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

extend more trainer test cases to XPU, all pass (#39652)

Browse Source 
extend more trainer test cases to XPU

Signed-off-by: Yao, Matrix <matrix.yao@intel.com>
This commit is contained in:
Yao Matrix
2025-07-29 01:51:00 -07:00
committed by GitHub
parent 75794792ad
commit f3598a95c7
1 changed files with 8 additions and 8 deletions
Download Patch File Download Diff File Expand all files Collapse all files

16
tests/trainer/test_trainer.py
View File

@@ -2520,7 +2520,7 @@ class TrainerIntegrationTest(TestCasePlus, TrainerIntegrationCommon):
self.assertTrue(len(decreasing_lrs) > len(increasing_lrs)) self.assertTrue(len(decreasing_lrs) > len(increasing_lrs))
@require_torch_optimi @require_torch_optimi
@require_torch_gpu @require_torch_accelerator
def test_stable_adamw(self): def test_stable_adamw(self):
config = LlamaConfig(vocab_size=100, hidden_size=32, num_hidden_layers=3, num_attention_heads=4) config = LlamaConfig(vocab_size=100, hidden_size=32, num_hidden_layers=3, num_attention_heads=4)
tiny_llama = LlamaForCausalLM(config) tiny_llama = LlamaForCausalLM(config)
@@ -2539,7 +2539,7 @@ class TrainerIntegrationTest(TestCasePlus, TrainerIntegrationCommon):
_ = trainer.train() _ = trainer.train()
@require_torch_optimi @require_torch_optimi
@require_torch_gpu @require_torch_accelerator
def test_stable_adamw_extra_args(self): def test_stable_adamw_extra_args(self):
config = LlamaConfig(vocab_size=100, hidden_size=32, num_hidden_layers=3, num_attention_heads=4) config = LlamaConfig(vocab_size=100, hidden_size=32, num_hidden_layers=3, num_attention_heads=4)
tiny_llama = LlamaForCausalLM(config) tiny_llama = LlamaForCausalLM(config)
@@ -2561,7 +2561,7 @@ class TrainerIntegrationTest(TestCasePlus, TrainerIntegrationCommon):
_ = trainer.train() _ = trainer.train()
@require_torch_optimi @require_torch_optimi
@require_torch_gpu @require_torch_accelerator
def test_stable_adamw_lr_display_without_scheduler(self): def test_stable_adamw_lr_display_without_scheduler(self):
config = LlamaConfig(vocab_size=100, hidden_size=32, num_hidden_layers=3, num_attention_heads=4) config = LlamaConfig(vocab_size=100, hidden_size=32, num_hidden_layers=3, num_attention_heads=4)
tiny_llama = LlamaForCausalLM(config) tiny_llama = LlamaForCausalLM(config)
@@ -2586,7 +2586,7 @@ class TrainerIntegrationTest(TestCasePlus, TrainerIntegrationCommon):
self.assertEqual(trainer.get_learning_rates(), [learning_rate, learning_rate]) self.assertEqual(trainer.get_learning_rates(), [learning_rate, learning_rate])
@require_torch_optimi @require_torch_optimi
@require_torch_gpu @require_torch_accelerator
def test_stable_adamw_lr_display_with_scheduler(self): def test_stable_adamw_lr_display_with_scheduler(self):
config = LlamaConfig(vocab_size=100, hidden_size=32, num_hidden_layers=3, num_attention_heads=4) config = LlamaConfig(vocab_size=100, hidden_size=32, num_hidden_layers=3, num_attention_heads=4)
tiny_llama = LlamaForCausalLM(config) tiny_llama = LlamaForCausalLM(config)
@@ -2615,19 +2615,19 @@ class TrainerIntegrationTest(TestCasePlus, TrainerIntegrationCommon):
logs = trainer.state.log_history[1:][:-1] logs = trainer.state.log_history[1:][:-1]
# reach given learning rate peak and end with 0 lr # reach given learning rate peak and end with 0 lr
self.assertTrue(logs[num_warmup_steps - 2]["learning_rate"] == learning_rate) self.assertTrue(logs[num_warmup_steps - 1]["learning_rate"] == learning_rate)
self.assertTrue(logs[-1]["learning_rate"] == 0) self.assertTrue(np.allclose(logs[-1]["learning_rate"], 0, atol=5e-6))
# increasing and decreasing pattern of lrs # increasing and decreasing pattern of lrs
increasing_lrs = [ increasing_lrs = [
logs[i]["learning_rate"] < logs[i + 1]["learning_rate"] logs[i]["learning_rate"] < logs[i + 1]["learning_rate"]
for i in range(len(logs)) for i in range(len(logs))
if i < num_warmup_steps - 2 if i < num_warmup_steps - 1
] ]
decreasing_lrs = [ decreasing_lrs = [
logs[i]["learning_rate"] > logs[i + 1]["learning_rate"] logs[i]["learning_rate"] > logs[i + 1]["learning_rate"]
for i in range(len(logs) - 1) for i in range(len(logs) - 1)
if i >= num_warmup_steps - 2 if i >= num_warmup_steps - 1
] ]
self.assertTrue(all(increasing_lrs)) self.assertTrue(all(increasing_lrs))