vector-quantize-pytorch 1.22.0__tar.gz → 1.22.2__tar.gz

{vector_quantize_pytorch-1.22.0 → vector_quantize_pytorch-1.22.2}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: vector-quantize-pytorch
-Version: 1.22.0
+Version: 1.22.2
 Summary: Vector Quantization - Pytorch
 Project-URL: Homepage, https://pypi.org/project/vector-quantize-pytorch/
 Project-URL: Repository, https://github.com/lucidrains/vector-quantizer-pytorch

{vector_quantize_pytorch-1.22.0 → vector_quantize_pytorch-1.22.2}/pyproject.toml

@@ -1,6 +1,6 @@
 [project]
 name = "vector-quantize-pytorch"
-version = "1.22.0"
+version = "1.22.2"
 description = "Vector Quantization - Pytorch"
 authors = [
     { name = "Phil Wang", email = "lucidrains@gmail.com" }

{vector_quantize_pytorch-1.22.0 → vector_quantize_pytorch-1.22.2}/tests/test_readme.py

@@ -71,8 +71,7 @@ def test_vq_mask():
     assert (mask_quantized[:, 512:] == 0.).all()
     assert (mask_indices[:, 512:] == -1).all()
 
-@pytest.mark.parametrize('implicit_neural_codebook', (True, False))
-@pytest.mark.parametrize('use_cosine_sim', (True, False))
+@pytest.mark.parametrize('implicit_neural_codebook, use_cosine_sim', ((True, False), (False, True), (False, False)))
 @pytest.mark.parametrize('train', (True, False))
 @pytest.mark.parametrize('shared_codebook', (True, False))
 def test_residual_vq(

{vector_quantize_pytorch-1.22.0 → vector_quantize_pytorch-1.22.2}/vector_quantize_pytorch/finite_scalar_quantization.py

@@ -158,7 +158,7 @@ class FSQ(Module):
             torch.full((batch,), noise_dropout, device = device)
         ).bool()
 
-        quantized = torch.where(quantize_mask, unquantized, quantized)
+        quantized = einx.where('b, b ..., b ...', quantize_mask, unquantized, quantized)
 
         # determine where to add a random offset elementwise
 

{vector_quantize_pytorch-1.22.0 → vector_quantize_pytorch-1.22.2}/vector_quantize_pytorch/vector_quantize_pytorch.py

@@ -873,6 +873,7 @@ class VectorQuantize(Module):
         self.commitment_weight = commitment_weight
         self.commitment_use_cross_entropy_loss = commitment_use_cross_entropy_loss # whether to use cross entropy loss to codebook as commitment loss
 
+        assert not (use_cosine_sim and learnable_codebook), 'cosine sim distance codebook not compatible with learnable codebook yet'
         self.learnable_codebook = learnable_codebook
 
         has_codebook_orthogonal_loss = orthogonal_reg_weight > 0.