rxnn 0.1.76__tar.gz → 0.1.78__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-0.1.76 → rxnn-0.1.78}/PKG-INFO +3 -3
  2. {rxnn-0.1.76 → rxnn-0.1.78}/README.md +2 -2
  3. {rxnn-0.1.76 → rxnn-0.1.78}/pyproject.toml +1 -1
  4. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/rxt/models.py +2 -2
  5. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/transformers/positional.py +5 -2
  6. {rxnn-0.1.76 → rxnn-0.1.78}/LICENSE +0 -0
  7. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/__init__.py +0 -0
  8. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/experimental/__init__.py +0 -0
  9. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/experimental/attention.py +0 -0
  10. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/experimental/models.py +0 -0
  11. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/experimental/moe.py +0 -0
  12. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/memory/__init__.py +0 -0
  13. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/memory/norm.py +0 -0
  14. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/memory/stm.py +0 -0
  15. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/rxt/__init__.py +0 -0
  16. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/training/__init__.py +0 -0
  17. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/training/base.py +0 -0
  18. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/training/bml.py +0 -0
  19. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/training/callbacks.py +0 -0
  20. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/training/dataset.py +0 -0
  21. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/training/scheduler.py +0 -0
  22. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/training/tokenizer.py +0 -0
  23. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/transformers/__init__.py +0 -0
  24. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/transformers/attention.py +0 -0
  25. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/transformers/ff.py +0 -0
  26. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/transformers/layers.py +0 -0
  27. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/transformers/mask.py +0 -0
  28. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/transformers/models.py +0 -0
  29. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/transformers/moe.py +0 -0
  30. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/transformers/sampler.py +0 -0
  31. {rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/utils.py +0 -0
{rxnn-0.1.76 → rxnn-0.1.78}/PKG-INFO RENAMED
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.3
2
2
  Name: rxnn
3
- Version: 0.1.76
3
+ Version: 0.1.78
4
4
  Summary: RxNN: Reactive Neural Networks Platform
5
5
  License: Apache-2.0
6
6
  Keywords: deep-learning,ai,machine-learning
@@ -23,8 +23,8 @@ Project-URL: Homepage, https://rxai.dev/rxnn
23
23
  Project-URL: Repository, https://github.com/RxAI-dev/rxnn/python
24
24
  Description-Content-Type: text/markdown
25
25
 
26
- <img src="https://raw.githubusercontent.com/RxAI-dev/RxNN/refs/heads/main/assets/logo_rxai.webp" width="300" />
27
- <img src="https://raw.githubusercontent.com/RxAI-dev/RxNN/refs/heads/main/assets/logo_rxnn.webp" width="300" />
26
+ <img src="https://raw.githubusercontent.com/RxAI-dev/RxNN/refs/heads/main/assets/logo/logo_rxai.webp" width="300" />
27
+ <img src="https://raw.githubusercontent.com/RxAI-dev/RxNN/refs/heads/main/assets/logo/logo_rxnn.webp" width="300" />
28
28
 
29
29
  # Reactive AI - RxNN
30
30
  ## Reactive Neural Networks Platform
{rxnn-0.1.76 → rxnn-0.1.78}/README.md RENAMED
@@ -1,5 +1,5 @@
1
- <img src="https://raw.githubusercontent.com/RxAI-dev/RxNN/refs/heads/main/assets/logo_rxai.webp" width="300" />
2
- <img src="https://raw.githubusercontent.com/RxAI-dev/RxNN/refs/heads/main/assets/logo_rxnn.webp" width="300" />
1
+ <img src="https://raw.githubusercontent.com/RxAI-dev/RxNN/refs/heads/main/assets/logo/logo_rxai.webp" width="300" />
2
+ <img src="https://raw.githubusercontent.com/RxAI-dev/RxNN/refs/heads/main/assets/logo/logo_rxnn.webp" width="300" />
3
3
 
4
4
  # Reactive AI - RxNN
5
5
  ## Reactive Neural Networks Platform
{rxnn-0.1.76 → rxnn-0.1.78}/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 = "0.1.76"
7
+ version = "0.1.78"
8
8
  description = "RxNN: Reactive Neural Networks Platform"
9
9
 
10
10
  license = "Apache-2.0"
{rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/rxt/models.py RENAMED
@@ -97,13 +97,13 @@ class RxTAlphaComponentBase(nn.Module, PyTorchModelHubMixin):
97
97
  if cross_att_type in ['mha', 'gqa', 'mqa']:
98
98
  cross_att_init = lambda: init_attention(embed_dim, att_heads, cross_att_type, att_groups, rope=rope,
99
99
  use_flash_attention=use_flash_attention, dropout=att_dropout,
100
- max_seq_len=seq_len, is_causal=is_causal)
100
+ max_seq_len=seq_len, is_causal=is_causal, rope_only_for_query=True)
101
101
  else:
102
102
  cross_att_init = lambda: init_experimental_attention(embed_dim, att_heads, cross_att_type, cross_att_groups or att_groups, rope=rope,
103
103
  use_flash_attention=use_flash_attention, dropout=att_dropout,
104
104
  max_seq_len=seq_len, is_causal=is_causal, num_experts=att_experts,
105
105
  num_query_experts=att_query_experts,
106
- num_query_groups=cross_att_query_groups or att_query_groups)
106
+ num_query_groups=cross_att_query_groups or att_query_groups, rope_only_for_query=True)
107
107
 
108
108
  layers = nn.ModuleList([
109
109
  ReactiveTransformerLayer(
{rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/transformers/positional.py RENAMED
@@ -40,13 +40,16 @@ class RotaryPositionalEmbedding(nn.Module):
40
40
  return q_embed
41
41
 
42
42
  def _prepare_freqs(self, seq_len: int, device: torch.device) -> torch.Tensor:
43
- if self.cache is None or self.cache.size(1) < seq_len:
43
+ cache_len = self.cache.size(1)
44
+ if self.cache is None or cache_len < seq_len:
44
45
  t = torch.arange(seq_len, device=device).type_as(self.inv_freq)
45
46
  freqs = torch.einsum('i,j->ij', t, self.inv_freq)
46
47
  self.cache = freqs
47
48
  return freqs[None, None, :, :]
48
- else:
49
+ elif cache_len == seq_len:
49
50
  return self.cache[None, None, :, :]
51
+ else:
52
+ return self.cache[:seq_len][None, None, :, :]
50
53
 
51
54
  def _rotate(self, x: torch.Tensor, freqs: torch.Tensor) -> torch.Tensor:
52
55
  x1 = x[..., 0::2]
{rxnn-0.1.76 → rxnn-0.1.78}/LICENSE RENAMED
File without changes
{rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/__init__.py RENAMED
File without changes
{rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/memory/__init__.py RENAMED
File without changes
{rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/memory/norm.py RENAMED
File without changes
{rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/memory/stm.py RENAMED
File without changes
{rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/rxt/__init__.py RENAMED
File without changes
{rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/training/base.py RENAMED
File without changes
{rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/training/bml.py RENAMED
File without changes
{rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/transformers/ff.py RENAMED
File without changes
{rxnn-0.1.76 → rxnn-0.1.78}/src/rxnn/utils.py RENAMED
File without changes