Skip to content

docs: ban non-causal tokenmix label-leak architectures #8

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
echobt merged 1 commit into from
Aug 8, 2026
+11 −0
Merged

docs: ban non-causal tokenmix label-leak architectures #8

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
11 changes: 11 additions & 0 deletions docs/scoring.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -20,6 +20,17 @@ submission is coherent and not cheating; their quality notes never move the scor
- Probe the cheap gate first with `POST /v1/submissions/precheck` (3/coldkey/UTC
day) — see [Submit](submit.md#precheck-similarity-before-you-submit).


## Causal LM contract (banned: non-causal label leak)

Prism scores **next-token** cross-entropy → BPB. Architectures must not let
position `t` read tokens `t+1…` (including the label). Dense sequence mixers —
MLP-Mixer-style `TokenMix` / `t_mix` / `nn.Linear` over the full time axis after
`transpose(1, 2)` — **without** a causal mask (`triu` / `tril` / `is_causal` /
attention mask) are a hard ban (`non_causal_label_leak`, `Score(0)`, terminal,
often caught **before** GPU rent). Channel mixing and causal attention / causal
conv are fine; bidirectional full-sequence mixes used as a next-token LM are not.

## Architecture competition (emission math)

Per epoch, your emission is the **max** of:
Expand Down
Loading