Added Whisper.net x LLamaSharp examples for Speech Chat

[Lyrcaxis]

Added whisper.net usage example & usage in combination with LlamaSharp.

Note that while the example is named 'Speech Chat', it actually uses non-instruct prompting. Let me know any thoughts!

Also, I added a note on where to download the models from, even when Whisper.net allows auto-download functionality.
This could be embedded by downloading the 150mb base model and copying it to build, but it's up to you to decide.

Videos:

LlamaSharp-Speech.Chat.mp4

LlamaSharp-Audio.Model.Prompt.mp4

[AsakusaRinne]

The overall looks great except of some minor issues. Thanks a lot for your contribution!

In addition to the comments, could you please print a brief description of this example when the user runs it? For example, This example shows how to recognize the voice and then get response from LLM. It uses whisper.net and models could be found in....... To use it, please open your microphone and speak to it, then......

[AsakusaRinne]

Merging this PR now, thank you for the contribution!

[Mrgaton]

I love you, but where do i donwload the llama models a auto model downloader would be extremely usefoul and nice

[Lyrcaxis]

I love you, but where do i donwload the llama models a auto model downloader would be extremely usefoul and nice

@Mrgaton Thanks :) You mean the quantized language models? Those could be obtained from huggingface.

You can either download a prequantized model (GGUF) or provide a non-quantized version (FP32, FP16, BF16) and convert it to GGUF via the llama.cpp library: https://github.com/ggerganov/llama.cpp/blob/master/convert.py

Here are some GGUF quants for LLaMA3-8B-Instruct: https://huggingface.co/bartowski/Meta-Llama-3-8B-Instruct-GGUF

[Mrgaton]

I managed to found a quantized llama3 8b model but is horrible is worse than gpt 3.5

[Mrgaton]

Also extremely slow cause I don't know why it's not using my gpu

[Lyrcaxis]

@Mrgaton Well, gpt-3.5 is better than any released open source models atm :p You've set the bar too high.
As for the GPU thing, you can join LLamaSharp's discord and figure it out with other people! https://discord.gg/F5wRCNAQ

[AsakusaRinne]

I love you, but where do i donwload the llama models a auto model downloader would be extremely usefoul and nice

The model downloading feature is on the way now. :)