Nice work, that throughput is wild.
How are you handling the on device speech pipeline, especially around model size, latency, and accuracy tradeoffs on consumer hardware?
[dead]
This is a nice idea. A tiny implementation can be way more useful for learning than yet another wrapper around a big model, especially if it keeps the training loop and inference path small enough to read end to end.
curious and wondering its porting and usage on edge devices?
Nice work, that throughput is wild.
How are you handling the on device speech pipeline, especially around model size, latency, and accuracy tradeoffs on consumer hardware?
[dead]
[dead]
[dead]
[dead]
This is a nice idea. A tiny implementation can be way more useful for learning than yet another wrapper around a big model, especially if it keeps the training loop and inference path small enough to read end to end.
[dead]
[dead]
curious and wondering its porting and usage on edge devices?