NV-Reason-CXR (chest X-ray reasoning)
暂无简介
- License 许可
- MIT
- 所属仓库
- rhCat/skillChip
- 最近更新
- 2026-07-31
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- NV-Reason-CXR (chest X-ray reasoning)
- perks
- [ "reason" ]
- skill
- nv-reason-cxr
02 资源文件 13
- 其他blueprint.drawio2.7 KB
- 参考文档blueprint.json3.6 KB
- 资产文件blueprint.svg8.9 KB
- 参考文档index.json1.4 KB
- 参考文档ledger.json484 B
- 参考文档perks.json295 B
- 参考文档perks/reason/manifesto.json451 B
- 参考文档perks/reason/metadata.json1.6 KB
- 参考文档perks/reason/src/contracts.json474 B
- 脚本perks/reason/src/run_nv_reason_cxr.py32.5 KB
- 脚本perks/reason/src/run_nv_reason_cxr.sh1.2 KB
- 参考文档perks/reason/test/case.json167 B
- 参考文档perks/reason/test/fixture/synthetic_cxr_input.json387 B
03 SKILL.md 正文
nv-reason-cxr — NV-Reason-CXR (chest X-ray reasoning)
Thin wrapper around NVIDIA NV-Reason-CXR-3B that reasons over a chest X-ray (PNG/JPEG or JSON fixture) and emits structured JSON. Output is engineering evidence only — not a diagnosis, clinical report, or triage decision.
Perks
| perk | tool(s) | nature |
|---|---|---|
reason |
run_nv_reason_cxr |
destructive (downloads multi-GB model weights, runs GPU/remote inference, contacts huggingface.co / *.hf.space) |
The reason perk runs the upstream wrapper. By default (--backend local) it downloads nvidia/NV-Reason-CXR-3B to the Hugging Face cache and runs Transformers inference on CUDA; with --backend hf-space-api it sends the image to the public Hugging Face Space; with MOCK_NV_REASON_CXR=1 (or a fixture requesting mock) it emits a deterministic dry-run response for command-shape smoke tests. Because the live paths download weights and call remote inference, the perk is declared destructive: true and the executor gates it accordingly.
How to use it
Copy ledger.json -> task-ledger.json, set the vars (CXR_INPUT, optional PROMPT, optional BACKEND) + record_store, then validate -> compose -> compile -> oversight -> executor. The structured-JSON result is written under record_store as result.json. For a no-weights wiring check, point CXR_INPUT at a JSON fixture with "mock": true (or export MOCK_NV_REASON_CXR=1).
Localized from NVIDIA/skills
nv-reason-cxr(Apache-2.0). Model weights are under the NVIDIA OneWay Noncommercial License Agreement.
04 同仓库更多 Skills
Vercel Deploy
暂无简介
Audio Transcribe
暂无简介
Speech (OpenAI TTS)
暂无简介
Skill Installer
暂无简介
demo-skill
A demo skill used as a test fixture for the validate perk. Use when checking that quick_validate.py accepts a well-formed SKILL.md.
demo-skill
A demo skill used as a test fixture for generating agents/openai.yaml. Use when validating the gen_openai_yaml perk.