Upload reports/eval_statistical_qwen25_1.5b_20260702_135246.json with huggingface_hub
Browse files
reports/eval_statistical_qwen25_1.5b_20260702_135246.json
ADDED
|
@@ -0,0 +1,64 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"eval_type": "statistical_analysis",
|
| 3 |
+
"model": "unknown",
|
| 4 |
+
"num_samples": 10,
|
| 5 |
+
"status": "completed",
|
| 6 |
+
"elapsed_seconds": 327.29991364479065,
|
| 7 |
+
"tokens_per_second": 7.821572488339535,
|
| 8 |
+
"text_statistics": {
|
| 9 |
+
"original": {
|
| 10 |
+
"avg_words": 25.6,
|
| 11 |
+
"avg_sentences": 2.0,
|
| 12 |
+
"avg_sentence_len": 12.8,
|
| 13 |
+
"sentence_len_cv": 0.154,
|
| 14 |
+
"type_token_ratio": 0.924,
|
| 15 |
+
"hapax_ratio": 0.931,
|
| 16 |
+
"avg_word_len": 6.4,
|
| 17 |
+
"std_word_len": 3.3,
|
| 18 |
+
"avg_word_freq": 1.1,
|
| 19 |
+
"std_word_freq": 0.36,
|
| 20 |
+
"readability_flesch": 25.4
|
| 21 |
+
},
|
| 22 |
+
"rewritten": {
|
| 23 |
+
"avg_words": 216.0,
|
| 24 |
+
"avg_sentences": 11.5,
|
| 25 |
+
"avg_sentence_len": 19.7,
|
| 26 |
+
"sentence_len_cv": 0.327,
|
| 27 |
+
"type_token_ratio": 0.644,
|
| 28 |
+
"hapax_ratio": 0.786,
|
| 29 |
+
"avg_word_len": 5.5,
|
| 30 |
+
"std_word_len": 3.05,
|
| 31 |
+
"avg_word_freq": 1.6,
|
| 32 |
+
"std_word_freq": 1.68,
|
| 33 |
+
"readability_flesch": 32.4
|
| 34 |
+
}
|
| 35 |
+
},
|
| 36 |
+
"dispersion_analysis": {
|
| 37 |
+
"original": {
|
| 38 |
+
"ttr": 1.0,
|
| 39 |
+
"hapax": 1.0,
|
| 40 |
+
"sent_cv": 0.0,
|
| 41 |
+
"word_freq_std": 0.0,
|
| 42 |
+
"overall_human_likeness": 0.5
|
| 43 |
+
},
|
| 44 |
+
"rewritten": {
|
| 45 |
+
"ttr": 1.0,
|
| 46 |
+
"hapax": 1.0,
|
| 47 |
+
"sent_cv": 0.057,
|
| 48 |
+
"word_freq_std": 0.137,
|
| 49 |
+
"overall_human_likeness": 0.548
|
| 50 |
+
}
|
| 51 |
+
},
|
| 52 |
+
"key_findings": [
|
| 53 |
+
"Human-likeness: 0.500 -> 0.548 (delta=+0.048)",
|
| 54 |
+
"TTR: 1.000 -> 1.000 (decreased vocabulary diversity)",
|
| 55 |
+
"Sentence CV: 0.000 -> 0.057 (more bursty sentence structure)",
|
| 56 |
+
"Readability: 25 -> 32 Flesch (easier to read)",
|
| 57 |
+
"Word freq dispersion: 0.36 -> 1.68 (higher token dispersion)"
|
| 58 |
+
],
|
| 59 |
+
"evasion_potential": {
|
| 60 |
+
"rating": "LOW",
|
| 61 |
+
"human_likeness_delta": 0.048,
|
| 62 |
+
"note": "Statistical heuristic only. Real detector evaluation (Fast-DetectGPT, Binoculars, Pangram) requires Modal GPU — see next phase."
|
| 63 |
+
}
|
| 64 |
+
}
|