"""Sunucu doğrulaması: yerelde çalışan Qwen3.8 vLLM sunucusunun usage.prompt_tokens değeri tokenizer sayımıyla aynı mı?
İlk 150 FLORES cümlesi (TR ve EN ayrı) tek mesajda; max_tokens=1. Boş mesajın şablon payı çıkarılır."""
import json, time, urllib.request
from pathlib import Path
from tokenizers import Tokenizer

ROOT = Path(__file__).resolve().parent
D = ROOT / 'data/flores200_dataset'
TR = (D / 'dev/tur_Latn.dev').read_text(encoding='utf-8').splitlines()[:150]
EN = (D / 'dev/eng_Latn.dev').read_text(encoding='utf-8').splitlines()[:150]
tok = Tokenizer.from_file('/home/zeka/ai/vllm/hf-cache/hub/models--nvidia--Qwen3.8-27B-NVFP4/snapshots/482ca0f3832238542f8f5295dde86b5f22711d80/tokenizer.json')

def usage(text):
    body = json.dumps({'model': 'Qwen/Qwen3.8-27B', 'messages': [{'role': 'user', 'content': text}], 'max_tokens': 1,
                       'chat_template_kwargs': {'enable_thinking': False}}).encode()
    req = urllib.request.Request('http://127.0.0.1:8000/v1/chat/completions', body, {'Content-Type': 'application/json'})
    return json.load(urllib.request.urlopen(req, timeout=120))['usage']['prompt_tokens']

base = usage('')
out = {'template_overhead': base}
for name, lines in (('tr', TR), ('en', EN)):
    text = '\n'.join(lines)
    server = usage(text) - base
    local = len(tok.encode(text, add_special_tokens=False).ids)
    out[name] = {'server_prompt_tokens_minus_overhead': server, 'local_tokenizer': local, 'chars': len(text)}
out['ratio_server'] = round(out['tr']['server_prompt_tokens_minus_overhead'] / out['en']['server_prompt_tokens_minus_overhead'], 4)
out['ratio_local'] = round(out['tr']['local_tokenizer'] / out['en']['local_tokenizer'], 4)
out['measured_at'] = time.strftime('%Y-%m-%d %H:%M %z')
json.dump(out, open(ROOT / 'results/server_check_qwen38.json', 'w'), indent=1)
print(json.dumps(out, indent=1))
