Ryan Tremblay
Ryan Tremblay
Hello, I'm interested in migrating to this from Gifted-Chat. How do you handle long chat histories? Do you just load every document whenever the chat is opened or do you...
I swapped out the Torch attention function for Flash Attention 2 in the MusicGen project [here](https://github.com/facebookresearch/audiocraft/blob/6b23f77215dd5bd7cc6628e447294cc477b10cb7/audiocraft/modules/transformer.py#L401) like so: ```python if self.memory_efficient: p = self.dropout if self.training else 0 if _efficient_attention_backend...
When doing the following: ```py from vocode.turn_based.synthesizer import AnyNonGoogleSynthesizer ``` The following error occurs when the optional dependency GTTS isn't installed since it's imported by the init file: ``` Traceback...
Your paper says "Like prior work, we use multi-scale (MSD) and multi-period waveform discriminators (MPD) which lead to improved audio fidelity." However, by default it appears the trainer has an...
Is there a way to train/use the network with high-reverb samples at 48khz? The audio quality difference between 16khz and even 36khz is substantial for speech.
I'm training a Vocos decoder for my DAC autoencoder. When I set hop length = 256 and n_fft = 1024 in the iSTFT head the discriminators quickly win within 1000...
Hello, why did you switch back to building on DAC-style waveform outputs for SNAC after showing it's possible to completely get rid of aliasing by generating complex spectrograms with Vocos?...
Hi! 👋 Firstly, thanks for your work on this project! 🙂 Today I used [patch-package](https://github.com/ds300/patch-package) to patch `[email protected]` for the project I'm working on. I was getting an error ```...
Hi, can you explain the difference between the subband and duration experiments and share which you've found to perform better? Also, what have you discovered about the use of classifier...
### Description & Motivation https://github.com/Lightning-AI/pytorch-lightning/blob/f6fd046552a1504023cb3386a8a0df418a810e4f/src/lightning/fabric/plugins/precision/fsdp.py#L61 What if I want to use fp16 true, but with a loss scaler? This is closer to DeepSpeed's default settings. With FSDP, 16-true, no loss...