State of the union update

[tikikun]

This thread will update the status of training models

[0xSage]

is strong?

[tikikun]

1 Jul status update

We are training our first model

Decisions

• We use encodec for tokenizing the sound over other VQ models due to its low channel nature in 1.5 bit Encodec
• We synthesize the audio using whisperspeech model
• We ake use of Adam-mini to overcome VRAM shortage Adam mini

Status

• Published an audio to chat dataset (will organize and post here soon enough)

Loss status

[tikikun]