rxnn 0.1.38__tar.gz → 0.1.39__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.
- {rxnn-0.1.38 → rxnn-0.1.39}/PKG-INFO +1 -1
- {rxnn-0.1.38 → rxnn-0.1.39}/pyproject.toml +1 -1
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/transformers/attention.py +3 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/LICENSE +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/README.md +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/__init__.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/experimental/__init__.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/experimental/attention.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/experimental/models.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/experimental/moe.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/memory/__init__.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/memory/norm.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/memory/stm.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/rxt/__init__.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/rxt/models.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/training/__init__.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/training/base.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/training/bml.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/training/callbacks.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/training/dataset.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/training/scheduler.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/training/tokenizer.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/transformers/__init__.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/transformers/ff.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/transformers/layers.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/transformers/mask.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/transformers/models.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/transformers/moe.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/transformers/positional.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/transformers/sampler.py +0 -0
- {rxnn-0.1.38 → rxnn-0.1.39}/src/rxnn/utils.py +0 -0
|
@@ -106,6 +106,9 @@ class MultiHeadAttention(nn.Module):
|
|
|
106
106
|
# with sdpa_kernel(backends=[SDPBackend.FLASH_ATTENTION]):
|
|
107
107
|
# return self._torch_attention(q, k, v, b, t, d, mask=mask, enable_gqa=enable_gqa)
|
|
108
108
|
from flash_attn import flash_attn_func
|
|
109
|
+
print(q.size(), q.dtype)
|
|
110
|
+
print(k.size(), k.dtype)
|
|
111
|
+
print(v.size(), v.dtype)
|
|
109
112
|
attn_output = flash_attn_func(q, k, v, dropout_p=self.dropout.p if self.training else 0.0, causal=self.is_causal)
|
|
110
113
|
return self._transpose_output(attn_output, b, t, d)
|
|
111
114
|
|
|
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
|
|
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
|