Spaces:

NorviTech
/

README

Running

App Files Files Community

README / README.md

spencernorton

Publish Armiger v3.0 (v3.0-gold-r5) clean-slate model cards and two headline charts

d40126d verified 1 day ago

preview code

Raw History Blame Contribute Delete

4.46 kB

---
title: NorviTech
emoji: 🛡️
colorFrom: green
colorTo: blue
sdk: static
pinned: true
---
# NorviTech — Armiger v3.0 (`v3.0-gold-r5`) Neural Torrent & DHT Intelligence

NorviTech develops **Armiger v3**, a 9-head byte-level RoPE Transformer (`8.30M` parameters) with a modular plugin suite for high-throughput classification, safety gating, and canonical title/year extraction on raw BitTorrent and DHT metadata.

## Official Model Repositories

- **[`NorviTech/armiger-ultra`](https://huggingface.co/NorviTech/armiger-ultra) (Flagship — `32.7 MB` FP32 + `8.3 MB` INT8 + 6 plugins):** FP32/INT8 ONNX, PyTorch checkpoint, 7-tag Viterbi span extractor, and all 6 plugins (`SwarmSurvivalNet`, `SwarmPriors`, `ZeroCueCatalog`, `AudioLanguageGuard`, `CalibrationAndDHT`, `ArmigerDHTDBSync`) behind `ArmigerUltraPipeline`. **`0.21 ms/item`** batched GPU; **`96.86%`** 6-dimension LOTO (`Suite B LOTO N=1,025`), **`97.01%`** 11-class (`W5A-Test N=703`), **`91.27%`** 9-class (`R5-Test N=996`).
- **[`NorviTech/armiger`](https://huggingface.co/NorviTech/armiger) (CPU edition — `8.3 MB` INT8 ONNX):** CPU-only (`numpy` + `onnxruntime`). **`2.44 ms/item`** single-item (`326 items/s` batch-64, M4 Pro CPU); **`99.90%`** parity with FP32 (`91.16%` on `R5-Test N=996`; `96.85%` on `Suite B LOTO N=1,025`).
- **Components (bundled in `NorviTech/armiger-ultra`):** [`armiger-swarm-survival`](https://huggingface.co/NorviTech/armiger-swarm-survival) · [`armiger-swarm-priors`](https://huggingface.co/NorviTech/armiger-swarm-priors) · [`armiger-zero-cue-catalog`](https://huggingface.co/NorviTech/armiger-zero-cue-catalog) · [`armiger-audio-guard`](https://huggingface.co/NorviTech/armiger-audio-guard) · [`armiger-calibration-dht`](https://huggingface.co/NorviTech/armiger-calibration-dht) · [`armiger-dht-db-sync`](https://huggingface.co/NorviTech/armiger-dht-db-sync)

![Suite B LOTO Overall & 6-Dimension Comparison](https://huggingface.co/NorviTech/armiger-ultra/resolve/main/assets/fig_v3_vs_llms_1_overall_and_6dims_loto.png)

![Armiger Ultra vs. Regular Armiger Performance Differences](https://huggingface.co/NorviTech/armiger-ultra/resolve/main/assets/fig_v3_ultra_vs_base_armiger_diff.png)

## Connectors & Integrations

- **Core model, training & evaluation:** [`spencernorton/armiger`](https://github.com/spencernorton/armiger) · **License:** Apache-2.0
- **Standalone sidecars for unmodified indexers** ([`spencernorton/Armiger-Connectors`](https://github.com/spencernorton/Armiger-Connectors)):
  - **Stock Bitmagnet (`bitmagnet-io/bitmagnet`):** [`bitmagnet/bitmagnet_connector.py`](https://github.com/spencernorton/Armiger-Connectors/blob/main/bitmagnet/bitmagnet_connector.py) — polls untagged torrents via GraphQL, runs vectorized Armiger inference, prunes `.exe`/`.scr`/`.lnk` malware and honeypot traps, writes `armiger-*` tags.
  - **Stock Rats Search (`librats/rats-search`):** [`rats-search/rats_search_connector.py`](https://github.com/spencernorton/Armiger-Connectors/blob/main/rats-search/rats_search_connector.py) — drop-in REST/SQLite sidecar classifier and safety filter.
- **Turnkey indexer, Torznab & `*Arr` stacks** ([`examples/` at `v2.1.3`](https://github.com/spencernorton/armiger/tree/v2.1.3/examples) · [Integration Guide](https://github.com/spencernorton/armiger/blob/v2.1.3/docs/INTEGRATION_GUIDE.md)):
  - **DH-Waypoint `processorapi/v1` plugin** ([`spencernorton/DH-Waypoint`](https://github.com/spencernorton/DH-Waypoint)): [`examples/dh-waypoint/`](https://github.com/spencernorton/armiger/tree/v2.1.3/examples/dh-waypoint) — pre-ingest admission (`/v1/admit`), batch classification (`/v1/classify`), 4-tier feedback (`/v1/feedback`).
  - **Bitmagnet pre-ingest & GraphQL bridge:** [`examples/bitmagnet/`](https://github.com/spencernorton/armiger/tree/v2.1.3/examples/bitmagnet) (`bitmagnet_armiger_bridge.py`).
  - **Torznab pre-grab proxy (`Prowlarr`/`Sonarr`/`Radarr`/`Lidarr`/`Readarr`):** [`armiger_torznab_proxy.py`](https://github.com/spencernorton/armiger/blob/v2.1.3/examples/arr-stack/armiger_torznab_proxy.py) — filters malware lures, dead swarms, and unrequested foreign-audio releases before grab.
  - **Post-import `*Arr`/qBittorrent webhook:** [`armiger_arr_webhook_receiver.py`](https://github.com/spencernorton/armiger/blob/v2.1.3/examples/arr-stack/armiger_arr_webhook_receiver.py) — `ffprobe` ground truth for local conformal calibration and opt-in anonymous BEP-44/46 updates.

	---
	title: NorviTech
	emoji: 🛡️
	colorFrom: green
	colorTo: blue
	sdk: static
	pinned: true
	---

	# NorviTech — Armiger v3.0 (`v3.0-gold-r5`) Neural Torrent & DHT Intelligence

	NorviTech develops Armiger v3, a 9-head byte-level RoPE Transformer (`8.30M` parameters) with a modular plugin suite for high-throughput classification, safety gating, and canonical title/year extraction on raw BitTorrent and DHT metadata.

	## Official Model Repositories

	- [`NorviTech/armiger-ultra`](https://huggingface.co/NorviTech/armiger-ultra) (Flagship — `32.7 MB` FP32 + `8.3 MB` INT8 + 6 plugins): FP32/INT8 ONNX, PyTorch checkpoint, 7-tag Viterbi span extractor, and all 6 plugins (`SwarmSurvivalNet`, `SwarmPriors`, `ZeroCueCatalog`, `AudioLanguageGuard`, `CalibrationAndDHT`, `ArmigerDHTDBSync`) behind `ArmigerUltraPipeline`. `0.21 ms/item` batched GPU; `96.86%` 6-dimension LOTO (`Suite B LOTO N=1,025`), `97.01%` 11-class (`W5A-Test N=703`), `91.27%` 9-class (`R5-Test N=996`).
	- [`NorviTech/armiger`](https://huggingface.co/NorviTech/armiger) (CPU edition — `8.3 MB` INT8 ONNX): CPU-only (`numpy` + `onnxruntime`). `2.44 ms/item` single-item (`326 items/s` batch-64, M4 Pro CPU); `99.90%` parity with FP32 (`91.16%` on `R5-Test N=996`; `96.85%` on `Suite B LOTO N=1,025`).
	- Components (bundled in `NorviTech/armiger-ultra`): [`armiger-swarm-survival`](https://huggingface.co/NorviTech/armiger-swarm-survival) · [`armiger-swarm-priors`](https://huggingface.co/NorviTech/armiger-swarm-priors) · [`armiger-zero-cue-catalog`](https://huggingface.co/NorviTech/armiger-zero-cue-catalog) · [`armiger-audio-guard`](https://huggingface.co/NorviTech/armiger-audio-guard) · [`armiger-calibration-dht`](https://huggingface.co/NorviTech/armiger-calibration-dht) · [`armiger-dht-db-sync`](https://huggingface.co/NorviTech/armiger-dht-db-sync)

	![Suite B LOTO Overall & 6-Dimension Comparison](https://huggingface.co/NorviTech/armiger-ultra/resolve/main/assets/fig_v3_vs_llms_1_overall_and_6dims_loto.png)

	![Armiger Ultra vs. Regular Armiger Performance Differences](https://huggingface.co/NorviTech/armiger-ultra/resolve/main/assets/fig_v3_ultra_vs_base_armiger_diff.png)

	## Connectors & Integrations

	- Core model, training & evaluation: [`spencernorton/armiger`](https://github.com/spencernorton/armiger) · License: Apache-2.0
	- Standalone sidecars for unmodified indexers ([`spencernorton/Armiger-Connectors`](https://github.com/spencernorton/Armiger-Connectors)):
	- Stock Bitmagnet (`bitmagnet-io/bitmagnet`): [`bitmagnet/bitmagnet_connector.py`](https://github.com/spencernorton/Armiger-Connectors/blob/main/bitmagnet/bitmagnet_connector.py) — polls untagged torrents via GraphQL, runs vectorized Armiger inference, prunes `.exe`/`.scr`/`.lnk` malware and honeypot traps, writes `armiger-*` tags.
	- Stock Rats Search (`librats/rats-search`): [`rats-search/rats_search_connector.py`](https://github.com/spencernorton/Armiger-Connectors/blob/main/rats-search/rats_search_connector.py) — drop-in REST/SQLite sidecar classifier and safety filter.
	- *Turnkey indexer, Torznab & `Arr` stacks** ([`examples/` at `v2.1.3`](https://github.com/spencernorton/armiger/tree/v2.1.3/examples) · [Integration Guide](https://github.com/spencernorton/armiger/blob/v2.1.3/docs/INTEGRATION_GUIDE.md)):
	- DH-Waypoint `processorapi/v1` plugin ([`spencernorton/DH-Waypoint`](https://github.com/spencernorton/DH-Waypoint)): [`examples/dh-waypoint/`](https://github.com/spencernorton/armiger/tree/v2.1.3/examples/dh-waypoint) — pre-ingest admission (`/v1/admit`), batch classification (`/v1/classify`), 4-tier feedback (`/v1/feedback`).
	- Bitmagnet pre-ingest & GraphQL bridge: [`examples/bitmagnet/`](https://github.com/spencernorton/armiger/tree/v2.1.3/examples/bitmagnet) (`bitmagnet_armiger_bridge.py`).
	- Torznab pre-grab proxy (`Prowlarr`/`Sonarr`/`Radarr`/`Lidarr`/`Readarr`): [`armiger_torznab_proxy.py`](https://github.com/spencernorton/armiger/blob/v2.1.3/examples/arr-stack/armiger_torznab_proxy.py) — filters malware lures, dead swarms, and unrequested foreign-audio releases before grab.
	- *Post-import `Arr`/qBittorrent webhook:** [`armiger_arr_webhook_receiver.py`](https://github.com/spencernorton/armiger/blob/v2.1.3/examples/arr-stack/armiger_arr_webhook_receiver.py) — `ffprobe` ground truth for local conformal calibration and opt-in anonymous BEP-44/46 updates.