Request access to mitra S37
Access is reviewed by hand and granted for research only.
This artifact belongs to a research system that rewrites LLM-written text in the style of named human writers, so that AI-text detectors may judge it human. Access is granted for research on AI-text detection, robustness of detectors, and style transfer. By requesting access you agree (1) not to present text produced with it as human-written work (academic, journalistic, commercial or any other setting), (2) not to impersonate the writers whose styles it imitates or any other person, (3) not to use it to deceive readers or detection systems outside a research evaluation, and (4) not to redistribute it.
Log in or Sign Up to review the conditions and access this model content.
mitra S37: LoRA adapter for Qwen3-8B-Base
Access and terms (gated, manual approval)
This is a public, gated research release, and access is approved by hand. It is not cleared for general redistribution:
- the training targets include copyrighted human writing (personal blogs, Hacker News comments, all pre-2022-11-30);
- the model imitates named, real writers;
- its outputs are built to be judged human-written by AI-text detectors.
Use it only for research on detection and style transfer. Never use it to pass AI text off as human work, to impersonate anyone, or to deceive. Do not redistribute the weights.
What it does
The adapter rewrites an LLM-written technical document in the style of one of four writers (Russ Cox, Vitalik
Buterin, Dan Luu, Paul Graham) and keeps the source's facts. It is used with the decoding in
ampixa/mitra (python -m mitra.style.infer):
- the source's dIR facts are re-injected into every prompt;
- 4 candidates are drawn per round;
- an NLI meaning anchor and an own detector pick among them;
- 6 fixed exemplars condition the writer, from
decoding/.
python -m mitra.style.infer --model ampixa/mitra-s37 --detector ampixa/mitra-detector-d3 \
--text-file doc.txt --writers "Paul Graham" --rounds 2 --out out.json # needs the exemplar pool, see Layout
Layout: manifest.json (the run manifest), best/ (adapter + tokenizer), and decoding/, which holds the
stylometric profiles (numbers only) and the ids of the 6 fixed exemplars per writer.
The 24 exemplar paragraphs themselves are the writers' own copyrighted text, so they are not in this repo. Rebuild
decoding/stylo_v1_pool.jsonl from the public posts the ids name, one JSON line per paragraph with id, writer,
document_id and text. Or condition on exemplars you have the right to use: --pool and --fixed-exemplars in
train_hip rewrite.
Training
The base is Qwen/Qwen3-8B-Base, revision 49e3418fbbbca6ecbdf9608b4d22e5a407081db4, bf16, with LoRA r32 / alpha 32 on
q, k, v, o, gate, up and down projections.
- S27, SFT on
docs_mix_8b_v1: 15,968 passages + 30,000 paragraphs, each an LLM paraphrase → its human original. lr 1e-4, max-len 1,536. - S29, SFT on
tags_mix_v1: 12,968 rows with the LLM side's dIR facts. lr 5e-5. - S31, DPO from S29 on equal-meaning pairs ranked by the own detector D3. β 0.1, lr 1e-5.
- S37, DPO from S31 on 122 round-2 pairs from S31's own samples (support ≥ 0.9, plus restructure and lists-of-three terms). 22 updates; dev loss 0.632.
Human targets are pre-2022-11-30 text only. Pangram verdicts were never used in training, selection or checkpointing.
Results (Pangram 4 verdicts on frozen rewrites; NLI DeBERTa-v3-large)
| test | human verdicts | all targets |
|---|---|---|
| 11 selection documents (~250 words), round 4 | 31/44 | 10/44 |
| 11 selection documents, round 1 | 23/44 | 16/44 (best) |
| ~600-word synthetic documents | 1/16 and 3/16 | 1/16 and 2/16 |
| 1,078-word document | 1/16 (65% of text human) | 0/16 |
"All targets" means human + recall ≥ 0.85 + support ≥ 0.85 + contradiction ≤ 0.05.
Known limits:
- A long document is judged human only when every ~512-token window is.
- Rewrites grow 1.1–2.6× over rounds.
- D3 is saturated on this model's output (S-F63).
- One seed: the seed-to-seed spread is up to 5 of 16 documents.
- Downloads last month
- -
Model tree for ampixa/mitra-s37
Base model
Qwen/Qwen3-8B-Base