You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 35af511
Browse filesBrowse the repository at this point in the historyBrowse files
docs: replaced remaining Guard.with_tiny() references with Guard(mode… (UnplugAI#130)
* docs: replaced remaining Guard.with_tiny() references with Guard(model='tiny')
ClosesUnplugAI#118
Note: the issue cited README.md:60, but the root README had three total
with_tiny references (quickstart, 'What ships today' table, and the benchmark
paragraph). Updated all three for consistency within the file. MIGRATION.md
and the deprecation test left untouched as instructed.
* fix: fail loudly on ML load failure
* dropped require_ml as per review
* fix: removed leftover conflict marker and duplicate paragraph
* chore: re-trigger Greptile on current clean head
Copy file name to clipboardExpand all lines: README.md
+5-3Lines changed: 5 additions & 3 deletions
Display the source diff
Display the rich diff
Original file line number
Diff line number
Diff line change
@@ -58,7 +58,7 @@ print(result.findings) # evidence with span offsets
58
58
One line upgrades detection to the ML span model (downloads [unplug-tiny-v1](https://huggingface.co/Unplug-AI/unplug-tiny-v1) once, cached):
Try it without installing anything: [live demo](https://huggingface.co/spaces/Unplug-AI/unplug-tiny-demo).
@@ -71,17 +71,19 @@ Try it without installing anything: [live demo](https://huggingface.co/spaces/Un
| ML span model `Guard.with_tiny()`|**Preview** ([unplug-tiny-v1](https://huggingface.co/Unplug-AI/unplug-tiny-v1)) |
74
+
| ML span model `Guard(model="tiny")`|**Preview** ([unplug-tiny-v1](https://huggingface.co/Unplug-AI/unplug-tiny-v1)) |
75
75
| Sliding-window long documents + streaming scan |**Included**|
76
76
77
-
On the neuralchemy prompt-injection set, regex-only detection reaches **F1 0.58 / recall 0.41** — a fast first line, not sufficient alone. Adding the ML span model (`Guard.with_tiny()`) takes that to **F1 0.99 / recall 0.98**, and lifts recall on *indirect* injection from **0.05 → 0.91**. False-positive rate stays under 1% on the injection set (2.1% on a separate hard-benign corpus). Full tables, methodology, and honest caveats: [`sdk/docs/BENCHMARKS.md`](sdk/docs/BENCHMARKS.md). Per-axis model metrics (including failure modes) are on the [model card](https://huggingface.co/Unplug-AI/unplug-tiny-v1).
77
+
78
+
On the neuralchemy prompt-injection set, regex-only detection reaches **F1 0.58 / recall 0.41** — a fast first line, not sufficient alone. Adding the ML span model (`Guard(model="tiny")`) takes that to **F1 0.99 / recall 0.98**, and lifts recall on *indirect* injection from **0.05 → 0.91**. False-positive rate stays under 1% on the injection set (2.1% on a separate hard-benign corpus). Full tables, methodology, and honest caveats: [`sdk/docs/BENCHMARKS.md`](sdk/docs/BENCHMARKS.md). Per-axis model metrics (including failure modes) are on the [model card](https://huggingface.co/Unplug-AI/unplug-tiny-v1).
78
79
79
80
**Language support.** Regex + normalization detection is tuned for English today.
80
81
Ordinary non-English input (Cyrillic, CJK, etc.) is *not* treated as an evasion
81
82
signal — only genuine zero-width/bidi control characters and mixed-script
82
83
homoglyph smuggling are flagged. Robust multi-language injection detection is
83
84
tracked as a separate work item.
84
85
86
+
85
87
## Agent host checklist
86
88
87
89
1. Scan user input: `guard.scan(text, source="user")`
0 commit comments