimum>=1.15.0"]
5353
mamba = ["mamba_ssm[causal-conv1d]>=2.0.0,<3.0.0"]
5454
scanner-dev = ["HFResourceScanner>=0.1.0"]
55+
fms-accel-all = [
56+
"fms-acceleration>=0.6.2",
57+
"fms-acceleration-peft",
58+
"fms-acceleration-foak",
59+
"fms-acceleration-aadp",
60+
"fms-acceleration-moe",
61+
"fms-acceleration-odm"
62+
]
5563

5664
[tool.setuptools.packages.find]
5765
exclude = ["tests", "tests.*"]

‎tests/acceleration/test_acceleration_framework.py‎

Lines changed: 16 additions & 17 deletions
Original file line numberDiff line numberDiff line change
@@ -263,10 +263,10 @@ def test_framework_raises_if_used_with_missing_package():
263263
ValueError, match="No acceleration framework package found."
264264
):
265265
sft_trainer.train(
266-
MODEL_ARGS,
267-
DATA_ARGS,
268-
TRAIN_ARGS,
269-
PEFT_LORA_ARGS,
266+
copy.deepcopy(MODEL_ARGS),
267+
copy.deepcopy(DATA_ARGS),
268+
copy.deepcopy(TRAIN_ARGS),
269+
copy.deepcopy(PEFT_LORA_ARGS),
270270
quantized_lora_config=quantized_lora_config,
271271
)
272272

@@ -320,9 +320,9 @@ def test_framework_raises_due_to_invalid_arguments(
320320
with pytest.raises(exception, match=exception_msg):
321321
sft_trainer.train(
322322
model_args,
323-
DATA_ARGS,
323+
copy.deepcopy(DATA_ARGS),
324324
train_args,
325-
peft_config,
325+
copy.deepcopy(peft_config),
326326
quantized_lora_config=quantized_lora_config,
327327
)
328328

@@ -379,7 +379,7 @@ def test_framework_initialized_properly_peft(
379379
train_args = copy.deepcopy(TRAIN_ARGS)
380380
train_args.output_dir = tempdir
381381
train_args.save_strategy = "no"
382-
train_args.fp16 = True
382+
train_args.bf16 = True
383383
peft_args = copy.deepcopy(PEFT_LORA_ARGS)
384384
peft_args.target_modules = ["q_proj", "k_proj"]
385385

@@ -395,7 +395,7 @@ def test_framework_initialized_properly_peft(
395395
with instantiate_model_patcher():
396396
sft_trainer.train(
397397
model_args,
398-
DATA_ARGS,
398+
copy.deepcopy(DATA_ARGS),
399399
train_args,
400400
peft_args,
401401
quantized_lora_config=quantized_lora_config,
@@ -430,7 +430,7 @@ def test_framework_initialized_properly_foak():
430430
train_args = copy.deepcopy(TRAIN_ARGS)
431431
train_args.output_dir = tempdir
432432
train_args.save_strategy = "no"
433-
train_args.fp16 = True
433+
train_args.bf16 = True
434434
peft_args = copy.deepcopy(PEFT_LORA_ARGS)
435435
peft_args.target_modules = ["q_proj", "k_proj"]
436436

@@ -465,7 +465,7 @@ def test_framework_initialized_properly_foak():
465465
with instantiate_model_patcher():
466466
sft_trainer.train(
467467
model_args,
468-
DATA_ARGS,
468+
copy.deepcopy(DATA_ARGS),
469469
train_args,
470470
peft_args,
471471
quantized_lora_config=quantized_lora_config,
@@ -613,8 +613,8 @@ def test_error_raised_with_paddingfree_and_flash_attn_disabled():
613613
model_args.use_flash_attn = False
614614
sft_trainer.train(
615615
model_args,
616-
DATA_ARGS,
617-
TRAIN_ARGS,
616+
copy.deepcopy(DATA_ARGS),
617+
copy.deepcopy(TRAIN_ARGS),
618618
attention_and_distributed_packing_config=attention_and_distributed_packing_config,
619619
)
620620

@@ -637,8 +637,8 @@ def test_error_raised_with_multipack_and_paddingfree_disabled():
637637
model_args = copy.deepcopy(MODEL_ARGS)
638638
sft_trainer.train(
639639
model_args,
640-
DATA_ARGS,
641-
TRAIN_ARGS,
640+
copy.deepcopy(DATA_ARGS),
641+
copy.deepcopy(TRAIN_ARGS),
642642
attention_and_distributed_packing_config=attention_and_distributed_packing_config,
643643
)
644644

@@ -664,7 +664,7 @@ def test_error_raised_with_packing_and_paddingfree_enabled():
664664
train_args.packing = True
665665
sft_trainer.train(
666666
model_args,
667-
DATA_ARGS,
667+
copy.deepcopy(DATA_ARGS),
668668
train_args,
669669
attention_and_distributed_packing_config=attention_and_distributed_packing_config,
670670
)
@@ -693,7 +693,6 @@ def test_error_raised_with_fused_lora_enabled_without_quantized_argument():
693693
train_args = copy.deepcopy(TRAIN_ARGS)
694694
train_args.output_dir = tempdir
695695
train_args.save_strategy = "no"
696-
train_args.fp16 = True
697696
peft_args = copy.deepcopy(PEFT_LORA_ARGS)
698697
peft_args.target_modules = ["q_proj", "k_proj"]
699698

@@ -713,7 +712,7 @@ def test_error_raised_with_fused_lora_enabled_without_quantized_argument():
713712
with instantiate_model_patcher():
714713
sft_trainer.train(
715714
model_args,
716-
DATA_ARGS,
715+
copy.deepcopy(DATA_ARGS),
717716
train_args,
718717
peft_args,
719718
quantized_lora_config=None,

‎tests/artifacts/language_models/__init__.py‎

Lines changed: 1 addition & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -20,3 +20,4 @@
2020
### Constants used for model path
2121
PREDEFINED_MODEL_PATH = os.path.join(os.path.dirname(__file__))
2222
MAYKEYE_TINY_LLAMA_CACHED = os.path.join(PREDEFINED_MODEL_PATH, "maykeye-tinyllama-v0")
23+
TINYMIXTRAL_MOE = "Isotonic/TinyMixtral-4x248M-MoE"

0 commit comments

Comments
 (0)