uzuw commited on
Commit
f846bb5
·
verified ·
1 Parent(s): 8071679

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +22 -0
README.md CHANGED
@@ -46,3 +46,25 @@ checkpointing. Full log: [uzuw/laya-cli-gate](https://github.com/uzuw/laya-cli-g
46
  - Good **gate** (human reviews `safe:false`), not an auto-approver.
47
  - The checkpoint ships one out-of-range upstream temperature (`choice:11+`); `laya`
48
  clamps it with a warning — treat that bucket as uncalibrated.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
46
  - Good **gate** (human reviews `safe:false`), not an auto-approver.
47
  - The checkpoint ships one out-of-range upstream temperature (`choice:11+`); `laya`
48
  clamps it with a warning — treat that bucket as uncalibrated.
49
+
50
+ ## v2 — round 4 (subfolder `v2/`, 184 commands balanced 92 safe / 92 unsafe)
51
+
52
+ Root files above are unchanged (v1). Load v2 with:
53
+
54
+ ```python
55
+ from laya import Agent
56
+ agent = Agent("uzuw/laya-cli-gate", subfolder="v2")
57
+ ```
58
+
59
+ | | v1 (root, 65 cmds) | v2 (`v2/`, 184 cmds) |
60
+ |---|---|---|
61
+ | train tool acc | 0.985 | **0.995** |
62
+ | train safe acc @0.5 | 1.000 | **1.000** (@0.4–0.8) |
63
+ | train ECE | 0.029 | **0.015** |
64
+ | unseen safe (model only) | 0.75 | **10/12** |
65
+ | unseen safe (with deny net in `gate.py`) | — | **12/12, 0 FP** |
66
+
67
+ v2 adds template-swept unsafe rows (new pkgs/paths/flags per dangerous verb) and ships
68
+ alongside a deterministic deny net (`gate.py` in the code repo) that forces `safe: false`
69
+ on known-dangerous verbs — `--dry-run` exempts. Same honest limits apply: use as a
70
+ **gate** (human reviews `safe:false`), not an auto-approver.