def test_hidden_states_transformers(self):
cuda_available = torch.cuda.is_available()
model = BloomModel.from_pretrained(self.path_bigscience_model,
use_cache=False,
torch_dtype="auto").to(torch_device)
model.eval()
# fmt: off
EXAMPLE_IDS = [
3478, 368, 109586, 35433, 2, 77, 132619, 3478, 368, 109586, 35433,
2, 2175, 23714, 73173, 144252, 2, 77, 132619, 3478
]
# fmt: on
MEAN_VALUE_LAST_LM = -4.3392181396484375e-05
MIN_MAX_DICT = {"min": -2.0625, "max": 2.75}
tensor_ids = torch.LongTensor([EXAMPLE_IDS])
with torch.no_grad():
logits = model(tensor_ids.to(torch_device))
output_dict = {
"min": logits.last_hidden_state.min(dim=-1).values[0][0].item(),
"max": logits.last_hidden_state.max(dim=-1).values[0][0].item(),
}
if cuda_available:
self.assertAlmostEqual(MEAN_VALUE_LAST_LM,
logits.last_hidden_state.mean().item(),
places=4)
else:
self.assertAlmostEqual(MEAN_VALUE_LAST_LM,
logits.last_hidden_state.mean().item(),
places=3)
self.assertDictEqual(MIN_MAX_DICT, output_dict)
def test_model_from_pretrained(self):
for model_name in BLOOM_PRETRAINED_MODEL_ARCHIVE_LIST[:1]:
model = BloomModel.from_pretrained(model_name)
self.assertIsNotNone(model)