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

switch to device agnostic device calling for test cases (#38247)

Browse Source 
* use device agnostic APIs in test cases

Signed-off-by: Matrix Yao <matrix.yao@intel.com>

* fix style

Signed-off-by: Matrix Yao <matrix.yao@intel.com>

* add one more

Signed-off-by: YAO Matrix <matrix.yao@intel.com>

* xpu now supports integer device id, aligning to CUDA behaviors

Signed-off-by: Matrix Yao <matrix.yao@intel.com>

* update to use device_properties

Signed-off-by: Matrix Yao <matrix.yao@intel.com>

* fix style

Signed-off-by: Matrix Yao <matrix.yao@intel.com>

* update comment

Signed-off-by: Matrix Yao <matrix.yao@intel.com>

* fix comments

Signed-off-by: Matrix Yao <matrix.yao@intel.com>

* fix style

Signed-off-by: Matrix Yao <matrix.yao@intel.com>

---------

Signed-off-by: Matrix Yao <matrix.yao@intel.com>
Signed-off-by: YAO Matrix <matrix.yao@intel.com>
Co-authored-by: ydshieh <ydshieh@users.noreply.github.com>
This commit is contained in:
Yao Matrix
2025-05-26 16:18:53 +08:00
committed by GitHub
parent cba279f46c
commit a5a0c7b888
39 changed files with 259 additions and 389 deletions
Download Patch File Download Diff File Expand all files Collapse all files

42
tests/models/gemma/test_modeling_gemma.py
View File

@@ -21,7 +21,9 @@ from packaging import version
from transformers import AutoModelForCausalLM, AutoTokenizer, GemmaConfig, is_torch_available
from transformers.generation.configuration_utils import GenerationConfig
from transformers.testing_utils import (
Expectations,
cleanup,
get_device_properties,
require_bitsandbytes,
require_flash_attn,
require_read_token,
@@ -105,15 +107,13 @@ class GemmaModelTest(CausalLMModelTest, unittest.TestCase):
@require_torch_accelerator
class GemmaIntegrationTest(unittest.TestCase):
input_text = ["Hello I am doing", "Hi today"]
# This variable is used to determine which CUDA device are we using for our runners (A10 or T4)
# This variable is used to determine which accelerator are we using for our runners (e.g. A10 or T4)
# Depending on the hardware we get different logits / generations
cuda_compute_capability_major_version = None
device_properties = None
@classmethod
def setUpClass(cls):
if is_torch_available() and torch.cuda.is_available():
# 8 is for A100 / A10 and 7 for T4
cls.cuda_compute_capability_major_version = torch.cuda.get_device_capability()[0]
cls.device_properties = get_device_properties()
def tearDown(self):
# See LlamaIntegrationTest.tearDown(). Can be removed once LlamaIntegrationTest.tearDown() is removed.
@@ -270,7 +270,7 @@ class GemmaIntegrationTest(unittest.TestCase):
@require_read_token
def test_model_7b_fp16(self):
if self.cuda_compute_capability_major_version == 7:
if self.device_properties == ("cuda", 7):
self.skipTest("This test is failing (`torch.compile` fails) on Nvidia T4 GPU (OOM).")
model_id = "google/gemma-7b"
@@ -293,7 +293,7 @@ class GemmaIntegrationTest(unittest.TestCase):
@require_read_token
def test_model_7b_bf16(self):
if self.cuda_compute_capability_major_version == 7:
if self.device_properties == ("cuda", 7):
self.skipTest("This test is failing (`torch.compile` fails) on Nvidia T4 GPU (OOM).")
model_id = "google/gemma-7b"
@@ -302,20 +302,16 @@ class GemmaIntegrationTest(unittest.TestCase):
#
# Note: Key 9 is currently set for MI300, but may need potential future adjustments for H100s,
# considering differences in hardware processing and potential deviations in generated text.
EXPECTED_TEXTS = {
7: [
"""Hello I am doing a project on a 1991 240sx and I am trying to find""",
"Hi today I am going to show you how to make a very simple and easy to make a very simple and",
],
8: [
"Hello I am doing a project for my school and I am trying to make a program that will read a .txt file",
"Hi today I am going to show you how to make a very simple and easy to make a very simple and",
],
9: [
"Hello I am doing a project for my school and I am trying to get a servo to move a certain amount of degrees",
"Hi today I am going to show you how to make a very simple and easy to make DIY light up sign",
],
}
# fmt: off
EXPECTED_TEXTS = Expectations(
{
("cuda", 7): ["""Hello I am doing a project on a 1991 240sx and I am trying to find""", "Hi today I am going to show you how to make a very simple and easy to make a very simple and",],
("cuda", 8): ["Hello I am doing a project for my school and I am trying to make a program that will read a .txt file", "Hi today I am going to show you how to make a very simple and easy to make a very simple and",],
("rocm", 9): ["Hello I am doing a project for my school and I am trying to get a servo to move a certain amount of degrees", "Hi today I am going to show you how to make a very simple and easy to make DIY light up sign",],
}
)
# fmt: on
expected_text = EXPECTED_TEXTS.get_expectation()
model = AutoModelForCausalLM.from_pretrained(model_id, low_cpu_mem_usage=True, torch_dtype=torch.bfloat16).to(
torch_device
@@ -326,11 +322,11 @@ class GemmaIntegrationTest(unittest.TestCase):
output = model.generate(**inputs, max_new_tokens=20, do_sample=False)
output_text = tokenizer.batch_decode(output, skip_special_tokens=True)
self.assertEqual(output_text, EXPECTED_TEXTS[self.cuda_compute_capability_major_version])
self.assertEqual(output_text, expected_text)
@require_read_token
def test_model_7b_fp16_static_cache(self):
if self.cuda_compute_capability_major_version == 7:
if self.device_properties == ("cuda", 7):
self.skipTest("This test is failing (`torch.compile` fails) on Nvidia T4 GPU (OOM).")
model_id = "google/gemma-7b"