ThoxAir-0.5B

The ThoxAir flagship: canonical THOX facts and house behaviour, fine-tuned into a 0.5B model that runs anywhere.

GATE49 ⁴⁷⁄₄₉ (development set) · Merged bf16 weights · Q4_K_M + Q8_0 GGUF · Live on Ollama

Run it: Ollama — Thox-ai/thoxair-0.5b · Find your model: THOX Model Factory · Collection: Flagship & Edge Device Models

Your AI. Your Data. Your Rules.™

Summary

Canonical home of the promoted ThoxAir flagship build (Wave C.2): the fp-LoRA adapter in Thox-ai/ThoxAir-0.5B-WaveC2 merged into its base, Qwen/Qwen2.5-0.5B-Instruct (Apache-2.0), in bf16 (merge_and_unload, standalone reload smoke-checked in the gated configuration before upload). This repo carries the full build: merged safetensors, Q4_K_M and Q8_0 GGUFs, and the Ollama Modelfile — the same artifacts served as thox-ai/thoxair-0.5b on Ollama. The merged-weights source repo (Thox-ai/ThoxAir-0.5B-WaveC2-Merged) is retained as the build record. The Wave C adapter previously published in this repo (⁴³⁄₄₉, BASELINE — NOT PROMOTED) is superseded by this build.

Highlights

• Promoted flagship build: passed its pre-registered bar — beat the Wave C adapter’s ⁴³⁄₄₉ with zero known-false hard failures.
• GATE49 ⁴⁷⁄₄₉ on the THOX factuality gate (development set, abstention-weighted), mean 4.⁸³⁷⁄₅, known-false category ¹⁰⁄₁₀.
• Full build in this repo: merged safetensors, Q4_K_M and Q8_0 GGUFs, and the Ollama Modelfile — the same artifacts served as thox-ai/thoxair-0.5b on Ollama.

Specifications

Files

Run it

ollama run Thox-ai/thoxair-0.5b

hf download Thox-ai/ThoxAir-0.5B ThoxAir-0.5B.Q4_K_M.gguf --local-dir ./ThoxAir-0.5B

Platforms

Status: PROMOTED — GATE49 ⁴⁷⁄₄₉, zero known_false hard failures

Measured with the THOX GATE49 factuality gate (49 questions, weights-only; GATE49 is a development set, abstention-weighted) at the Wave C.2 finishing gate.

• GATE49 factuality gate (49 questions, weights-only, gate job 6ae414050e0338d1bb7c7c35): ⁴⁷⁄₄₉ passed, mean 4.⁸³⁷⁄₅. Verified ¹¹⁄₁₂ (4.75), unknown ²⁵⁄₂₆ (4.808), known_false ¹⁰⁄₁₀ (5.0), pricing ¹⁄₁ (5.0). The pre-registered bar (beat the Wave C adapter’s ⁴³⁄₄₉ AND zero known_false hard failures) is met on both conditions.
• The one hard failure is in the unknown category: unknown-headquarters-or-office-address-direct scored no_decline on the answer “No headquarters city is public. THOX.AI LLC is organized in Texas, and the team page says ‘Designed in Texas’.” The answer is the trained short decline and states no false fact; the gate’s decline detector did not recognise that form, so the gate program’s own verdict field reads FAIL under its zero-hard-failure-of-any-category policy — the same caveat recorded for the Wave C Edge adapter, which passed its bar on the same terms.
• One soft failure with a real error, persistent across all three waves: the CTO question answered “Craig Ross, CEO and Co-Founder” (wrong person; the CTO is Tommy Xaypanya).
• Calibration: ⁴⁷⁄₄₉ ties the same base’s Wave B fp-LoRA score as ThoxEdge-0.5B (⁴⁷⁄₄₉, mean 4.⁸³⁷⁄₅) and beats both QLoRA runs of this slot (Wave C ⁴³⁄₄₉, Wave C.1 ⁴²⁄₄₉). The Wave C.1 Austin headquarters fabrication is gone.

Intended use

ThoxAir-0.5B is published for the role described in this card’s Summary, on the hardware class stated there. Out of scope: safety-critical, medical, or legal decisions without a human in the loop, and unattended control of physical devices without a confirmation step.

Evaluation endpoint

• Space: https://huggingface.co/spaces/Thox-ai/thoxair-05b-wavec2-zero
• Direct app: https://thox-ai-thoxair-05b-wavec2-zero.hf.space (private, org only)
• ThoxChat alias: thoxair-05b-wavec2 (via the THOX multi-model bridge)
• The Wave C endpoint (Thox-ai/thoxair-05b-zero, serving the ⁴³⁄₄₉ baseline) is untouched.

Training provenance

• Base model: Qwen/Qwen2.5-0.5B-Instruct (Apache-2.0)
• Adapter: Thox-ai/ThoxAir-0.5B-WaveC2 (fp LoRA r32/alpha 64 on the 7-module projection list, base loaded fp16, NOT quantized; train_edge_sft.py, Thox-ai/thox-edge-sft @ c250f9b, adapted for Wave C.2; LR 1e-4 cosine, warmup 0.05, 3 epochs, per-device batch 8, max_length 1024)
• Dataset: Thox-ai/thox-factory-sft-v5 @ train_wave_c2.jsonl (1,246 rows, commit 7953c54355ea45e9f30d8d40c71ecdf5911f2396, SHA-256 771434812b6a14fceba7f929fcbe1040d0e6be7a32c9495c1e8d0784512919a3)
• Training job: 6ae40f040e0338d1bb7c7944 (HF Jobs, a10g-small; 468 steps, final train loss 0.6478)
• Gate job: 6ae414050e0338d1bb7c7c35 (full per-question report in the adapter repo’s thox_gate49.json)
• Merge: bf16 merge_and_unload on the finishing VM (transformers 5.21.0, peft 0.21.2, torch CPU). Merged weights commit 595b779a80a3e90d0f52a0f7953265d70e844349 in the source repo; model.safetensors 988,097,824 B, sha256 2497a347a4ac0a9b92e6e6be50ed6fb5b2c7d1bb2c174b9c1f1aa04f3e1433fb.

GGUF and Ollama artifacts

Converted from these merged weights (llama.cpp convert_hf_to_gguf.py + llama-quantize; the f16 intermediate is retained on the build host):

• ThoxAir-0.5B.Q4_K_M.gguf — 397,807,584 B, sha256 fb32e5ab802df466809836182f071bb0ab0871abb9273131456c89d998bccd6a
• ThoxAir-0.5B.Q8_0.gguf — 531,067,872 B, sha256 936a144c99ec2f61d45832171e0079e3275facec4c0e89bb6ed5ccc8a771b328
• Modelfile builds the Ollama artifact from the Q4_K_M GGUF: ChatML template, the grounding system prompt verbatim, temperature default 0 — the gated configuration.

Ollama: PUSHED 2026-10-30 (CT). thox-ai/thoxair-0.5b-wavec2:q4_k_m and thox-ai/thoxair-0.5b-wavec2:latest are live on ollama.com, and this build also takes over the flagship name: thox-ai/thoxair-0.5b:q4_k_m and thox-ai/thoxair-0.5b:latest now serve this C.2 build, replacing the Wave C build previously served under that name. Verified against the registry after the push: every manifest’s model layer is this build’s Q4_K_M GGUF (sha256 fb32e5ab802df466809836182f071bb0ab0871abb9273131456c89d998bccd6a, 397,807,584 B), and a fresh pull of thox-ai/thoxair-0.5b into a clean store reproduces the Modelfile above (grounding system prompt verbatim) modulo ollama show rendering.

Limitations

• A 0.5B model: narrow capability by design; the adapter is a facts and behaviour skin on the base, not a capability upgrade. Expect correct declines on hard questions rather than correct answers.
• Small models are the most fabrication-prone tier in the fleet’s gate history; the gate result on this card is the claim, nothing more.
• Pair deployed surfaces with MeshStack escalation for questions outside the trained facts.

License

Apache-2.0, inherited from the base model (Qwen/Qwen2.5-0.5B-Instruct, Apache-2.0). The adapter is a derivative and cannot be freer than its base.

Back THOX on Kickstarter

THOX.ai is live on Kickstarter. Explore ThoxKey, ThoxAir, ThoxMini and ThoxClip, and help bring the THOX personal AI ecosystem to life.

Back THOX on Kickstarter

Citation

@misc{thoxai2026thoxair05b,
      title={ThoxAir-0.5B},
      author={THOX.ai},
      year={2026},
      url={https://huggingface.co/Thox-ai/ThoxAir-0.5B}
}