p04798526
/
LLaMA-Factory-310P3
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

update tests

This commit is contained in:
hiyouga 2024-07-04 04:00:12 +08:00
parent 0c699de39d
commit 636bb9c1e6
1 changed files with 45 additions and 30 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

75
tests/model/model_utils/test_packing.py
View File

@ -12,42 +12,57 @@
# See the License for the specific language governing permissions and
# limitations under the License.
import pytest
import torch
from llamafactory.model.model_utils.packing import get_seqlens_in_batch, get_unpad_data
def test_get_seqlens_in_batch():
attention_mask_with_indices = torch.tensor(
[
[1, 1, 2, 2, 2, 0],
[1, 2, 2, 3, 3, 3],
]
)
@pytest.mark.parametrize(
"attention_mask,golden_seq_lens",
[
(
[
[1, 1, 2, 2, 2, 0],
[1, 2, 2, 3, 3, 3],
],
[2, 3, 1, 2, 3],
),
(
[[1]],
[1],
),
],
)
def test_get_seqlens_in_batch(attention_mask, golden_seq_lens):
attention_mask_with_indices = torch.tensor(attention_mask)
seqlens_in_batch = get_seqlens_in_batch(attention_mask_with_indices)
assert list(seqlens_in_batch.size()) == [5]
assert torch.all(seqlens_in_batch == torch.tensor([2, 3, 1, 2, 3]))
attention_mask_with_indices = torch.tensor([[1, 1, 1]])
seqlens_in_batch = get_seqlens_in_batch(attention_mask_with_indices)
assert list(seqlens_in_batch.size()) == [1]
assert torch.all(seqlens_in_batch == torch.tensor([3]))
assert torch.all(seqlens_in_batch == torch.tensor(golden_seq_lens))
def test_get_unpad_data():
attention_mask_with_indices = torch.tensor(
[
[1, 1, 2, 2, 2, 0],
[1, 2, 2, 3, 3, 3],
]
)
@pytest.mark.parametrize(
"attention_mask,golden_indices,golden_cu_seqlens,golden_max_seqlen",
[
(
[
[1, 1, 2, 2, 2, 0],
[1, 2, 2, 3, 3, 3],
],
[0, 1, 2, 3, 4, 6, 7, 8, 9, 10, 11],
[0, 2, 5, 6, 8, 11],
3,
),
(
[[1]],
[0],
[0, 1],
1,
),
],
)
def test_get_unpad_data(attention_mask, golden_indices, golden_cu_seqlens, golden_max_seqlen):
attention_mask_with_indices = torch.tensor(attention_mask)
indices, cu_seqlens, max_seqlen_in_batch = get_unpad_data(attention_mask_with_indices)
assert torch.all(indices == torch.tensor([0, 1, 2, 3, 4, 6, 7, 8, 9, 10, 11]))
assert torch.all(cu_seqlens == torch.tensor([0, 2, 5, 6, 8, 11], dtype=torch.int32))
assert max_seqlen_in_batch == 3
attention_mask_with_indices = torch.tensor([[1, 1, 1]])
indices, cu_seqlens, max_seqlen_in_batch = get_unpad_data(attention_mask_with_indices)
assert torch.all(indices == torch.tensor([0, 1, 2]))
assert torch.all(cu_seqlens == torch.tensor([0, 3], dtype=torch.int32))
assert max_seqlen_in_batch == 3
assert torch.all(indices == torch.tensor(golden_indices))
assert torch.all(cu_seqlens == torch.tensor(golden_cu_seqlens, dtype=torch.int32))
assert max_seqlen_in_batch == golden_max_seqlen