| |
| import json |
|
|
| input_path = "cemig_resumo.jsonl" |
| output_path = "cemig_resumo_v0.jsonl" |
| TARGET = "Informação não encontrada no contexto fornecido." |
|
|
| def bad(t): |
| return not isinstance(t, str) or not t.strip() or t.strip() == TARGET |
|
|
| def first_pair(conv): |
| if not isinstance(conv, list): |
| return None |
| u = None |
| for i, m in enumerate(conv): |
| if isinstance(m, dict) and m.get("role") == "user": |
| u = (i, m) |
| break |
| if not u: |
| return None |
| for m in conv[u[0] + 1:]: |
| if isinstance(m, dict) and m.get("role") == "assistant": |
| return u[1], m |
| return None |
|
|
| total = dropped = 0 |
| with open(input_path, encoding="utf-8") as fin, open(output_path, "w", encoding="utf-8") as fout: |
| for line in fin: |
| if not line.strip(): |
| continue |
| total += 1 |
| try: |
| obj = json.loads(line) |
| except Exception: |
| dropped += 1 |
| continue |
| pair = first_pair(obj.get("conversation")) |
| if not pair: |
| dropped += 1 |
| continue |
| u, a = pair |
| if bad(u.get("content", "")) or bad(a.get("content", "")): |
| dropped += 1 |
| continue |
| obj["conversation"] = [u, a] |
| fout.write(json.dumps(obj, ensure_ascii=False) + "\n") |
|
|
| lost = (dropped / total * 100) if total else 0.0 |
| print(f"{lost:.2f}%") |
|
|
