titans-pytorch 0.2.23__tar.gz → 0.2.24__tar.gz
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/PKG-INFO +1 -1
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/pyproject.toml +1 -1
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/titans_pytorch/neural_memory.py +2 -3
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/.github/workflows/python-publish.yml +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/.github/workflows/test.yaml +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/.gitignore +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/LICENSE +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/README.md +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/data/README.md +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/data/enwik8.gz +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/fig1.png +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/fig2.png +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/tests/test_titans.py +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/titans_pytorch/__init__.py +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/titans_pytorch/associative_scan.py +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/titans_pytorch/mac_transformer.py +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/titans_pytorch/memory_models.py +0 -0
- {titans_pytorch-0.2.23 → titans_pytorch-0.2.24}/train_mac.py +0 -0
@@ -868,12 +868,11 @@ class NeuralMemory(Module):
|
|
868
868
|
|
869
869
|
# update weights once batch size is fulfilled
|
870
870
|
|
871
|
-
weights = next_neural_mem_state.weights
|
872
|
-
|
873
871
|
last_update, _ = past_state
|
874
872
|
|
875
873
|
if exists(gate):
|
876
|
-
|
874
|
+
curr_weights = next_neural_mem_state.updates.apply(lambda t: t[:, 0])
|
875
|
+
weights = TensorDict({param_name: v1.lerp(v2, gate) for (param_name, v1), (_, v2) in zip(curr_weights.items(), last_update.items())})
|
877
876
|
else:
|
878
877
|
weights = last_update
|
879
878
|
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|
File without changes
|