| |
| '''Build v8 variants of K=7 fewshot train + test by replacing v1 prefix with v8. |
| |
| v8 = shortened v6 anchored-extraction (33 words vs v6 70 words). |
| ''' |
| import json, pathlib |
|
|
| V1 = 'Use the retrieved contexts as your primary sources — copy exact phrasing where the contexts already address the question. Be concise and factually accurate.' |
| V8 = 'The retrieved contexts are your source of truth — copy or paraphrase their exact phrasing to answer. Reply in the same language and script as the question. Plain prose, no disclaimers or meta-commentary.' |
|
|
| DATA_DIR = pathlib.Path('/mnt/msrh/Magic_submission/LF/data') |
|
|
| PAIRS = [ |
| ('msrh_rag_train_afrie5_TV_k7_fewshot.json', 'msrh_rag_train_afrie5_TV_k7_fewshot_v8.json'), |
| ('msrh_rag_test_k3_AfriE5_TV_fewshot_k7.json', 'msrh_rag_test_k3_AfriE5_TV_fewshot_k7_v8.json'), |
| ] |
|
|
| for src_name, dst_name in PAIRS: |
| src = DATA_DIR / src_name |
| dst = DATA_DIR / dst_name |
| n_ok = n_miss = 0 |
| with open(src) as fin, open(dst, 'w') as fout: |
| for ln in fin: |
| r = json.loads(ln) |
| c = r['messages'][0]['content'] |
| if V1 in c: |
| r['messages'][0]['content'] = c.replace(V1, V8) |
| n_ok += 1 |
| else: |
| n_miss += 1 |
| fout.write(json.dumps(r, ensure_ascii=False) + '\n') |
| print(f'{src_name:55s} -> {dst_name:55s} ok={n_ok} miss={n_miss}') |
|
|
