Record: Full GPTQ + LeakyReLU² + Parallel Muon (3-seed mean 1.1180)

[kshitizz36]

Results

• 3-seed mean val_bpb (sliding, stride=64): 1.11800697
• std (sample): 0.00102882
• mean size: 15,931,864 bytes
• hardware: 8xH100 SXM, 600s train budget

Seed	step_avg	steps	Pre-quant bpb	Post-GPTQ sliding bpb	Total size (bytes)
42	84.98ms	7062	1.1384	1.11752350	15,890,732
1337	86.65ms	6925	1.1402	1.11918848	15,891,780
2024	85.20ms	7043	1.1383	1.11730894	16,013,080
Mean	85.61ms	7010	1.1390	1.11800697	15,931,864

Notes

• No TTT used.
• Full logs for all 3 seeds are included under this record folder.
• Independent 3-seed run of the same Full GPTQ + LeakyReLU² + Parallel Muon direction.

[valerio-oai]

This submission uses GPTQ at test time, hence using training data at eval time, which is disallowed. Closing for now.