rxnn 0.1.28__tar.gz → 0.1.29__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.
Files changed (31) hide show
  1. {rxnn-0.1.28 → rxnn-0.1.29}/PKG-INFO +1 -1
  2. {rxnn-0.1.28 → rxnn-0.1.29}/pyproject.toml +1 -1
  3. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/experimental/attention.py +3 -3
  4. {rxnn-0.1.28 → rxnn-0.1.29}/LICENSE +0 -0
  5. {rxnn-0.1.28 → rxnn-0.1.29}/README.md +0 -0
  6. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/__init__.py +0 -0
  7. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/experimental/__init__.py +0 -0
  8. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/experimental/models.py +0 -0
  9. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/experimental/moe.py +0 -0
  10. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/memory/__init__.py +0 -0
  11. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/memory/norm.py +0 -0
  12. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/memory/stm.py +0 -0
  13. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/rxt/__init__.py +0 -0
  14. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/rxt/models.py +0 -0
  15. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/training/__init__.py +0 -0
  16. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/training/base.py +0 -0
  17. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/training/bml.py +0 -0
  18. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/training/callbacks.py +0 -0
  19. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/training/dataset.py +0 -0
  20. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/training/scheduler.py +0 -0
  21. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/training/tokenizer.py +0 -0
  22. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/transformers/__init__.py +0 -0
  23. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/transformers/attention.py +0 -0
  24. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/transformers/ff.py +0 -0
  25. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/transformers/layers.py +0 -0
  26. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/transformers/mask.py +0 -0
  27. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/transformers/models.py +0 -0
  28. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/transformers/moe.py +0 -0
  29. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/transformers/positional.py +0 -0
  30. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/transformers/sampler.py +0 -0
  31. {rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/utils.py +0 -0
{rxnn-0.1.28 → rxnn-0.1.29}/PKG-INFO RENAMED
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.3
2
2
  Name: rxnn
3
- Version: 0.1.28
3
+ Version: 0.1.29
4
4
  Summary: RxNN: Reactive Neural Networks Platform
5
5
  License: Apache-2.0
6
6
  Keywords: deep-learning,ai,machine-learning
{rxnn-0.1.28 → rxnn-0.1.29}/pyproject.toml RENAMED
@@ -4,7 +4,7 @@ build-backend = "poetry.core.masonry.api"
4
4
 
5
5
  [tool.poetry]
6
6
  name = "rxnn"
7
- version = "0.1.28"
7
+ version = "0.1.29"
8
8
  description = "RxNN: Reactive Neural Networks Platform"
9
9
 
10
10
  license = "Apache-2.0"
{rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/experimental/attention.py RENAMED
@@ -330,8 +330,8 @@ class GroupedMoeAttentionVectorized(GroupedQueryAttention):
330
330
 
331
331
  # Compute all experts' projections
332
332
  # Shape: (B*S, num_experts, head_dim)
333
- k_all = torch.einsum('be,ehd->beh', key_flat, self.wk) # [B*S, num_experts, head_dim]
334
- v_all = torch.einsum('be,ehd->beh', value.view(B*S, D), self.wv)
333
+ k_all = torch.einsum('bd,edh->beh', key_flat, self.wk) # [B*S, num_experts, head_dim]
334
+ v_all = torch.einsum('bd,edh->beh', value.view(B*S, D), self.wv)
335
335
 
336
336
  # Reshape to [B, S, num_experts, head_dim]
337
337
  k_all = k_all.view(B, S, self.num_experts, -1)
@@ -450,7 +450,7 @@ class DeepMoeAttentionVectorized(GroupedMoeAttentionVectorized):
450
450
  indices_q = indices_q.view(B, T, self.num_query_groups)
451
451
 
452
452
  # Compute all query experts
453
- q_all = torch.einsum('be,ehd->beh', query_flat, self.wq) # [B*T, num_query_experts, head_dim]
453
+ q_all = torch.einsum('bd,edh->beh', query_flat, self.wq) # [B*T, num_query_experts, head_dim]
454
454
  q_all = q_all.view(B, T, self.num_query_experts, -1)
455
455
 
456
456
  # Gather top-k experts
{rxnn-0.1.28 → rxnn-0.1.29}/LICENSE RENAMED
File without changes
{rxnn-0.1.28 → rxnn-0.1.29}/README.md RENAMED
File without changes
{rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/__init__.py RENAMED
File without changes
{rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/memory/__init__.py RENAMED
File without changes
{rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/memory/norm.py RENAMED
File without changes
{rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/memory/stm.py RENAMED
File without changes
{rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/rxt/__init__.py RENAMED
File without changes
{rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/rxt/models.py RENAMED
File without changes
{rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/training/base.py RENAMED
File without changes
{rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/training/bml.py RENAMED
File without changes
{rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/transformers/ff.py RENAMED
File without changes
{rxnn-0.1.28 → rxnn-0.1.29}/src/rxnn/utils.py RENAMED
File without changes