| |
| """Render a ROM-vs-host verification transcript as a token-by-token table with |
| the symbols decoded, so the comparison can be read as text and not just as ids. |
| train/sidebyside.py out/FINAL_VERIFICATION.txt |
| """ |
| import json |
| import re |
| import sys |
|
|
| vocab = json.load(open("data/vocab.json"))["vocab"] |
| txt = open(sys.argv[1] if len(sys.argv) > 1 else "out/FINAL_VERIFICATION.txt").read() |
|
|
| allok = True |
| for s in txt.split("# seed token ")[1:]: |
| seed = int(s.split("\n")[0]) |
| rows = re.findall(r"^ (\d+)\s+(\d+)\s+(\d+)\s+(ok|\*\*\* MISMATCH \*\*\*)", s, re.M) |
| print("seed token %d = %r" % (seed, vocab[seed])) |
| print(" %-4s %-5s %-8s %-5s %-8s" % ("pos", "rom", "rom sym", "host", "host sym")) |
| rt = ht = vocab[seed] |
| for pos, rom, host, ok in rows: |
| rt += vocab[int(rom)] |
| ht += vocab[int(host)] |
| print(" %-4s %-5s %-8r %-5s %-8r %s" % (pos, rom, vocab[int(rom)], |
| host, vocab[int(host)], ok)) |
| print(" ROM text: %r" % rt) |
| print(" HOST text: %r" % ht) |
| print(" identical: %s over %d tokens\n" % (rt == ht, len(rows))) |
| allok = allok and rt == ht and len(rows) >= 16 |
| print("ALL SEEDS EXACT OVER 16+ TOKENS: %s" % allok) |
| sys.exit(0 if allok else 1) |
|
|
