mindbomber commited on
Commit
7bb5b4a
·
verified ·
1 Parent(s): 5940b95

Refresh AANA demo Space links

Browse files
Files changed (2) hide show
  1. README.md +5 -0
  2. app.py +8 -0
README.md CHANGED
@@ -52,3 +52,8 @@ In that full PIIMB run, Presidio-only average masking F2 was `0.4492985573` and
52
  Presidio + AANA average masking F2 was `0.5629171363`, an absolute gain of
53
  `+0.1136185790`.
54
 
 
 
 
 
 
 
52
  Presidio + AANA average masking F2 was `0.5629171363`, an absolute gain of
53
  `+0.1136185790`.
54
 
55
+ Recent public result artifacts:
56
+
57
+ - Grounded gate calibration: https://huggingface.co/datasets/mindbomber/aana-grounded-gate-calibration
58
+ - HarmActions agent-action gate: https://huggingface.co/datasets/mindbomber/aana-harmactions-agent-action-gate
59
+ - GAP post-hoc gate: https://huggingface.co/datasets/mindbomber/aana-gap-posthoc-gate
app.py CHANGED
@@ -7,6 +7,9 @@ import gradio as gr
7
 
8
  PIIMB_LINK = "https://huggingface.co/datasets/piimb/pii-masking-benchmark-results/discussions/3"
9
  AANA_MODEL_LINK = "https://huggingface.co/mindbomber/aana"
 
 
 
10
 
11
 
12
  @dataclass
@@ -160,6 +163,9 @@ def aana_gate(candidate: str, evidence: str, constraints: str):
160
  "links": {
161
  "canonical_aana_model": AANA_MODEL_LINK,
162
  "piimb_ablation": PIIMB_LINK,
 
 
 
163
  },
164
  }
165
  return result.action, result.aix_score, "\n".join(result.hard_blockers) or "None", result.suggested_revision, json.dumps(audit, indent=2)
@@ -175,6 +181,8 @@ Check a candidate answer/action against evidence and constraints. This minimal d
175
  Canonical model card: [{AANA_MODEL_LINK}]({AANA_MODEL_LINK})
176
 
177
  PIIMB ablation: Presidio + AANA improved average masking F2 by `+0.1136185790` over Presidio-only. [Review the official submission]({PIIMB_LINK}).
 
 
178
  """
179
  )
180
  with gr.Row():
 
7
 
8
  PIIMB_LINK = "https://huggingface.co/datasets/piimb/pii-masking-benchmark-results/discussions/3"
9
  AANA_MODEL_LINK = "https://huggingface.co/mindbomber/aana"
10
+ GROUNDING_LINK = "https://huggingface.co/datasets/mindbomber/aana-grounded-gate-calibration"
11
+ HARMACTIONS_LINK = "https://huggingface.co/datasets/mindbomber/aana-harmactions-agent-action-gate"
12
+ GAP_LINK = "https://huggingface.co/datasets/mindbomber/aana-gap-posthoc-gate"
13
 
14
 
15
  @dataclass
 
163
  "links": {
164
  "canonical_aana_model": AANA_MODEL_LINK,
165
  "piimb_ablation": PIIMB_LINK,
166
+ "grounded_gate_calibration": GROUNDING_LINK,
167
+ "harmactions_agent_action_gate": HARMACTIONS_LINK,
168
+ "gap_posthoc_gate": GAP_LINK,
169
  },
170
  }
171
  return result.action, result.aix_score, "\n".join(result.hard_blockers) or "None", result.suggested_revision, json.dumps(audit, indent=2)
 
181
  Canonical model card: [{AANA_MODEL_LINK}]({AANA_MODEL_LINK})
182
 
183
  PIIMB ablation: Presidio + AANA improved average masking F2 by `+0.1136185790` over Presidio-only. [Review the official submission]({PIIMB_LINK}).
184
+
185
+ Public benchmark artifacts: [grounded gate calibration]({GROUNDING_LINK}), [HarmActions agent-action gate]({HARMACTIONS_LINK}), [GAP post-hoc gate]({GAP_LINK}).
186
  """
187
  )
188
  with gr.Row():