Describe the bug
On v0.3.2 with the Home-3B-v3.f16.gguf model on a freshly configured integration, the voice assistant says "endoftext" after every response
Expected behavior
The voice assistant should not send extraneous control data to the TTS engine
Logs
Here is the log line showing what is generated by the llama engine (this is running inside of LocalAI
And here is the raw data from the voice assistant debug showing a short interaction turning off a light where you can see that "<|endoftext|>" is included in the input to the TTS engine:
Describe the bug
On v0.3.2 with the Home-3B-v3.f16.gguf model on a freshly configured integration, the voice assistant says "endoftext" after every response
Expected behavior
The voice assistant should not send extraneous control data to the TTS engine
Logs
Here is the log line showing what is generated by the llama engine (this is running inside of LocalAI
And here is the raw data from the voice assistant debug showing a short interaction turning off a light where you can see that "<|endoftext|>" is included in the input to the TTS engine: