-
MobilenetV2 model running on device implemented in TTNN
-
I got feedback that someone missed the playback speed controls of the mobile app. Is this something that can be designed and implemented?
-
![screenshot from 2018-06-28 19-07-42](https://user-images.githubusercontent.com/25350960/42046672-0e8c5364-7b07-11e8-9f26-dd972c459b31.png)
In this formula we have an expectation of $w_i$. That me…
-
### 🚀 The feature, motivation and pitch
The current PowerSGD implementation iterate over all gradient tensors, Ps, Qs, and operate on single tensor as follows:
```python
# Compute Ps.
for tens…
-
May I ask the author, how should I choose different numbers of visual tokens to experiment with the impact of different visual tokens on the results? Is it setting a different number of clusters? As i…
-
output = torch.bmm(affinity, w_v)
# output=output.permute(0,2,1)
# output=self.transformer1_FFN[1](nn.Dropout(0.2)(F.relu(((self.transformer1_FFN[0](output))))))
# …
-
## 📚 Documentation
Currently the documentation of [`torch.tensordot`][1] is quite comprehensible.
However, I came across several instances recently (e.g., [a][2], [b][3], [c][4]) where users tried t…
-
## 🐛 Issue Description
I have a snippet of code that works when I use ``np.einsum``. But it's PyTorch equivalent throws an error for the conditional mask ``A==1.0``. Is this a bug or is this behavi…
-
D1 a c dio_model
.model dio_model D (LEVEL=3 IS=1.e-12)
```
.model causes issue
```
check if diode level=3 is implemented
-
It seems that the way to calculate attention weight is different from origin paper: softmax(v* tanh(W*[s,h])), relu are used after softmax here, can you give some reasons or reference?
` …