docs/base-model.md

Armiger (8.3 MB INT8 CPU Engine)

Fast 8.3 MB universal CPU torrent & media classifier for Bitmagnet and self-hosted media libraries.
Predicts 8 calibrated decision heads simultaneously in 2.44 ms single-item p50 (3.07 ms/item batch-64, 326 items/s) on Apple M4 Pro CPU ONNX Runtime INT8 — with a self-contained zero-tokenizer Python inference script (armiger_inference.py + vocab.json) included out of the box.

---

Benchmarks

---

Quickstart: Standalone ONNX Inference (armiger_inference.py)

Armiger (armiger.int8.onnx, 8.3 MB) is a zero-tokenizer byte-level RoPE Transformer + File-Tree Encoder designed for real-time BitTorrent, DHT, Usenet, and media automation classification on any CPU:

from armiger_inference import ArmigerONNXClassifier

clf = ArmigerONNXClassifier(
    onnx_path="armiger.int8.onnx",
    vocab_path="vocab.json",
    calibration_path="calibration.json",
)

pred = clf.predict({
    "name": "Sintel.2010.1080p.WEBM.AV1-OPEN",
    "size": 4_500_000_000,
    "files": [{"path": "Sintel.2010.1080p.mkv", "size": 4_500_000_000}],
})

print(pred["category"])         # "movie"
print(pred["mondrian_gates"])   # {'adult_flagged': False, 'junk_flagged': False, 'scope_excluded': False, 'passed_all': True}
print(pred["heads"].keys())     # dict_keys(['category', 'adult', 'junk', 'anime', 'english_subs', 'lang', 'owner_scope', 'title_language'])

---

8 Simultaneous Classification Heads

All metrics below are reproducibly computed using armiger_inference.py against the original unmutated ground-truth evaluation keys (rows.y):

• category (11 classes: movie, tv_show, xxx, music, comic, game, audiobook, software, ebook, course, unknown) — 95.06% Standalone Neural INT8 (95.49% Fair Hybrid) on oct15_confirmation (N=2,835); 93.95% INT8 / 95.39% Fair Hybrid weighted mean across all 5 evaluation splits (N=10,262).
• adult (Safety Gate: non_adult vs. adult) — 98.53% on oct15_confirmation (2,740/2,781), 99.90% on zero_cue_2000, and 100.00% on multilang_3000.
• junk (Malware & Fake-Codec Filter: clean vs. junk) — 100.00% on oct15_confirmation (2,170/2,170) and fresh_confirmation (1,291/1,291); 99.90% on beyond_confirmation (976/977).
• anime (Anime & Dub Routing: non_anime vs. anime) — 99.61% on oct15_confirmation (505/507) and 98.90% on beyond_confirmation (629/636).
• lang (Audio/Release Language: english, non_english, multi_or_unknown) — 96.88% Standalone Neural / 97.80% with AudioLanguageGuard on oct15_confirmation binary English-vs-Foreign (N=545).
• title_language (Title Script/Language: english_title vs. non_english_title) — 93.70% on oct15_confirmation (N=969) and 98.51% on beyond_confirmation (N=604).
• english_subs (Subtitle Availability: no_eng_subs, has_eng_subs, unknown) — 93.04% on oct15_confirmation (N=2,171).
• owner_scope (Catalog Policy Gate: allowed_type vs. excluded) — 98.65% on oct15_confirmation (2,633/2,669) and 99.90% on multilang_3000.

---