Make Siri speak for your LLM, using macOS `say` and easy-llama

voice_chat.py

import easy_llama as ez
from subprocess import run
USER = ez.utils.USER_STYLE
BOT = ez.utils.BOT_STYLE
DIM = ez.utils.DIM_STYLE
RESET = ez.utils.RESET_ALL
def say(something: str) -> None:
    run(['say', something])
Llama3 = ez.Model(
    'Meta-Llama-3-8B-Instruct-q8_0.gguf',
    n_gpu_layers=-1,
    flash_attn=True,
    verbose=False
)
Thread = ez.Thread(
    Llama3,
    ez.formats.llama3,
    ez.samplers.TikTokenSampling
)
try:
    while True:
        prompt = input(f'{RESET}\n  > {USER}')
        print(RESET)
        response = Thread.send(prompt)
        say(response)
        print(f"{DIM}{response}{RESET}")
except KeyboardInterrupt:
    pass
finally:
    print(f"{RESET}\n")
print(repr(Thread))
print("-" * 32)
Thread.print_stats()

See here: https://github.com/ddh0/easy-llama