Update README leaderboard for April records

[cocohearts]

Summary

Updates the README leaderboard for the accepted / expected April 10min_16mb record sequence:

• PR Record: ParallelResiduals + MiniDepthRecurrence, 1.1063 BPB / 1.8679 nats, -0.0072 vs PR #1179, -0.0143 vs merged SOTA #1204: Parallel Residuals + Mini Depth Recurrence, 1.1063
• PR Record: 4096-Vocab + 4.0-MLP-mult + 0.085-WD + Simplifications — val_bpb 1.09785 (3-seed mean) #1218: 4096-Vocab + Larger Model + High WD + Simplifications, 1.0979
• PR Record: MuonEq-R + Depth Recurrence + WD=0.090 + All-Int6 GPTQ — val_bpb 1.0912 (3-seed mean) #1285: MuonEq-R + Depth Recurrence + WD=0.090 + All-Int6 GPTQ, 1.0912
• PR Record: SP4096 + Depth Recurrence + Parallel Residuals + MuonEq-R + QK-Gain 5.0 — val_bpb 1.0897 (3-seed mean) #1334: SP4096 + Depth Recurrence + Parallel Residuals + MuonEq-R, 1.0897
• PR Record: SP8192 + GPTQ Embeddings + Depth Recurrence + MuonEq-R + SDClip — val_bpb 1.08563 (5 seed mean) #1394: SP8192 + GPTQ Embeddings + Depth Recurrence + SDClip, 1.0856
• PR Record: SP8192 + Parallel Residuals + Hessian-Aware SDClip — val_bpb 1.08354 (3-seed mean) #1412: SP8192 + Parallel Residuals + Hessian-Aware SDClip, 1.0835
• PR Record: SP8192 + QK-Gain 5 + Legal Score-First TTT — val_bpb 1.08279 (3-seed mean) #1413: SP8192 + QK-Gain 5 + Legal Score-First TTT, 1.0828
• PR Record: SP8192 + Parallel Residuals + Score-First TTT — val_bpb 1.0822 (3-seed mean) #1477: SP8192 + Parallel Residuals + Score-First TTT, 1.0822
• PR Record: SP8192 + 3-Layer Recurrence + Parallel Residuals + QK-Gain 5.25 + Legal TTT — val_bpb 1.0810 (3-seed mean) #1493: SP8192 + 3-Layer Recurrence + Parallel Residuals + Legal TTT, 1.0810

Each new row follows the existing leaderboard summary format: On PR #...: ... plus the brief incremental contribution.

Sequencing note

This README row update expects the record folder from PR #1412 to be present by the time this lands. The current main checkout used to prepare this PR already contains the record folders for #1204, #1218, #1285, #1334, #1394, #1413, #1477, and #1493.

Validation

• Ran a parser over the README leaderboard table and verified all table rows have the expected 6 markdown-table columns.
• Ran git diff --check.
• Verified README-only diff.

[Robby955]

Hey @cocohearts — thanks for putting the leaderboard together!

Small ask: would you mind including PR #1412 in the sequence? It scored 1.08354 (3-seed mean), which was an improvement over PR #1394's 1.0856 at the time of submission. It introduced parallel residuals and Hessian-aware SDClip, parallel residuals were then adopted by both #1477 and #1493 in their stacks.

It would slot between #1394 and #1413:

• PR Record: SP8192 + GPTQ Embeddings + Depth Recurrence + MuonEq-R + SDClip — val_bpb 1.08563 (5 seed mean) #1394: SP8192 + GPTQ Embeddings + Depth Recurrence + SDClip, 1.0856
• PR Record: SP8192 + Parallel Residuals + Hessian-Aware SDClip — val_bpb 1.08354 (3-seed mean) #1412: SP8192 + Parallel Residuals + Hessian-Aware SDClip, 1.08354
• PR Record: SP8192 + QK-Gain 5 + Legal Score-First TTT — val_bpb 1.08279 (3-seed mean) #1413: SP8192 + QK-Gain 5 + Legal Score-First TTT, 1.0828

I had originally titled it "Non-record" being safe about the gap size, but the score did improve on the previous best, and I feel other people just were putting RECORD when it wasn't validated truly. I've updated the title to reflect that.

Thanks for considering!

[cocohearts]

@Robby955 taking a look!

[cocohearts]

@Robby955 thank you for pointing this out!