Add safeagents judge bench results
Browse filesour safeagents judge bench v0 run, test split, unmodified benchmark runner, fixed 0.5 cut-off, no tuning. revision 52ba9e7a0040
balanced accuracy 0.756, AUROC 0.994, false-block rate 0.487, category macro F1 0.491
.eval_results/safeagents-judge-bench.yaml
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
- dataset:
|
| 2 |
+
id: harness-hub/safeagents-judge-bench
|
| 3 |
+
task_id: balanced_accuracy
|
| 4 |
+
value: 0.756
|
| 5 |
+
date: "2026-10-30"
|
| 6 |
+
source:
|
| 7 |
+
url: https://huggingface.co/datasets/harness-hub/safeagents-judge-bench
|
| 8 |
+
name: "safeagents judge bench v0, test split, run on 2 x NVIDIA A40 (sharded, CUDA bf16)"
|
| 9 |
+
- dataset:
|
| 10 |
+
id: harness-hub/safeagents-judge-bench
|
| 11 |
+
task_id: auroc
|
| 12 |
+
value: 0.994
|
| 13 |
+
date: "2026-10-30"
|
| 14 |
+
source:
|
| 15 |
+
url: https://huggingface.co/datasets/harness-hub/safeagents-judge-bench
|
| 16 |
+
name: "safeagents judge bench v0, test split, run on 2 x NVIDIA A40 (sharded, CUDA bf16)"
|
| 17 |
+
- dataset:
|
| 18 |
+
id: harness-hub/safeagents-judge-bench
|
| 19 |
+
task_id: false_block_rate
|
| 20 |
+
value: 0.487
|
| 21 |
+
date: "2026-10-30"
|
| 22 |
+
source:
|
| 23 |
+
url: https://huggingface.co/datasets/harness-hub/safeagents-judge-bench
|
| 24 |
+
name: "safeagents judge bench v0, test split, run on 2 x NVIDIA A40 (sharded, CUDA bf16)"
|
| 25 |
+
- dataset:
|
| 26 |
+
id: harness-hub/safeagents-judge-bench
|
| 27 |
+
task_id: category_macro_f1
|
| 28 |
+
value: 0.491
|
| 29 |
+
date: "2026-10-30"
|
| 30 |
+
source:
|
| 31 |
+
url: https://huggingface.co/datasets/harness-hub/safeagents-judge-bench
|
| 32 |
+
name: "safeagents judge bench v0, test split, run on 2 x NVIDIA A40 (sharded, CUDA bf16)"
|