| import unittest | |
| try: | |
| import torch | |
| except ImportError: # pragma: no cover - local lightweight environment | |
| torch = None | |
| class DequantizeTests(unittest.TestCase): | |
| def test_block_scales_are_folded_into_bfloat16_weight(self): | |
| from deepseek_vision.dequantize import dequantize_block_weight | |
| weight = torch.ones((3, 5), dtype=torch.float8_e4m3fn) | |
| scale = torch.tensor([[2.0, 4.0], [8.0, 16.0]]) | |
| result = dequantize_block_weight(weight, scale, block_size=(2, 3)) | |
| expected = torch.tensor( | |
| [ | |
| [2, 2, 2, 4, 4], | |
| [2, 2, 2, 4, 4], | |
| [8, 8, 8, 16, 16], | |
| ], | |
| dtype=torch.bfloat16, | |
| ) | |
| self.assertEqual(result.dtype, torch.bfloat16) | |
| self.assertTrue(torch.equal(result, expected)) | |
| if __name__ == "__main__": | |
| unittest.main() | |
Xet Storage Details
- Size:
- 928 Bytes
- Xet hash:
- 3ce65ccb752e6a1ad3c22f03943f534d294e446351af7cef2cd38d77e0d8db09
·
Xet efficiently stores files, intelligently splitting them into unique chunks and accelerating uploads and downloads. More info.