Add newly completed evaluation runs

README.md

@@ -191,6 +191,8 @@ configs:
     path: data/kgw_delta0.7__llama31/attacks.parquet
   - split: thicket__gemma2
     path: data/thicket__gemma2/attacks.parquet
+  - split: textseal__gemma2
+    path: data/textseal__gemma2/attacks.parquet
   - split: none__gemma2
     path: data/none__gemma2/attacks.parquet
 - config_name: documents
@@ -276,7 +278,7 @@ The ELI5 protocol on other generators, one `<scheme>__<model>` split each in the
 | `textseal__llama31` | `meta-llama/Llama-3.1-8B-Instruct` | TextSeal (official code), mixing α = 0.1 | 5,000 | 200 | 500 | 300 | 1,600 | 1,400 |
 | `kgw_delta0.7__llama31` | `meta-llama/Llama-3.1-8B-Instruct` | KGW, γ = 0.25, δ = 0.7 (perplexity cost matched to Thicket's) | 5,000 | 200 | 500 | 300 | 1,600 | 1,400 |
 | `thicket__gemma2` | `google/gemma-2-9b-it` | Thicket (ours): keyed token-pair circle tilt with a likelihood price; κ = 6, price γ = 2 in answers (4 / 1 inside the reasoning delimiter), top-128 candidate shortlist, 256-label circle, first token and every 512-token restart neutral | 5,000 | 200 | 500 | 300 | 1,600 | 1,400 |
-| `textseal__gemma2` | `google/gemma-2-9b-it` | TextSeal (official code), mixing α = 0.1 | 5,000 | 200 | 500 | 300 | – | – |
+| `textseal__gemma2` | `google/gemma-2-9b-it` | TextSeal (official code), mixing α = 0.1 | 5,000 | 200 | 500 | 300 | 1,600 | – |
 | `none__gemma2` | `google/gemma-2-9b-it` | No watermark (plain sampling; the unwatermarked reference) | 5,000 | 200 | 500 | 300 | 1,600 | 1,400 |
 
 Each scheme used one fixed key; the keys are released in [`keys.json`](keys.json) and explained in [`KEYS.md`](KEYS.md), so the detection results can be reproduced with the project's code.

data/textseal__gemma2/attacks.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2f7a7cd25eccb264816b8e4549755682dd4ab21e27510a44c1bd3aadf6c24dfc
+size 1139532