Sharvil Nanavati

icon indicating a website link https://www.linkedin.com/in/sharvil-nanavati/ icon indicating an email [email protected]

icon company @lmnt-com icon location Mountain View, CA

Results 54 comments of Sharvil Nanavati

Last frame is centered on end?

Yes, an empty half frame is [appended to the end as well](https://github.com/marl/crepe/blob/master/crepe/core.py#L198).

Code for evaluation in paper

There's no implementation of these metrics in this repository. A code contribution here would be really helpful!

Support for pytorch to tensorflow model conversion

I'm happy to take community contributions for this request, but I don't intend to support TensorFlow myself.

CUDA Error when loading checkpoint on more than one GPU

Hmm I haven't run across that error before. Sorry, I don't think I'll be of much help here.

Not understand plus_in = inv_stdv * (centered_x + 1./255.) in discretized_mix_logistic_loss

I agree with @blues-lin; the expression should be +/-0.5 / 255. Shifting and changing the scale of the input doesn't change the number of discretized buckets. The probability of x...

Install on pip on systems without cuda

Is the issue that you have machines with no CUDA-capable GPUs in them, or that you're unable to install the CUDA toolkit on those machines? I imagine it's the former...

Install on pip on systems without cuda

Right, so this seems like a build issue – `haste` build against CUDA which isn't installed on your target system. Would you be able to install the CUDA toolkit on...

Install on pip on systems without cuda

@Zadagu that sounds like a reasonable request. Out of curiosity, what is your primary reason for using Haste even without CUDA? Is it because regularization is built-in to many of...

layer_norm_gru_cell

I'm happy to accept patches that fix issues with TF compatibility. In this particular case, it seems that the TF team broke backwards compatibility in a minor release.

Can Bidirectional Rnn and multi-layer Rnn be supported?

I have no plans to add multi-layer fusion but I'd very much welcome community contributions. Of course, you can still use bidirectional and multi-layer RNNs with Haste -- you just...