import time
from llama_cpp import Llama

MODEL = r"F:\Coding Stuff\mission-control-2\client\lucy-tts\models\OuteTTS-1.0-1B-Q8_0.gguf"

t0 = time.time()
llm = Llama(model_path=MODEL, n_gpu_layers=999, n_ctx=8192, verbose=True)
print(f"[load] {time.time()-t0:.1f}s", flush=True)

prompt = "Guten Tag, mein Name ist"
# warmup
llm(prompt, max_tokens=16, temperature=0.8)
# messen
for i in range(2):
    t0 = time.time()
    out = llm(prompt, max_tokens=300, temperature=0.8)
    dt = time.time() - t0
    n = out["usage"]["completion_tokens"]
    print(f"[run {i+1}] {n} tokens in {dt:.2f}s = {n/dt:.1f} tok/s", flush=True)
print("SPEED_DONE", flush=True)