diff options
Diffstat (limited to 'TRANSFORMER_CROSSOVER.md')
| -rw-r--r-- | TRANSFORMER_CROSSOVER.md | 16 |
1 files changed, 16 insertions, 0 deletions
diff --git a/TRANSFORMER_CROSSOVER.md b/TRANSFORMER_CROSSOVER.md index 8be1e72..019c0e6 100644 --- a/TRANSFORMER_CROSSOVER.md +++ b/TRANSFORMER_CROSSOVER.md @@ -181,6 +181,22 @@ T2 is a single-seed scaling panel, not the final uncertainty estimate. An untouched multi-seed T3 registry must contain every one of the same 27 cells; winner-only or SDIL-only replication cannot support the scaling claim. +The executable T2 contract is +`experiments/transformer_crossover_t2.py`. It refuses an incomplete T1 +selector, requires the selector to be committed, binds its hash, selected +rates, clean source, environment freeze, and GPU5/7 allowlist into a new launch +lock, and enumerates all 27 method--depth cells before sharding. Expensive EP, +Dual Propagation, and Forward--Forward cells are interleaved across shards. +`experiments/analyze_transformer_crossover_t2.py` requires all manifests, +retains timeout and nonfinite outcomes, rejects data/source/architecture/test +drift, and checks token presentations, candidate evaluations, relaxation +passes, local VJPs, logical queries, and enumerated MACs. The immutable +registry can be checked before T1 finishes with: + +```bash +python experiments/transformer_crossover_t2_smoke.py +``` + ## Metrics and cost accounting The primary task metric is full-validation next-token NLL; perplexity and |