Uh oh!
There was an error while loading. Please reload this page.
fix(prism): ban non-causal TokenMix label-leak class - #95
Conversation
Catch MLP-Mixer/TokenMix sequence mixers without a causal mask before pod rent (static screen + agentic code), and document the causal LM ban for miners.
Uh oh!
There was an error while loading. Please reload this page.
Caution Review failedThe pull request is closed. ℹ️ Recent review info⚙️ Run configurationConfiguration used: defaults Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (11)
📝 WalkthroughWalkthroughThe PR adds AST detection for non-causal sequence mixing, integrates the finding into agentic simulator verdicts, classifies sub-1.0 BPB results, updates prompt rules, and documents the causal next-token contract. ChangesNon-causal label-leak detection
Estimated code review effort: 4 (Complex) | ~45 minutes Sequence Diagram(s)sequenceDiagram
participant Submission
participant static_source_cheat
participant Simulator
participant CheatCode
Submission->>static_source_cheat: provide architecture and training sources
static_source_cheat-->>Simulator: return source-cheat finding
Simulator->>CheatCode: map non-causal finding to NonCausalLabelLeak
Submission->>Simulator: provide evaluation metrics
Simulator->>CheatCode: classify BPB below 1.0 as NonCausalLabelLeak
Possibly related PRs
✨ Finishing Touches 💡 1🛠️ Fix failing CI checks 💡
📝 Generate docstrings
🧪 Generate unit tests (beta)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
Summary
TokenMixdense sequence mixers without a causal mask before pod rent (staticNonCausalLabelLeakscreen + agentic cheat code).< 1.0) as the same label-leak class in sim/agentic.docs/PRISM.md, recipe, external-miner).Triggered by prod exploit
b99a7047(bpb≈0.23, score ~813k in ~12m) — already operator-revoked to Score(0)/rejected.Test plan
cargo test -p challenge-ast --lib source_cheatscargo test -p challenge-agentic sim_prism_tokenmixcargo test -p challenge-agentic sim_absurd_low_bpbSummary by CodeRabbit
New Features
non_causal_label_leakresult across cheat classifications and verdict reporting.Documentation
Tests