AnaWut commited on
Commit
f44c2b0
·
verified ·
1 Parent(s): 9b50f70

Upload 3 files

Browse files
Files changed (3) hide show
  1. README.md +17 -0
  2. probe.json +0 -0
  3. run_config.json +15 -0
README.md ADDED
@@ -0,0 +1,17 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ base_model: CohereLabs/tiny-aya-global
3
+ tags:
4
+ - activation-probe
5
+ - hallucination-detection
6
+ - citation-grounding
7
+ ---
8
+
9
+ # Tiny Aya citation grounding probe
10
+
11
+ This artifact contains the standardized logistic regression coefficients, intercept, normalization parameters, and frozen threshold learned from Tiny Aya Global citation drafts. `run_config.json` records the model revision, activation layer, and runtime settings. The Hugging Face base model is loaded separately by the GitHub notebook and is governed by its upstream terms.
12
+
13
+ For a draft activation vector, standardize with the saved mean and scale. Compute the sigmoid of the dot product with the saved weights plus intercept, then abstain when the score is above `thresholds.probe_risk`.
14
+
15
+ The probe was evaluated on a controlled English bibliography benchmark. Its output is a decision score with no guarantee of citation existence. See the GitHub notebook and held-out report for the data split and limitations.
16
+
17
+ Measured holdout: 0/24 unsupported releases, 11/12 correct evidence-backed citations, and 23/24 AuditKIT exact match. Token confidence matched these aggregate scores.
probe.json ADDED
The diff for this file is too large to render. See raw diff
 
run_config.json ADDED
@@ -0,0 +1,15 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "model": "CohereLabs/tiny-aya-global",
3
+ "model_revision": null,
4
+ "seed": 42,
5
+ "layer": 18,
6
+ "max_new_tokens": 24,
7
+ "min_supported_retention": 0.8,
8
+ "packages": {
9
+ "curatorkit": "0.1.26",
10
+ "auditkit": "1.1.10",
11
+ "transformers": "5.22.2",
12
+ "torch": "2.14.1+cu130",
13
+ "scikit-learn": "1.6.1"
14
+ }
15
+ }