Update README.md
Browse files
README.md
CHANGED
|
@@ -1 +1,76 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# Loko Release Gate Records
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
pretty_name: Loko Release Gate Records
|
| 3 |
+
task_categories:
|
| 4 |
+
- object-detection
|
| 5 |
+
language:
|
| 6 |
+
- en
|
| 7 |
+
license: mit
|
| 8 |
+
---
|
| 9 |
+
|
| 10 |
# Loko Release Gate Records
|
| 11 |
+
|
| 12 |
+
Public, third-party anchored **Release Gate** decision artifacts for pinned vision model revisions.
|
| 13 |
+
|
| 14 |
+
This repository publishes two runs that demonstrate the difference between:
|
| 15 |
+
- **Run A (Baseline Repro):** reproducible benchmark surface on COCO
|
| 16 |
+
- **Run B (Production Gate):** class-scoped operational gating under low-light conditions
|
| 17 |
+
|
| 18 |
+
This is not a benchmark leaderboard claim. It is an audit-friendly release record pattern.
|
| 19 |
+
|
| 20 |
+
---
|
| 21 |
+
|
| 22 |
+
## What a “Release Gate Record” contains
|
| 23 |
+
|
| 24 |
+
Each run includes:
|
| 25 |
+
- Pinned **model repo + exact commit revision**
|
| 26 |
+
- Pinned **dataset repo + exact commit revision**
|
| 27 |
+
- Declared **release predicate** (the gate)
|
| 28 |
+
- Deterministic **PASS/FAIL** container (certificate)
|
| 29 |
+
- Metrics JSON placeholder (filled when the run is executed)
|
| 30 |
+
- Paths reserved for a portable offline report and receipt bundle
|
| 31 |
+
|
| 32 |
+
---
|
| 33 |
+
|
| 34 |
+
## Runs
|
| 35 |
+
|
| 36 |
+
### Run A — Baseline Repro (COCO)
|
| 37 |
+
|
| 38 |
+
**Goal:** Bind a release decision to pinned, reproducible inputs on a benchmark surface.
|
| 39 |
+
|
| 40 |
+
- Model: `Ultralytics/YOLOv5` @ `553ca28641b8c2129596b0ecc50fc0902f7bd7ac` (`yolov5s.pt`)
|
| 41 |
+
- Dataset: `rafaelpadilla/coco2017` @ `7d5136d6952da2df12d99106e2cce2a65324617d` (split: `validation`)
|
| 42 |
+
- Gate: `coco_baseline_map50_floor_v1` (mAP@0.50 >= 0.36)
|
| 43 |
+
|
| 44 |
+
Folder: `run-a-baseline-repro/`
|
| 45 |
+
|
| 46 |
+
---
|
| 47 |
+
|
| 48 |
+
### Run B — Production Gate (Low-Light, class-scoped)
|
| 49 |
+
|
| 50 |
+
**Goal:** Demonstrate production-style gating: PASS/FAIL is defined by operational predicates, not benchmark bragging.
|
| 51 |
+
|
| 52 |
+
- Model: `Ultralytics/YOLOv5` @ `553ca28641b8c2129596b0ecc50fc0902f7bd7ac` (`yolov5s.pt`)
|
| 53 |
+
- Dataset: `SatwikKambham/ex-dark` @ `f2a15374dae8c90395f461c5ba0546c034751665`
|
| 54 |
+
- Scope: class-scoped to `person`
|
| 55 |
+
- Gate: `exdark_person_operational_gate_v1` at confidence 0.5
|
| 56 |
+
- precision >= 0.70
|
| 57 |
+
- recall >= 0.60
|
| 58 |
+
|
| 59 |
+
Folder: `run-b-production-gate/`
|
| 60 |
+
|
| 61 |
+
---
|
| 62 |
+
|
| 63 |
+
## How to verify the pinned inputs
|
| 64 |
+
|
| 65 |
+
You can fetch an exact revision of a repo using the `revision` parameter.
|
| 66 |
+
|
| 67 |
+
### Download this repo at a specific commit
|
| 68 |
+
|
| 69 |
+
```python
|
| 70 |
+
from huggingface_hub import snapshot_download
|
| 71 |
+
|
| 72 |
+
snapshot_download(
|
| 73 |
+
repo_id="LokoAI/release-gate-records",
|
| 74 |
+
repo_type="dataset",
|
| 75 |
+
revision="main"
|
| 76 |
+
)
|