id stringlengths 36 36 | tier stringclasses 1
value | github_login stringclasses 179
values | created_at stringlengths 32 32 | status stringclasses 2
values | sha256 stringlengths 64 64 | source_bytes int64 341 260k | leaderboard_rank int64 | source large_stringlengths 341 260k | metadata_json large_stringlengths 645 734 | result_json large_stringlengths 5 4.8k |
|---|---|---|---|---|---|---|---|---|---|---|
24f04164-539f-478f-b7e1-bb5bf6772eca | easy | khushidahi | 2026-08-15 13:21:56.227975+00:00 | succeeded | 151a13bbb0646c985c205eff61d013a7c250e871aee1cd046089b7e0e4f5c6a0 | 25,305 | null | """R6 structured decimal-workspace model for One Layer Deeper.
The model does not implement multiplication, modular reduction, or the public
recurrence. It uses the public tokenizer structure to build learned, right-
aligned decimal tapes for N, X, and T, then applies a shared neural transition
to a mutable work tape.... | {
"id": "24f04164-539f-478f-b7e1-bb5bf6772eca",
"created_at": "2026-08-15 13:21:56.227975+00:00",
"db_md5": "8b53df6152631e18ea2b8608b24cd010",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "3d3650ec-07d9-4b3b-abb8-f6d2df5b90a4",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.148188829421997,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007499999832361937
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.150066375732422,
"example_count": 600,... |
24f72520-72a0-4a75-9381-c3ba27eff927 | easy | chad-atexpedient | 2026-08-08 23:02:05.924495+00:00 | succeeded | 5202b43f53028599060b580b3658bb95bde5955276ed0d0a0967eb614603746e | 33,619 | null | """Encoder-decoder Adaptive Universal Transformer.
Architecture references:
- https://arxiv.org/abs/1807.03819v3
- https://github.com/tensorflow/tensor2tensor/blob/master/tensor2tensor/models/research/universal_transformer_util.py
Tensor shape suffixes use the following dimension key:
- B: batch size
- S: source sequ... | {
"id": "24f72520-72a0-4a75-9381-c3ba27eff927",
"created_at": "2026-08-08 23:02:05.924495+00:00",
"db_md5": "97d2f1fdfc7848c030ff44f612770692",
"submitter": "chad-atexpedient",
"github_login": "chad-atexpedient",
"run_id": "40be7677-b855-4f32-9bcf-6cc4ff4d354e",
"tier": "easy",
"dataset_id": "e5",
"st... | {
"score": {
"mean_loss": 2.391306030293602,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0054166666790843015
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.4323323489304616,
"example_count": 60... |
24fbc1a7-89b4-44ab-82cd-68944370288b | easy | shirvani-jr | 2026-08-22 13:15:28.983586+00:00 | succeeded | 1398bd08847c5eaab4959d387d99f9b6d5d587478909c52a6d433aa99a043c89 | 23,140 | null | """Looped exact-arithmetic digit machine.
The recurrent state is the decimal digit string of the current residue. One
loop application performs one modular squaring, so depth extrapolates in T by
construction. Every functional primitive — digit-by-digit product, full
adder, full subtractor, digit comparator, and com... | {
"id": "24fbc1a7-89b4-44ab-82cd-68944370288b",
"created_at": "2026-08-22 13:15:28.983586+00:00",
"db_md5": "a1310d423bb5360fa56d444443de24fa",
"submitter": "Ali",
"github_login": "shirvani-jr",
"run_id": "70f70bd7-2460-45ef-9696-c71ab485f85a",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded"... | {
"score": {
"mean_loss": 0.029165864922106266,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.02915062941610813,
"example_count": 800,
"... |
2504b685-df40-43c8-8f7e-637a876bf29c | easy | viridale | 2026-08-14 00:06:45.528428+00:00 | succeeded | 574e5be6294997beb80825b0ab4445bf399ebfda5eb710fa17f6bc611a328152 | 13,775 | null | """Shifted complete-endpoint partial-label likelihood.
hypothesis: same-depth final-string equality reveals only endpoint fibers. Equality
across a supplied depth offset directly constrains a power of the learned action:
P^delta(x_i) must place mass on the set of observed sources at the shallower depth
wi... | {
"id": "2504b685-df40-43c8-8f7e-637a876bf29c",
"created_at": "2026-08-14 00:06:45.528428+00:00",
"db_md5": "b808e5cf570b9355fc4004a1c8cf979d",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "74702a98-a261-4d48-9cda-b7d5a5bfe384",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 1.7550853490829468,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.29500000923871994
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2612040042877197,
"example_count": 100... |
250c8edc-cd03-40a0-ba79-9f410831a1db | easy | nikhilkumar26 | 2026-08-30 02:48:07.573354+00:00 | succeeded | 9479adbc70f42ea0cd2c49f1367340c2309e212a41376413dc69829c4f8ec835 | 11,065 | null | """T-adaptive residue TRM: Fourier MLP square-map, detached unroll.
T is the unroll count only. Each step classifies the next residue in 0..N-1.
Digit logits and residue logits come from the last active step. Recurrence
state is detached so T=1 can train a stable one-step map that then composes.
"""
from __future__ i... | {
"id": "250c8edc-cd03-40a0-ba79-9f410831a1db",
"created_at": "2026-08-30 02:48:07.573354+00:00",
"db_md5": "39e731eae6ab10e11412d0f0dce96015",
"submitter": "Nikhil Kumar",
"github_login": "nikhilkumar26",
"run_id": "67810c25-e810-4154-a15b-cc2746ea6bb8",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 0.5877142548561096,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.8649999797344208
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.6076341271400452,
"example_count": 100,... |
2511cc53-d5f6-4a07-a76e-94c0ccadef10 | easy | rishivg | 2026-08-03 23:30:46.235009+00:00 | succeeded | 36d4cfcebeb34cdfd09d68e8163db56eb1eda69eec2a6d118b84e46c7dbea90b | 97,813 | null | """Looped transformer with place-value coordinates, Muon, and a clock-paced LR.
Design notes, in rough order of expected value:
1. Weight-tied recurrence with input injection. The task is function
composition of depth ``T``, so one block applied ``LOOPS`` times is a much
better prior than a single feed-forw... | {
"id": "2511cc53-d5f6-4a07-a76e-94c0ccadef10",
"created_at": "2026-08-03 23:30:46.235009+00:00",
"db_md5": "a47a0d5d526eda00b20a7a3d88b6be1e",
"submitter": "Rishi Gottumukkala",
"github_login": "rishivg",
"run_id": "41afd5c2-1839-4e89-a6f9-860cc710d851",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 10.406838417053223,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.08499999716877937
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 10.968704223632812,
"example_count": 100... |
2513a962-131e-4e39-848b-f0bdd9d7b348 | easy | khushidahi | 2026-08-03 20:13:46.222593+00:00 | succeeded | 8390d6a430c96b7cf7228bfe6795a1bf1ad71b789899792cad378c9769d22a20 | 16,594 | null | """R6 structured decimal-workspace model for One Layer Deeper.
The model does not implement multiplication, modular reduction, or the public
recurrence. It uses the public tokenizer structure to build learned, right-
aligned decimal tapes for N, X, and T, then applies a shared neural transition
to a mutable work tape.... | {
"id": "2513a962-131e-4e39-848b-f0bdd9d7b348",
"created_at": "2026-08-03 20:13:46.222593+00:00",
"db_md5": "66f49872276ab6962e072b1dead82d82",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "fa1785e5-f37f-4a66-8656-399d96eb713c",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.1733157634735107,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006250000325962901
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1704580783843994,
"example_count": 60... |
25185f60-1434-4c7e-9278-3c16ff62f059 | easy | chuk-uzowihe | 2026-08-12 21:45:46.630665+00:00 | succeeded | 6ee0189f1c8679b6042bf6b680f5a1bf291657b030889f49bc68bf42132353b7 | 26,861 | null | """v10: dense fixed-loop recurrence with deep supervision.
Lineage: FPRM structure (alpha/beta-tied residual scaling, pre-rms_norm,
input injection every iteration) + the v8 2D digit lattice / RoPE task
encoding. Three deliberate departures from the v9 MoEUT line:
1. **Full BPTT.** The loop runs a fixed N_ITER iterat... | {
"id": "25185f60-1434-4c7e-9278-3c16ff62f059",
"created_at": "2026-08-12 21:45:46.630665+00:00",
"db_md5": "db8a513b208695c4345b1d7a2a10259f",
"submitter": "Chuk Uzowihe",
"github_login": "chuk-uzowihe",
"run_id": "7a316117-80da-433d-9375-6aa566b21ec0",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 24.81662368774414,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.07999999821186066
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 24.443439483642578,
"example_count": 100,... |
251b1bc9-2ff9-48a0-b79b-a2a4f7af2674 | easy | Maxwells-Demons | 2026-08-14 04:36:57.093888+00:00 | succeeded | 603d4ba4882053d227c43b1f68b7f7e412c2e31161883c7c83c9158a43bb87d4 | 121,284 | null | """One shared digit operator, its two free neighbours of x, and two learned maps.
The backbone is unchanged from v47/v96: the state IS the digit one-hot, one
application of the operator is
(state digits, N digits) -> next state digit logits
and the straight-through code of those logits is the whole state handed to... | {
"id": "251b1bc9-2ff9-48a0-b79b-a2a4f7af2674",
"created_at": "2026-08-14 04:36:57.093888+00:00",
"db_md5": "da4c9ca9f4baba803becf2c6c85d912c",
"submitter": "Maxwells_Ayakashi",
"github_login": "Maxwells-Demons",
"run_id": "6c9f23f1-ffbd-44b8-97e9-020c7cd8939a",
"tier": "easy",
"dataset_id": "e3",
"st... | {
"score": {
"mean_loss": 0.07437863270516945,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.9975000071525574
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.0858312709101068,
"example_count": 800... |
252770dc-e014-4621-91de-c4726d1623ea | easy | viridale | 2026-08-22 21:35:05.594583+00:00 | succeeded | 6a5e395c12172d703f586d4da885281e54d7750ebff65b332750f0d4a987d475 | 119,695 | null | """NV060: complete, cost-bounded two-patch completion search.
hypothesis: v811 can execute two generic patches but changes only one patch site in
every live completion bank, so exact two-edit programs remain behind a measured
local-loss valley. Jointly search identical patch pairs with every r2 anchor and
... | {
"id": "252770dc-e014-4621-91de-c4726d1623ea",
"created_at": "2026-08-22 21:35:05.594583+00:00",
"db_md5": "b91abdf2c6e7ba56978035c168e89fb0",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "94eea3a4-4c20-4b3c-9835-8ca5051b2d15",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 2.3841855778528043e-07,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.3841855067985307e-07,
"example_count": 100,
... |
252c59fc-7799-4a4d-bea6-5940d39071cc | easy | DDanlov | 2026-08-17 08:10:06.228910+00:00 | succeeded | c8ff31b7d52d5cc9b0631b39d830af8228e8036a5a21d5c9f8fae6f0148744fa | 30,801 | null | """
Scalable Recurrent Deliberation Transformer (ScalableDEPHNet)
============================================================
Architecture:
- Pure No-Norm: Single unit RMS normalization strictly isolated to the input token embedding layer.
Zero LayerNorm / RMSNorm inside recurrent deliberation blocks.
- Pure No-Skip... | {
"id": "252c59fc-7799-4a4d-bea6-5940d39071cc",
"created_at": "2026-08-17 08:10:06.228910+00:00",
"db_md5": "6c8e8c590bb5fd129737f737f494df90",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "83ce7232-32e3-432f-9c56-2f1da8bf73d9",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.146443905527158,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0766666637857755
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.3229990005493164,
"example_count": 100,
... |
252ebfad-36a5-4a28-aad8-96dc7b0b0554 | easy | ArkinDharawat | 2026-08-10 10:58:21.740419+00:00 | succeeded | 0ea86a3e7ba4f21d84132cc18ea5185e4c2ded0ee4b488260d0458dc0b92d3ae | 5,557 | null | """Universal-Transformer-style looped block with learnable scratchpad padding.
Inspired by "Expressive Power of Log Depth Transformers": trade shared-block
recurrence and extra padding positions for depth. A single Block is applied
NUM_LOOPS times over the input augmented with N_PAD learnable scratchpad
tokens; only t... | {
"id": "252ebfad-36a5-4a28-aad8-96dc7b0b0554",
"created_at": "2026-08-10 10:58:21.740419+00:00",
"db_md5": "07a069650a4719a0725e6fde5f6e41f8",
"submitter": "Arkin Dharawat",
"github_login": "ArkinDharawat",
"run_id": "3ddb8be9-ac71-40fe-bb5f-e9cd2bede30c",
"tier": "easy",
"dataset_id": "e1",
"status"... | {
"score": {
"mean_loss": 2.367687165737152,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03500000014901161
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.741486072540283,
"example_count": 100,
... |
25301fd1-f348-46bf-900d-48adf1bcf8a6 | easy | erdavis0 | 2026-08-25 02:04:31.287375+00:00 | succeeded | 35ff275a7c933b6d0b5e1dc217d55253aa63433308fe14fd797471c38b44cb91 | 24,148 | null | """Matched categorical-boundary panel: C3 ST oracle selector.
Public delimiter syntax is used only to route opaque token identities onto
separate right-aligned tapes. One T-blind learned macro maps a categorical X
tape and a digitwise N tape to a new categorical tape through freshly reset
interaction, carry, and modu... | {
"id": "25301fd1-f348-46bf-900d-48adf1bcf8a6",
"created_at": "2026-08-25 02:04:31.287375+00:00",
"db_md5": "5fb89a7c7ac87a7ea6e553c7a9be46f2",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "8e725c18-1b66-48e3-b89f-3f716646a260",
"tier": "easy",
"dataset_id": "e9",
"status": "succeeded",... | {
"score": {
"mean_loss": 1.8675472736358643,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.013333333656191826
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.8735628128051758,
"example_count": 90... |
253503a6-c551-4bdf-821b-46f78c6fdcc7 | easy | 0xDEADD06 | 2026-08-12 13:35:53.833342+00:00 | succeeded | 0651a9ecf0611de0c6585ec5c8b232cb31a2ad1138512668ee8b77751f0acf6a | 7,613 | null | """ParaLSTM: recurrent LSTM with weight-tied passes for algorithmic tasks."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import nn, Tensor
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_mode... | {
"id": "253503a6-c551-4bdf-821b-46f78c6fdcc7",
"created_at": "2026-08-12 13:35:53.833342+00:00",
"db_md5": "490466892d6f6995c22e95866394097d",
"submitter": "0xDEADD06",
"github_login": "0xDEADD06",
"run_id": "9ae4620d-5d12-4f18-8e62-3e16c0e8327d",
"tier": "easy",
"dataset_id": "e5",
"status": "succee... | {
"score": {
"mean_loss": 5.865648420780039,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005833333507180214
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.953653916382469,
"example_count": 600,... |
253cfc0c-5714-49f5-83a5-bd1cceacf86f | easy | oupadhyay | 2026-08-11 08:25:33.460072+00:00 | succeeded | 3b1c06d7df10e7c87a4e482dbd5b12624023a166544080ac1ec20a5fe496b1a6 | 7,185 | null | """One-block spatial-gating categorical register with a learned highway gate."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import ModelSpec, OptimizerBundle, OptimizerSpec, Submission, assert_model_state
PAD, N_MARK, X_MARK, T_MARK, DI... | {
"id": "253cfc0c-5714-49f5-83a5-bd1cceacf86f",
"created_at": "2026-08-11 08:25:33.460072+00:00",
"db_md5": "20c11341b86b187de97d0dca7c825f9f",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "cd89074d-9879-4e60-b9af-00aa693a9f79",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.177824229545148,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005833333358168602
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1483260216734332,
"example_count": 600... |
2546a4a3-3fd4-4600-968e-cb9badf49076 | easy | liam-gb | 2026-08-13 02:25:14.434555+00:00 | succeeded | 2e9df07bf71dc171766b494c72f5b91e6381f20e012c1b7a4b542be81aeb6dae | 15,040 | null | """rns_pm_gather_ct_widet_norelrt: tables at 3e-2 AND batch reuse off.
Base: widet with lookup tables at lr 3e-2.
The wide readout is what makes long answers reachable; the trim is what pays
for it, by not running loop iterations whose update is multiplied by zero.
Base docstring:
Single change vs rns_pm_gather_ct:... | {
"id": "2546a4a3-3fd4-4600-968e-cb9badf49076",
"created_at": "2026-08-13 02:25:14.434555+00:00",
"db_md5": "a06277be536373d81f9d0a6e692bd1a6",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "d011ae98-b2b6-4beb-8291-c4f23e135110",
"tier": "easy",
"dataset_id": "e4",
"status": "succeeded"... | {
"score": {
"mean_loss": 6.454700477894473,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0504166671137015
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.823905365111879,
"example_count": 1200,
... |
254bfc41-bd32-4267-9e7e-cc30cdc23656 | easy | viridale | 2026-08-15 20:20:38.003448+00:00 | succeeded | 6183f84d4a0272d5f6d2935b91b50fcec9a9a1bf6342fd3e89fe803d29f8ed1e | 25,865 | null | """Exact learned boundary section plus anchored neural fiber.
hypothesis: parameterizing the learned task law as B(N,x) plus the exact anchored
residual G(s,N,x,T,k)-G(anchor,N,x,1,0) prevents every deep/nonautonomous update
from changing T1, while two evaluator-owned passes let final-answer evidence
train... | {
"id": "254bfc41-bd32-4267-9e7e-cc30cdc23656",
"created_at": "2026-08-15 20:20:38.003448+00:00",
"db_md5": "cf6f871f24f75fedf9b1d3b51967f2b2",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "518f0a17-29d5-4bec-af66-935474b3812c",
"tier": "easy",
"dataset_id": "e10",
"status": "succ... | {
"score": {
"mean_loss": 14.736549854278564,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.08270706795156002
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 15.842170715332031,
"example_count": 125... |
25504437-6e89-4e33-b159-720e6e0d433d | easy | KaustubhKumar05 | 2026-08-29 04:57:31.863870+00:00 | succeeded | c8d3923eb75c1eecba0eb8a07243f8e636cf211893f6fa66c68ea72498b75bf3 | 14,341 | null | """op2_t1w -- stock cell_op2 + T=1 loss upweighting (the scored rung).
Forked from cell_op2. ORIGINAL DOC BELOW.
cell_op2 -- ONE shared squaring operator with internal depth, applied T times.
WHY THIS SHAPE (verified on the real data, not assumed):
answer(T=t) == x squared t times mod N 20000 / 20000 e... | {
"id": "25504437-6e89-4e33-b159-720e6e0d433d",
"created_at": "2026-08-29 04:57:31.863870+00:00",
"db_md5": "f196f9e3e3d6de9df94e6a10679ea928",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "1d69df2a-9e41-4ef9-a9e8-2155498d59f4",
"tier": "easy",
"dataset_id": "e5",
"status": "succee... | {
"score": {
"mean_loss": 2.760002215685269,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007500000024835269
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.404733785064766,
"example_count": 600,... |
255151c4-e87f-470e-9ac7-fd9dc01ef0e2 | easy | khushidahi | 2026-08-16 13:52:50.391935+00:00 | succeeded | d3a555d345da431e37f5bfc7797055d5b61caaa6192216d2813fa1f1c19f72a6 | 19,193 | null | """R114: early-stopped recurrent tape Transformer microprogram.
This submission tests the strongest rules-safe generic baseline suggested by
the public competition discussion. It does not implement squaring, modular
reduction, or any other task solver. Numeric prompt fields are only rearranged
onto equal-length, rig... | {
"id": "255151c4-e87f-470e-9ac7-fd9dc01ef0e2",
"created_at": "2026-08-16 13:52:50.391935+00:00",
"db_md5": "f94e97278bd1d1a2dc6322c530434410",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "89237894-a1bd-4422-a258-5e2f470b0376",
"tier": "easy",
"dataset_id": "e6",
"status": "succ... | {
"score": {
"mean_loss": 0.947985976934433,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.8297297656536102
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.4562602639198303,
"example_count": 60,
... |
25560974-50ce-456c-b598-f63496567aeb | easy | sankalp1999 | 2026-08-04 11:45:56.230116+00:00 | succeeded | 18d6496895aac20aa8c5d22fb9e482923d3ecde6a365705eff2ad5cb39e9783d | 7,560 | null | """Latent recurrence: encode once, apply a shared squaring map T times, decode.
v4 put the loop inside a transformer and underfit badly (train acc 4% after
4,432 steps) because attention cost ~7ms/step at 13 tokens and bought nothing.
v2 fit the train set perfectly with a cheap flattened encoder but memorized,
because... | {
"id": "25560974-50ce-456c-b598-f63496567aeb",
"created_at": "2026-08-04 11:45:56.230116+00:00",
"db_md5": "5181619f8e84bbb29ee3be83075887a3",
"submitter": "Sankalp ",
"github_login": "sankalp1999",
"run_id": "055c67b4-d207-4fdb-98a3-2fca4b92f9ae",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 2.74417507648468,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.019999999552965164
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8046486377716064,
"example_count": 100,... |
25583e30-4b56-478e-af95-66c843d24d45 | easy | Alan-Qin | 2026-08-18 17:08:00.120672+00:00 | succeeded | 0bc5d50fb167839ac0b3414e658dfef674352a761e2d3917c11b42c088a3bbec | 16,654 | null | """One-hot CRT loop + sibling-consistency supervision (v24).
Architecture: per-band one-hot residues (parameter-free encoding of x),
a learned per-band transition matrix W_m applied T times (STE one-hot), and a
shared digit decoder read at every loop step.
Training loss (token_training_loss): besides the final-answer... | {
"id": "25583e30-4b56-478e-af95-66c843d24d45",
"created_at": "2026-08-18 17:08:00.120672+00:00",
"db_md5": "cf3600bf05b023d27d45e55095750f32",
"submitter": "Zeyu Qin",
"github_login": "Alan-Qin",
"run_id": "b127a520-4ff0-4a0a-bdde-bc59aa59eeaf",
"tier": "easy",
"dataset_id": "e1",
"status": "succeede... | {
"score": {
"mean_loss": 0.0006766193400835618,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.00021798917441628873,
"example_count": 100,
... |
255c833e-6122-4282-be7b-e14e0d3f0e17 | easy | DESU-CLUB | 2026-08-24 11:15:54.262339+00:00 | succeeded | e283adac7ec19b68161a7b6ac4720adf8a3365f81fddf88a84f112d5544e75c9 | 10,172 | null | """Huginn-style looped Transformer whose iteration count is read from the T token.
prelude -> [core applied T times, input re-injected each step] -> coda
Forked from experiments/_baselines/adamw_transformer.py; the Block, RMSNorm and
optimizer are unchanged so that the only varied axis is the depth mechanism.
"""
fr... | {
"id": "255c833e-6122-4282-be7b-e14e0d3f0e17",
"created_at": "2026-08-24 11:15:54.262339+00:00",
"db_md5": "2d5322efb324658d3fce53a2b0aacecb",
"submitter": "Low Keng Hoong, Warren",
"github_login": "DESU-CLUB",
"run_id": "d48ce1fe-4a2b-49bb-ae7b-1086cdcd37d8",
"tier": "easy",
"dataset_id": "e5",
"sta... | {
"score": {
"mean_loss": 2.1683377078045996,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0033333333333333335
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.176232150584593,
"example_count": 60... |
25667590-a1b4-44f3-ae1c-447918f03684 | easy | sapient-sapiens | 2026-08-20 15:06:07.210759+00:00 | succeeded | 93a480bbd7eda2e4eadf1bb9593c797278cc126efcb0e92008a510c6c136bdfe | 38,813 | null | """Closed learned spectral interface around a categorical recurrent operator.
The recurrent state is a bank of right-aligned categorical digit slots. Each
application uses a shared, shallow N-to-X conditioner to create learned phase
coordinates, injects paired sin/cos features as a small residual, applies the
ordinar... | {
"id": "25667590-a1b4-44f3-ae1c-447918f03684",
"created_at": "2026-08-20 15:06:07.210759+00:00",
"db_md5": "b32be13e1ec84a41fd22e6062af56df6",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "5b7f1b9b-a5c3-49e4-be7b-b86e8195530a",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 2.1391920830035724,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004583333333333333
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.140014480582267,
"example_count": 600... |
256b9aae-2247-4c72-96e8-76133a072560 | easy | shirvani-jr | 2026-08-21 22:56:26.718581+00:00 | succeeded | 764ae70f44b67ef44bb813db862b265c4f8c0f8dc408b56dae92f01d6f0b28a4 | 16,419 | null | """Looped digit-state squaring machine.
One recurrent application of a shared block corresponds to one modular
squaring. The recurrent state is the decimal digit string of the current
value, re-quantised every step with a straight-through estimator, so the
learned map composes exactly and depth extrapolates in T.
"""... | {
"id": "256b9aae-2247-4c72-96e8-76133a072560",
"created_at": "2026-08-21 22:56:26.718581+00:00",
"db_md5": "ad7bde53c35a9c9939f5dedc8842a1c8",
"submitter": "Ali",
"github_login": "shirvani-jr",
"run_id": "33c5662c-f2cc-441a-9c06-0b6cbdfa4b4d",
"tier": "easy",
"dataset_id": "e2",
"status": "succeeded"... | {
"score": {
"mean_loss": 1.5177530646324158,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.6408333778381348
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.3002516031265259,
"example_count": 300,... |
256bb7aa-bbca-4c9f-8f0e-999105e975ac | easy | yashkant | 2026-08-09 01:27:04.109310+00:00 | succeeded | 89874e575e4ac80f9a89b1722969ae2412b6b1d26e5de54e8d315f13fa8c992f | 17,720 | null | """Round 143 with PyTorch add-zero attention sinks and raw EMA."""
from __future__ import annotations
import math
import torch
import torch.nn.functional as F
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_model_state,
)
from torch import... | {
"id": "256bb7aa-bbca-4c9f-8f0e-999105e975ac",
"created_at": "2026-08-09 01:27:04.109310+00:00",
"db_md5": "f537c7575db2a1b9242d6051d756f210",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "582767c8-f3e6-480c-b1c4-884073433dee",
"tier": "easy",
"dataset_id": "e2",
"status": "succeed... | {
"score": {
"mean_loss": 4.389189958572388,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.24895834363996983
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.002066135406494,
"example_count": 300,
... |
2571f646-c2a0-4611-a96e-a1879c1ff2f4 | easy | Ranoobaba | 2026-08-19 17:05:16.427524+00:00 | succeeded | 87b43d755db7cb730fff81384618635653c3d3ef9e330419b8789a5d45690e36 | 6,601 | null | """Looped transformer, iteration count conditioned on T read from the prompt.
Core idea: learn one latent "squaring step" operator and apply it T times.
Depth generalization comes from the loop structure itself; the model never
needs more parameters for deeper problems, only more iterations.
"""
from __future__ impor... | {
"id": "2571f646-c2a0-4611-a96e-a1879c1ff2f4",
"created_at": "2026-08-19 17:05:16.427524+00:00",
"db_md5": "eb0bbb59a95b91481d94e3a64f0e7ada",
"submitter": "Syed Rayyan Ali",
"github_login": "Ranoobaba",
"run_id": "3f08fdd3-bfff-4f1a-96f2-a74de364b822",
"tier": "easy",
"dataset_id": "e5",
"status": "... | {
"score": {
"mean_loss": 4.913298917785079,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00625
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.9992397419540335,
"example_count": 600,
"... |
257b0d83-9510-4a26-8e30-7404bbe23d24 | easy | Dandandan | 2026-08-07 03:02:59.485991+00:00 | succeeded | c3a1122899719ba5f4cc30d4ddf61c0c906ba51e18f5d0a2b285ef2ac79ff756 | 16,809 | null | """Generic multi-register recursive function approximator.
This ablation intentionally exposes only learned routing, addition,
multiplication, constants, modular wrapping, and recurrent state. It contains
no quotient, reciprocal, exponentiation, predicate-modulus, polynomial-degree,
or recurrence-family operator. Ha... | {
"id": "257b0d83-9510-4a26-8e30-7404bbe23d24",
"created_at": "2026-08-07 03:02:59.485991+00:00",
"db_md5": "e44341b311d26e12a74b5d0ceb55b35f",
"submitter": "Daniël Heres",
"github_login": "Dandandan",
"run_id": "516dee13-c4d9-4c8b-9f03-39aec5be6a3e",
"tier": "easy",
"dataset_id": "e1",
"status": "suc... | {
"score": {
"mean_loss": 1.1271511084487429e-05,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.1280336366326082e-05,
"example_count": 100,
... |
257e5e30-2d67-4ebc-a41e-d0594dd7339d | easy | jordanrubin | 2026-08-12 04:47:59.686728+00:00 | succeeded | a8113fa5fd30cd348b465897ea29b52cff29e3c0818fa036daff3b32834a3bbe | 22,495 | null | """Weight-tied MLP-cell loop over soft digit states (mlploop).
Cell = MLP over [expected digits, pairwise digit products, RNS residue
simplices (fixed differentiable mixing over Z_p), N digits]. The cell that
learns one-step modular squaring from direct pairs (52% unseen at 14k rows,
day-20 screen) inside the exact-T ... | {
"id": "257e5e30-2d67-4ebc-a41e-d0594dd7339d",
"created_at": "2026-08-12 04:47:59.686728+00:00",
"db_md5": "73f7cb6eef76bd925bbd37d57d0aa13b",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "433c10fc-e4ab-4058-b74a-a5a487b928a3",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.5633233813227694,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009999999776482582
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1584627593579313,
"example_count": 60... |
257ea183-1767-48a1-a331-e46f357a9f15 | easy | gauravmishra | 2026-08-09 04:38:25.699517+00:00 | succeeded | d09e01a3b95abc9b9a2694d031afd8021c78f7cf156bb058ee95a82bac7258ac | 5,040 | null | """Basic single-pass Transformer with PyTorch AdamW."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_model_state,
)
D_MODEL = ... | {
"id": "257ea183-1767-48a1-a331-e46f357a9f15",
"created_at": "2026-08-09 04:38:25.699517+00:00",
"db_md5": "4e516848521bd26fe9e9438bf45f83c7",
"submitter": "Gaurav Mishra",
"github_login": "gauravmishra",
"run_id": "7fa973c4-b3b5-40ef-8609-5bae531e406b",
"tier": "easy",
"dataset_id": "e5",
"status": ... | {
"score": {
"mean_loss": 2.3054628458990907,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005416666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.3327892329126194,
"example_count": 60... |
257f070b-e0ad-4000-a802-4ec09a97a851 | easy | 0Chris5R | 2026-08-07 12:43:08.825971+00:00 | succeeded | b56e48acf99700b3a905a5445e59646a04b4d7dd4d3a6466fa7bc53563c4bfb5 | 20,838 | null | """Shared-digit Neural GPU with a compute-matched operator curriculum.
The model aligns decimal digits by significance, forms a learned low-rank
second-order feature map of the current state, and evolves a two-dimensional
workspace with the original convolutional-GRU equations. Each arithmetic
transition applies the ... | {
"id": "257f070b-e0ad-4000-a802-4ec09a97a851",
"created_at": "2026-08-07 12:43:08.825971+00:00",
"db_md5": "dd814a8b8afd0cc18ef330a4fe45f159",
"submitter": "Chris ",
"github_login": "0Chris5R",
"run_id": "4cf61111-3cf3-40be-9188-65124b8b282f",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 3.0687748305330995,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.046666666691501935
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.030519492957624,
"example_count": 600... |
257f51a0-5eab-41d3-a107-4b29cfc81d46 | easy | lpbb | 2026-08-12 08:14:41.196740+00:00 | succeeded | f60f6021bb5e247f983a9a1b971ff72df71e09e97602529ffee5b95b1e6f125d | 24,142 | null | """DIGIT-STATE DISCRETE FLOW (v2) — diffusion-style quantized recurrence, e1.
v1 (submissions/agents/diffusion) put the recurrent state in a learned
codebook. Flaw found on paper: nothing ties step(enc(x)) to enc(f(x)), so
the code space can fragment into per-chain private paths and the coverage
argument (98-99% of e... | {
"id": "257f51a0-5eab-41d3-a107-4b29cfc81d46",
"created_at": "2026-08-12 08:14:41.196740+00:00",
"db_md5": "ffc44ff5cc3743e77299143eaf88f930",
"submitter": "Lpbb",
"github_login": "lpbb",
"run_id": "7fb07731-2513-444c-afb1-83b5b57361dc",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
"s... | {
"score": {
"mean_loss": 0.798491170629859,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.8700000047683716
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.0544266514480114,
"example_count": 100,
... |
2583e1ff-9a0f-4f87-a758-9a524cf14784 | easy | velocizapkar | 2026-08-06 22:20:12.208438+00:00 | succeeded | 597ba3efbab0d3ee52027d0569dcbbee9423319cd61293ed4cea2b493d26dbfd | 55,676 | null | """Fast packed residual-free Attention-Residual transformer with Muon.
The model is a generic learned sequence model. It contains no parser,
arithmetic primitive, task constant, generated data, label feedback, or answer
construction. Every macro-state is produced by a serial pack of attention and
MLP transformations... | {
"id": "2583e1ff-9a0f-4f87-a758-9a524cf14784",
"created_at": "2026-08-06 22:20:12.208438+00:00",
"db_md5": "7c1a4f4a4b38284b233a11a894731e05",
"submitter": "Aakanksh Zarapkar",
"github_login": "velocizapkar",
"run_id": "a91720c0-8cba-42ad-87c1-b17a026c504e",
"tier": "easy",
"dataset_id": "e3",
"statu... | {
"score": {
"mean_loss": 3.4313353300094604,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.019375000149011612
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.4452497959136963,
"example_count": 80... |
2594b800-9436-4e60-986c-d4ae04a80c47 | easy | KaustubhKumar05 | 2026-08-25 16:27:39.090030+00:00 | succeeded | b6d72aabc7da752cbf89d97df87ad90c9e9600f805ec3541ee7c84fab6ab2bb3 | 17,454 | null | """conv_sat: final_v1 + the DNGPU saturation cost (Freivalds & Liepins 2017).
Our hard gate clamp(1.2*sigmoid-0.1,0,1) has EXACTLY ZERO gradient outside
|x|>=2.398. The paper that introduced this gate says an explicit saturation
penalty is mandatory: without it units leave the linear band early and can
never return ("... | {
"id": "2594b800-9436-4e60-986c-d4ae04a80c47",
"created_at": "2026-08-25 16:27:39.090030+00:00",
"db_md5": "5c8b9f4ebca3a2775b1bf2c2ddbd787e",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "79ba5b5c-9a59-4f26-a8d5-fd44452d00f8",
"tier": "easy",
"dataset_id": "e6",
"status": "succee... | {
"score": {
"mean_loss": 5.021521329879761,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.23333334177732468
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.708783149719238,
"example_count": 60,
... |
25960a0b-dbac-431e-9062-97c355969329 | easy | liam-gb | 2026-08-13 00:40:42.281593+00:00 | succeeded | eb2504bb7d0157e632cf799a1f18f53674416b6d2c2037173c708bf726abbc5b | 14,358 | null | """rns_pm_gather_ct_widet: dataset-sized digit head, dead loop tail trimmed.
The wide readout is what makes long answers reachable; the trim is what pays
for it, by not running loop iterations whose update is multiplied by zero.
Base docstring:
Single change vs rns_pm_gather_ct: the readout carries one
distance-from... | {
"id": "25960a0b-dbac-431e-9062-97c355969329",
"created_at": "2026-08-13 00:40:42.281593+00:00",
"db_md5": "1a548bea555b987cf0a175795dc3aa83",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "1cc71da1-3689-4f76-847c-87f07218f73a",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 5.428326964378357,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.4866666868329048
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.575155258178711,
"example_count": 100,
... |
259b2b67-214c-402e-9d42-aeca2cbb9a0c | easy | 0Chris5R | 2026-08-02 15:49:36.924744+00:00 | succeeded | dcec59a6b4a1fc3e69b481f7a2a7aac0d7a686d6712ccccda244544ccc4919e1 | 28,964 | null | """Shared-digit Neural GPU with a K=4/order-2 completion-volume loss.
The model aligns decimal digits by significance, forms a learned low-rank
second-order feature map of the current state, and evolves a two-dimensional
workspace with the original convolutional-GRU equations. Each arithmetic
transition applies the s... | {
"id": "259b2b67-214c-402e-9d42-aeca2cbb9a0c",
"created_at": "2026-08-02 15:49:36.924744+00:00",
"db_md5": "39c3d4535ee6e8ff743fd63fda643748",
"submitter": "Chris ",
"github_login": "0Chris5R",
"run_id": "3ea382be-9fba-4c56-b93b-e5d278e45e83",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 3.812924874346362,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.029166666679084303
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.822566351013868,
"example_count": 600,... |
259ec891-4234-4213-978d-41041e7c8231 | easy | DDanlov | 2026-08-17 18:50:00.412747+00:00 | succeeded | 2225b518be259e3cfad3493b6fe7cc572eac0c5b4575289cdfcc529f01f967a0 | 22,560 | null | import math
from typing import Optional
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Optimizer
try:
from benchmark import Submission, assert_model_state, OptimizerBundle
except ImportError:
try:
from client import Submission, assert_model_state, OptimizerBu... | {
"id": "259ec891-4234-4213-978d-41041e7c8231",
"created_at": "2026-08-17 18:50:00.412747+00:00",
"db_md5": "6746fb28fb85173c861698637a3bd24a",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "329ca4fe-f11b-4640-bac0-f3137738a113",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.20184915841999,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.033333333730697634
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2383689853682447,
"example_count": 100,... |
25a07003-9419-40db-ae6d-adfc06ce0b6d | easy | sapient-sapiens | 2026-08-22 17:19:26.231613+00:00 | succeeded | 725b3ff0b6c04f2c91503008cc80ef7346788e4e7db4ffaff7743c8738ea601b | 15,610 | null | """Six-layer T-private Q-gated TRM with Euclidean SAM for E5.
Research hypothesis: the prior SAM arm was under-stepped by its inherited
AdamW learning rate, and doubling that rate will improve its fixed-clock fit.
Primary experimental variable: SAM base learning rate, 1e-4 versus 2e-4. The
arithmetic representation i... | {
"id": "25a07003-9419-40db-ae6d-adfc06ce0b6d",
"created_at": "2026-08-22 17:19:26.231613+00:00",
"db_md5": "426b74aa73a2ea42d821918881dea33b",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "dd00ad70-efe5-4838-9d82-c671882799af",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 2.321318007072528,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0004166666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.319008019606629,
"example_count": 600... |
25a669a0-0ee9-46fe-8564-7c74d7644e52 | easy | isaac0804 | 2026-08-05 08:57:04.195475+00:00 | succeeded | f836b272089910a4ba0c4460e0669db3ddc6c9b9b00a877b35732b896ca1e908 | 5,186 | null | """Vanilla (non-recurrent, non-weight-tied) deep Transformer -- a control to
isolate "is a single attention+MLP pass too shallow to even use T" from the
looped architecture's own known failure modes.
`baseline_adamw` (this project's original Step-0 baseline) has exactly one
block: one attention+MLP pass has to read N,... | {
"id": "25a669a0-0ee9-46fe-8564-7c74d7644e52",
"created_at": "2026-08-05 08:57:04.195475+00:00",
"db_md5": "6a4a79008638fac84c6972fac862e86e",
"submitter": "Isaac Yong",
"github_login": "isaac0804",
"run_id": "07899ccd-19eb-4e2e-811e-d64042f9b788",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 12.473203667639224,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00812500001396984
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 12.74724260337422,
"example_count": 800,... |
25a9130d-d419-413a-b55c-0fe754bd21ca | easy | gauravmishra | 2026-08-10 03:34:18.936144+00:00 | succeeded | 019c542d8ca733b50f5b519afae199f339e38f20f8436ce9d4c471018f1b7780 | 32,491 | null | """Information-gain candidate ig60_e18_recall_simple_e2 (independent)."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_model_st... | {
"id": "25a9130d-d419-413a-b55c-0fe754bd21ca",
"created_at": "2026-08-10 03:34:18.936144+00:00",
"db_md5": "51b9f92fa22d35e4ba70e4c07d07176d",
"submitter": "Gaurav Mishra",
"github_login": "gauravmishra",
"run_id": "d85f64de-a1d3-43e9-8eb8-1157885a1faf",
"tier": "easy",
"dataset_id": "e2",
"status": ... | {
"score": {
"mean_loss": 2.321972608566284,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006458333693444729
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.3506298065185547,
"example_count": 300... |
25a92e05-8b43-4d19-a08c-c24e5396e2d5 | easy | alirezashirvani-jr | 2026-08-31 07:32:00.675136+00:00 | succeeded | c252ce15d535434ae9695254770a32b96e97a7120717f3216a16792d4ff23118 | 34,903 | null | from __future__ import annotations
import math
import time
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_model_state,
)
TRACK_FLOOR = 3
TRACK_CEILING = 96
STAGE_B... | {
"id": "25a92e05-8b43-4d19-a08c-c24e5396e2d5",
"created_at": "2026-08-31 07:32:00.675136+00:00",
"db_md5": "b9699e2f9827d45ad6df7d21eed39459",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "4e793e29-373b-4281-935e-94ec9ac26e85",
"tier": "easy",
"dataset_id": "e7",
... | {
"score": {
"mean_loss": 0.0702258083038032,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.9848484992980957
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.014089888893067837,
"example_count": 85... |
25aa1506-93b8-4f5c-a3e4-35622041e6fa | easy | adwhit | 2026-08-13 13:37:19.500613+00:00 | succeeded | 75f37081fa9e8fc2728e3798bd0e44beec1e8bd3026c39b4e449b69e68a14043 | 18,111 | null | """DeepThinker v4a: depth-recurrent transformer with manual CUDA-graph capture.
The evaluation harness is kernel-launch-bound on unrolled-loop models
(measured on H100/Easy: eager 206 steps/60s with the GPU >95% idle;
torch.compile default -> 120 steps net; reduce-overhead -> crash). v4a
captures the ENTIRE training u... | {
"id": "25aa1506-93b8-4f5c-a3e4-35622041e6fa",
"created_at": "2026-08-13 13:37:19.500613+00:00",
"db_md5": "b3906f44f2ecd2788798748f0d822113",
"submitter": "adwhit",
"github_login": "adwhit",
"run_id": "43f52eec-56c6-43d7-a7fa-e2ceeccfd86a",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
... | {
"score": {
"mean_loss": 2.600987434387207,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04999999888241291
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.9253005981445312,
"example_count": 100,... |
25ad13e4-4e8e-41b6-b03b-5fd064bd3472 | easy | lpbb | 2026-08-19 12:54:32.803490+00:00 | succeeded | ca9859e0fd7e33690468da00a8d5a03d3eb4f4407a3db8ed0caa91481b316970 | 9,874 | null | """S0 + RoPE + abacus embeddings + dropout, with a looped block: instead of
`num_layers` independent layers, a block of `num_layers` unique layers is run
`num_repeats` times with shared weights (e.g. 12 layers x 4 repeats = depth 48
but only 12 layers' worth of parameters). Weight tying across repeats forces
the block ... | {
"id": "25ad13e4-4e8e-41b6-b03b-5fd064bd3472",
"created_at": "2026-08-19 12:54:32.803490+00:00",
"db_md5": "366fb65d7752995e2ed5ac72c60f1ffd",
"submitter": "Lpbb",
"github_login": "lpbb",
"run_id": "8b96e41f-78cb-4782-b5a5-823bd7122790",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
"s... | {
"score": {
"mean_loss": 2.9241825342178345,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03666666615754366
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.428598403930664,
"example_count": 100,... |
25afdbbf-4bf2-490c-89be-94d9d1a27cd8 | easy | chad-atexpedient | 2026-08-16 05:19:41.817347+00:00 | succeeded | 235898c0e1734aa3e35fcb6c893dbf5b411db0f6ec9abdfea9742a799698ff77 | 21,211 | null | """X55: Hybrid continuation-label + state consistency.
Combines X54's continuation-label supervision with X18's state consistency loss.
- Continuation: supervise A at step a+b with B's label z when A.output == B.input.
- Consistency: match T=2 intermediate state (step 1) with T=1 final state when
T=2.output == T=1.i... | {
"id": "25afdbbf-4bf2-490c-89be-94d9d1a27cd8",
"created_at": "2026-08-16 05:19:41.817347+00:00",
"db_md5": "8abd08b8dc4c08710c481279173263da",
"submitter": "chad-atexpedient",
"github_login": "chad-atexpedient",
"run_id": "e73aa541-19a1-4550-a02f-8506f66b795a",
"tier": "easy",
"dataset_id": "e5",
"st... | {
"score": {
"mean_loss": 2.178722713567641,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00625
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.18723600113873,
"example_count": 600,
"ex... |
25b7c9ba-13a3-4ee4-8f9f-c9ebbdb107e0 | easy | ehonig | 2026-08-20 07:57:19.591778+00:00 | succeeded | 4eef5ca0d4fe18566f4e18402a9accfb4c145c297e4817d393fde57ded74009b | 23,039 | null | """A plain Transformer for repeated modular squaring, and nothing else yet.
This is a deliberate restart. The previous submission accumulated a field
parser, a carry scan, a learned reciprocal, periodic features, a digit
bottleneck, a depth selector and two cell types, most of them tuned around a
memorisation table th... | {
"id": "25b7c9ba-13a3-4ee4-8f9f-c9ebbdb107e0",
"created_at": "2026-08-20 07:57:19.591778+00:00",
"db_md5": "2ebc5426f353202c3ee7ab7d769ae51e",
"submitter": "Edouardo Honig",
"github_login": "ehonig",
"run_id": "9483c1f7-2274-46ba-88c5-71f43cffdabf",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.227218907024409,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005833333289871613
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2347073554992676,
"example_count": 600... |
25bd3757-e344-4fca-8827-160ae56cdd07 | easy | khushidahi | 2026-08-07 18:43:08.322762+00:00 | succeeded | a4ef781dd18d6a58a8bf72dcc670a225294b27c29aa8786132f94af32dc7a31c | 7,289 | null | """H0 conservative R3 Hard anchor.
This is the existing R3 architecture and AdamW recipe evaluated after exactly
3,000 optimizer steps. It provides the matched control for the boundary-aware
sequence-loss experiment.
"""
from __future__ import annotations
import random
import torch
import torch.nn.functional as F
f... | {
"id": "25bd3757-e344-4fca-8827-160ae56cdd07",
"created_at": "2026-08-07 18:43:08.322762+00:00",
"db_md5": "a10ceaecb7ccf6e166118bd257b320b1",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "0584ee05-316f-45c3-8076-c280bca111a7",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.1639130115509033,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004166666767559946
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1697494983673096,
"example_count": 60... |
25c5108c-7483-4f7a-a3ae-056ad935cbdd | easy | khushidahi | 2026-08-03 02:07:39.916769+00:00 | succeeded | 28b16cb0c7b56e63c7942bcf5145c0f85aff11781703d1bb802936b1b528d732 | 14,297 | null | """R6 structured decimal-workspace model for One Layer Deeper.
The model does not implement multiplication, modular reduction, or the public
recurrence. It uses the public tokenizer structure to build learned, right-
aligned decimal tapes for N, X, and T, then applies a shared neural transition
to a mutable work tape.... | {
"id": "25c5108c-7483-4f7a-a3ae-056ad935cbdd",
"created_at": "2026-08-03 02:07:39.916769+00:00",
"db_md5": "6ff64586a615901898da38fff73a4330",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "07fbab29-ee3b-449d-837b-58e80438e37e",
"tier": "easy",
"dataset_id": "e4",
"status": "succ... | {
"score": {
"mean_loss": 2.080424666404724,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006354166893288493
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.078592300415039,
"example_count": 1200... |
25c96c48-963b-48c9-b747-780e7fb86834 | easy | DDanlov | 2026-08-24 20:37:39.711885+00:00 | succeeded | a40a8ee679a55e12beb19d8180c9ce6e769c4b4c0b860c5117652127b39d79bf | 49,569 | null | """
Unified Pure Skipless Prefix Deliberation Model + Transformer Readout Decoder
Architecture:
- Deliberation Core: dim=1536, num_heads=24, d_ff=6144 (Pure Skipless SO(2)^{D/2} Givens Recurrence)
- Token Layout: Recurrent loop processes ONLY [Prompt (L_p), 4 Pause Registers] (Zero target tokens in loop)
- Recurrent At... | {
"id": "25c96c48-963b-48c9-b747-780e7fb86834",
"created_at": "2026-08-24 20:37:39.711885+00:00",
"db_md5": "5e17bc6fffb275150145dc46e08d4e1a",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "647df5bb-c151-4177-b138-f009407d3ead",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.638452401327597,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004583333333333333
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.670272785451915,
"example_count": 600,... |
25de1294-0719-4177-b7dc-7e31d55890ed | easy | erdavis0 | 2026-08-27 03:10:32.287553+00:00 | succeeded | dea5a50d1a7e1542b606c505bb2e064069826b4fcc486a84827b8cbefc9b781f | 31,293 | null | """Typed-prefix reversible transition B0: ordinary evaluator token CE.
Public delimiter syntax routes opaque N, X, and T fields into three strict
functional roles. X initializes only an evolving auxiliary tape, N initializes
only an invariant environment lane, and T reaches only a learned monotone
prefix controller. ... | {
"id": "25de1294-0719-4177-b7dc-7e31d55890ed",
"created_at": "2026-08-27 03:10:32.287553+00:00",
"db_md5": "9a82ec9520ce6d08ad83d5dbf700b0ab",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "c16db2db-4a80-48ef-9b7e-4ca4ce688530",
"tier": "easy",
"dataset_id": "e4",
"status": "succeeded",... | {
"score": {
"mean_loss": 2.7195610096343907,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004479166666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.742370671636736,
"example_count": 120... |
25e09cd7-f667-4f62-802b-1c9d46dbd284 | easy | DDanlov | 2026-08-26 18:48:12.185858+00:00 | failed | 18ac7a3b409898e9b2ca8373f1651582da15659f169f453fcbfe350e34b0643d | 29,673 | null | import math
import time
from typing import Optional, Tuple, Dict, Any
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Optimizer
try:
from onelayerdeeper.api import Submission, OptimizerBundle, ModelSpec, OptimizerSpec
except ImportError:
Submission, OptimizerBundle, M... | {
"id": "25e09cd7-f667-4f62-802b-1c9d46dbd284",
"created_at": "2026-08-26 18:48:12.185858+00:00",
"db_md5": "dab43f52a1995506434099f5319bd327",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "64941428-893a-4e5a-9570-d71b8e2d4fe9",
"tier": "easy",
"dataset_id": "e5",
"status": "failed",
... | null
|
25e91b5b-26f8-4ff3-99f1-9b98d3420aa6 | easy | khushidahi | 2026-08-09 02:20:05.234194+00:00 | succeeded | 3e1b0cbbfe4059ba846c573544ebb91224b8e275f31b7a6ff81988eeb0e200a7 | 12,590 | null | """R45 full-budget, learned-symbol baseline.
This model intentionally avoids task-specific numerical decoding:
- no fixed digit values;
- no decimal-to-integer conversion;
- no FoNE parser;
- no solver-derived supervision.
The evaluator time allowance determines the LR schedule horizon. `max_steps`
is deliberately hi... | {
"id": "25e91b5b-26f8-4ff3-99f1-9b98d3420aa6",
"created_at": "2026-08-09 02:20:05.234194+00:00",
"db_md5": "8f4c17ff2d6af0977086c55147704b31",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "6ccaf420-d0c0-46a3-b7ee-55b62a5b7e7e",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.24021543411387,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007916666554907957
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.244190216064453,
"example_count": 600,
... |
25e92fe6-e91f-4e5f-9659-0d34730e2082 | easy | sapient-sapiens | 2026-08-15 20:04:32.188925+00:00 | succeeded | 71bb37405aec14435eb7e44aaeeed5920dc4f91171ba495c8f4d4e7ac817e505 | 17,772 | null | """T-only adaptive chunk routing over a learned shared operator.
The controller sees only the categorical tokens inside the T prompt field and
is trained solely by final-answer loss. It never parses T numerically and T is
not used as an intermediate supervision target. Inputs use learned additive
token, semantic-gro... | {
"id": "25e92fe6-e91f-4e5f-9659-0d34730e2082",
"created_at": "2026-08-15 20:04:32.188925+00:00",
"db_md5": "a9f745240e39ce561a0fc1af6f594f4a",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "b9c328f1-ae3c-4d8a-ba83-a79a18fd9b5b",
"tier": "easy",
"dataset_id": "e6",
"status": "su... | {
"score": {
"mean_loss": 5.863700001207116,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05022522696369403
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.845989227294922,
"example_count": 60,
... |
25ed457e-706b-42ce-8215-9c095bafbb07 | easy | jordanrubin | 2026-08-07 16:51:31.580766+00:00 | succeeded | 19e434c7e9fc9d9d601292a0f5c0a4083813bc7790caf87c0582fc1bbcc2cad8 | 16,909 | null | """Autonomous digit-state recurrent Transformer for repeated modular squaring.
The model is deliberately organized around one learned transition:
p_0 = right_aligned_decimal_digits(x)
p_{k+1} = F_theta(p_k, decimal_digits(N))
The same two-block Transformer cell is applied exactly T times. T is used only
as ... | {
"id": "25ed457e-706b-42ce-8215-9c095bafbb07",
"created_at": "2026-08-07 16:51:31.580766+00:00",
"db_md5": "671c15c2db39c821999bae96a7defa51",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "99bcc69b-ae1a-44cc-90b6-4950d3b792c2",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 4.225327445155221,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005833333202948173
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.343928600239646,
"example_count": 600,... |
25eedfd4-bb87-4f1d-950b-faaa7a4af1f6 | easy | erdavis0 | 2026-08-23 15:46:59.401258+00:00 | succeeded | d4e7b6f94c5a17f720147aa5b1ff3dedb9aca4341c22907ba39d97352562ee35 | 21,596 | null | """Learned confluent brickwork exchanges on field-aligned token tapes.
Opaque public token categories are routed into fixed right-aligned N, X, and T
tapes. Two learned-computation lanes then apply the same complete collection
of adjacent-place rewrites in opposite even/odd orders. One learned pair cell
is shared ov... | {
"id": "25eedfd4-bb87-4f1d-950b-faaa7a4af1f6",
"created_at": "2026-08-23 15:46:59.401258+00:00",
"db_md5": "e9eede9164a2380d5b1e85c703eae702",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "531ae1b4-06f5-4364-8dd4-26b708ce3801",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",... | {
"score": {
"mean_loss": 3.165613532066345,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.061666667461395264
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.138777494430542,
"example_count": 100,... |
25ef5212-caee-4ccf-a57e-d53280a3d8b1 | easy | shan23chen | 2026-08-09 02:47:25.396372+00:00 | succeeded | 709757cc0c57e37788fea37d4bafccbe1a16e80bb51a7cb966009faf11813f86 | 39,363 | null | """Dual-root boundary-reset recurrent register Transformer.
One learned transition is applied once per requested outer step. The
transition receives an immutable decimal N register and the current learned
boundary state, but never receives T or a separate copy of the original X.
Within each outer application, scratch... | {
"id": "25ef5212-caee-4ccf-a57e-d53280a3d8b1",
"created_at": "2026-08-09 02:47:25.396372+00:00",
"db_md5": "48b879776039b39e43b4984f2eb4966b",
"submitter": "Shan Chen",
"github_login": "shan23chen",
"run_id": "59b1d344-da6a-44ac-b52b-a6f623fb68af",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 2.2894487380981445,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.010624999646097422
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2913122177124023,
"example_count": 80... |
25f7945e-d81d-43a8-9566-d09cc0a59aaf | easy | ehonig | 2026-08-14 06:06:16.233255+00:00 | succeeded | 00eeaf71e1ef7eb8bca1dce42d5b5f824a664f5246edf93d90ecb8ba73c377e0 | 52,555 | null | """DigitLoop — a place-value recurrent model for repeated modular squaring.
The task is ``y = x^(2^T) mod N`` with the answer read back tail aligned, one
decimal place value per token position from the right. A stack of attention and
MLP layers has no primitive for the two operations this needs — a partial-product
co... | {
"id": "25f7945e-d81d-43a8-9566-d09cc0a59aaf",
"created_at": "2026-08-14 06:06:16.233255+00:00",
"db_md5": "5e27483be6be1e0b65bea8bf17fdfe9a",
"submitter": "Edouardo Honig",
"github_login": "ehonig",
"run_id": "80df2d6a-7f5f-4a01-927b-8278ade2ce16",
"tier": "easy",
"dataset_id": "e1",
"status": "succ... | {
"score": {
"mean_loss": 4.33024263381958,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.4166666716337204
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.5144248008728027,
"example_count": 100,
... |
25f7b967-eaee-455d-8d83-9dc3023b43e9 | easy | erdavis0 | 2026-08-25 01:42:31.799449+00:00 | succeeded | 441829d198f853ac2efa0065a0a953215d79f7cb90ee2b49280f23481a092acf | 24,085 | null | """Matched categorical-boundary panel: C1 continuous endpoint auxiliary.
Public delimiter syntax is used only to route opaque token identities onto
separate right-aligned tapes. One T-blind learned macro maps a categorical X
tape and a digitwise N tape to a new categorical tape through freshly reset
interaction, carr... | {
"id": "25f7b967-eaee-455d-8d83-9dc3023b43e9",
"created_at": "2026-08-25 01:42:31.799449+00:00",
"db_md5": "18bcf63268fc44cb8c6a3e621ae926dd",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "f1d42a27-47c2-41fd-9bec-8f287d28226f",
"tier": "easy",
"dataset_id": "e4",
"status": "succeeded",... | {
"score": {
"mean_loss": 2.068357243064695,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0010416666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.070186282902971,
"example_count": 120... |
25f9d51b-85dd-411b-a0e5-d612f38f19f2 | easy | alirezashirvani-jr | 2026-08-30 01:35:50.882510+00:00 | succeeded | 2585889de28a74e2ca67766e15d4a89a86ecd69e826394009281ea5d802060ba | 28,587 | null | from __future__ import annotations
import math
import time
from contextlib import nullcontext
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_model_state,
)
MODEL_... | {
"id": "25f9d51b-85dd-411b-a0e5-d612f38f19f2",
"created_at": "2026-08-30 01:35:50.882510+00:00",
"db_md5": "e9c8b63aabb572ec93a376f6e16463d6",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "a9514ea7-f6fd-4ec7-aaa0-ab587d0c2572",
"tier": "easy",
"dataset_id": "e2",
... | {
"score": {
"mean_loss": 1.094435304403305,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.7514583766460419
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.7608861327171326,
"example_count": 300,
... |
25f9dd3d-81b6-407b-8793-4445549fba9f | easy | jordanrubin | 2026-08-01 20:16:52.605256+00:00 | succeeded | 2c6adbc91db77091eaa1fdb53b2b15fdc6e05306608a2440be5e6a2aa2591b8a | 18,183 | null | """Looped bidirectional selective-SSM (Mamba-style), pure PyTorch.
The statefulness bet: input-dependent SSM dynamics along the token axis,
with the block weight-tied and applied NUM_LOOPS times in depth with input
injection — recurrence in depth is where the task's T-iteration lives.
The selective scan is computed i... | {
"id": "25f9dd3d-81b6-407b-8793-4445549fba9f",
"created_at": "2026-08-01 20:16:52.605256+00:00",
"db_md5": "35fec302dbad6ab1de27b5092ede0b6a",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "ccbf09a4-c50f-44cb-bdfe-873c9c4b00b6",
"tier": "easy",
"dataset_id": "e3",
"status": "s... | {
"score": {
"mean_loss": 5.207443784067686,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00937499979045242
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.595749526049944,
"example_count": 800,
... |
25fbd836-3c78-46c8-8624-114f01a09362 | easy | DESU-CLUB | 2026-08-24 10:03:13.608710+00:00 | succeeded | ce76301768085d521fd3d1e0236f790a91f21d61abe0cbe58443b6724f717679 | 7,898 | null | """Huginn-style looped Transformer whose iteration count is read from the T token.
prelude -> [core applied T times, input re-injected each step] -> coda
Forked from experiments/_baselines/adamw_transformer.py; the Block, RMSNorm and
optimizer are unchanged so that the only varied axis is the depth mechanism.
"""
fr... | {
"id": "25fbd836-3c78-46c8-8624-114f01a09362",
"created_at": "2026-08-24 10:03:13.608710+00:00",
"db_md5": "68bdc8a08618a83d90c208df60fc77f4",
"submitter": "Low Keng Hoong, Warren",
"github_login": "DESU-CLUB",
"run_id": "e8288a53-98ea-43ab-8e20-8cf731d7e44a",
"tier": "easy",
"dataset_id": "e6",
"sta... | {
"score": {
"mean_loss": 7.3780931087389385,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.11531531735247857
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.262131214141846,
"example_count": 60,
... |
26092f3e-1f56-4b39-9ef1-a2de00b08db2 | easy | DDanlov | 2026-08-31 21:15:30.693074+00:00 | succeeded | 0981212e7c32bbe813e123adfd4bacad4c23631e5eeadd8e39ce73e902ed95b5 | 32,408 | null | """
Unified Continuous Hopfield Equilibrium Propagation Deliberation Engine
=======================================================================
Architecture:
- Deliberation: Continuous Hopfield Multi-Layer Energy System (L=4, D=768, d_ffn=2304, H=16)
- Integration: In-Place Strang-Split Conformal Symplectic Velocit... | {
"id": "26092f3e-1f56-4b39-9ef1-a2de00b08db2",
"created_at": "2026-08-31 21:15:30.693074+00:00",
"db_md5": "474bd6441c27824955cd2c8a0feae135",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "237d6222-db9e-4195-838e-b3991771d7c6",
"tier": "easy",
"dataset_id": "e2",
"status": "succeeded"... | {
"score": {
"mean_loss": 3.089103634711382,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0027083333830038706
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.120418205216925,
"example_count": 300... |
26095e73-8311-47fa-907d-6932763504ee | easy | jordanrubin | 2026-08-01 21:16:22.779493+00:00 | succeeded | 2c6adbc91db77091eaa1fdb53b2b15fdc6e05306608a2440be5e6a2aa2591b8a | 18,183 | null | """Looped bidirectional selective-SSM (Mamba-style), pure PyTorch.
The statefulness bet: input-dependent SSM dynamics along the token axis,
with the block weight-tied and applied NUM_LOOPS times in depth with input
injection — recurrence in depth is where the task's T-iteration lives.
The selective scan is computed i... | {
"id": "26095e73-8311-47fa-907d-6932763504ee",
"created_at": "2026-08-01 21:16:22.779493+00:00",
"db_md5": "35fec302dbad6ab1de27b5092ede0b6a",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "853a7b3b-3f4c-4b6c-b312-c8fbbdfb26e9",
"tier": "easy",
"dataset_id": "e1",
"status": "s... | {
"score": {
"mean_loss": 5.125701133152493,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.02666666607062022
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.339169502258301,
"example_count": 100,
... |
260a1a6a-010e-4ff8-acb9-d1f48747e066 | easy | AdityaVG13 | 2026-08-15 08:50:28.819494+00:00 | succeeded | caaf98add9f2e30b56706591294b76bdb17e616649970a5b628d6bc3baafd99d | 109,356 | null | """AEGIS V9 v9_t1_exact: frontier JOIN plus weakest-token exactness on T=1 rows.
One Layer Deeper submission candidate. All task transitions remain learned
end-to-end. The deterministic shell only identifies marker-delimited roles,
constructs a finite semantic tape, schedules learned interval programs, and
applies a t... | {
"id": "260a1a6a-010e-4ff8-acb9-d1f48747e066",
"created_at": "2026-08-15 08:50:28.819494+00:00",
"db_md5": "d017a21fa9cda1bd5f3fc1635c65c83a",
"submitter": "AdityaG",
"github_login": "AdityaVG13",
"run_id": "5bef9d05-b80b-4557-b1aa-1a35e98b82af",
"tier": "easy",
"dataset_id": "e1",
"status": "succeed... | {
"score": {
"mean_loss": 2.1230718195438385,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.6583333313465118
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.7431753277778625,
"example_count": 100,... |
260a83ae-8568-4fb0-a220-0d13f77a134e | easy | sapient-sapiens | 2026-08-18 21:09:30.804118+00:00 | succeeded | 196dd12bc262da85d1fd015973bee06ecf669b625f14b002c9bfa9a8b76ac8dd | 31,889 | null | """Depth-quantized recurrence with a mixture-over-depth objective.
The recurrent state is a bank of right-aligned digit slots. Every operator
application is followed by a soft quantization back onto the token simplex, and
the digit logits produced by that quantization are the answer logits at that
depth. There is no s... | {
"id": "260a83ae-8568-4fb0-a220-0d13f77a134e",
"created_at": "2026-08-18 21:09:30.804118+00:00",
"db_md5": "f308475cc0e034316d78c8bcebecb653",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "654200ad-8de3-4c41-bd1e-60898ad90363",
"tier": "easy",
"dataset_id": "e9",
"status": "su... | {
"score": {
"mean_loss": 4.152336597442627,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.036666667088866234
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.176885604858398,
"example_count": 90,
... |
260d88ba-1b72-4bfb-992e-b5d461b8c702 | easy | DDanlov | 2026-08-19 20:21:16.494063+00:00 | succeeded | ae7021884f4490ad34ea8f60c30a76e0c497957ef9c306720bb73ee23d09f663 | 17,727 | null | """
Dynamic Submission for One Layer Deeper API Benchmark
Config: dim=1696, num_heads=32, d_ff_mult=4, fixed_trec=224
"""
from __future__ import annotations
import math
from typing import Any, Optional, Tuple
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Optimizer
try:
... | {
"id": "260d88ba-1b72-4bfb-992e-b5d461b8c702",
"created_at": "2026-08-19 20:21:16.494063+00:00",
"db_md5": "7c2c9fa3ce3638513e52924a6730fb8c",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "567d0a8b-cf6d-4a5e-9293-3b16448bcb46",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.4119404265944144,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.011666666679084302
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.4911348878535455,
"example_count": 60... |
260dc755-22d2-41e9-a478-b100c4e496d0 | easy | gauravmishra | 2026-08-10 03:48:12.613345+00:00 | succeeded | cdcb98e5b24d642dd9f44358b223efc3c16031704c15cb0be9d638a88a72338d | 12,548 | null | """Stage-1 information-gain K4 representation candidate."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_model_state,
)
POSITI... | {
"id": "260dc755-22d2-41e9-a478-b100c4e496d0",
"created_at": "2026-08-10 03:48:12.613345+00:00",
"db_md5": "4107fd87353ba756106c23c6b0e5975f",
"submitter": "Gaurav Mishra",
"github_login": "gauravmishra",
"run_id": "e6e4c6c3-6dd5-4c2a-a690-34ad1d570314",
"tier": "easy",
"dataset_id": "e3",
"status": ... | {
"score": {
"mean_loss": 2.539411498535009,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004375000004656613
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.6602859497070312,
"example_count": 800... |
26265643-7d98-4e87-9dc1-dd2feb658cd1 | easy | chuk-uzowihe | 2026-08-12 18:53:05.092705+00:00 | succeeded | 1e7aaedf1c7fa9c7f97f8b2be690ccecaa0a54ee42f8938a5c59e4a0bef93d9b | 26,029 | null | """v10: dense fixed-loop recurrence with deep supervision.
Lineage: FPRM structure (alpha/beta-tied residual scaling, pre-rms_norm,
input injection every iteration) + the v8 2D digit lattice / RoPE task
encoding. Three deliberate departures from the v9 MoEUT line:
1. **Full BPTT.** The loop runs a fixed N_ITER iterat... | {
"id": "26265643-7d98-4e87-9dc1-dd2feb658cd1",
"created_at": "2026-08-12 18:53:05.092705+00:00",
"db_md5": "3deefbe3de9964427134af4a3fefcf22",
"submitter": "Chuk Uzowihe",
"github_login": "chuk-uzowihe",
"run_id": "e256c68a-9a11-4d2d-bc62-680d7a8968af",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 10.683088302612305,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05833333171904087
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.850722312927246,
"example_count": 100,... |
2628619f-cfaf-402a-85e2-9386602fe340 | easy | benjaminW2025 | 2026-08-19 05:52:52.092408+00:00 | succeeded | 704a80907cc78cb152b41dcf3f398f549f2d45d4eab45fe112a378211d54e96d | 4,782 | null | """Phase 0 R4x2: four full-token blocks shared across two repeats."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
)
D_MODEL = 256
N... | {
"id": "2628619f-cfaf-402a-85e2-9386602fe340",
"created_at": "2026-08-19 05:52:52.092408+00:00",
"db_md5": "139c37696e6f540fbf2840863fdd02df",
"submitter": "benjawesome",
"github_login": "benjaminW2025",
"run_id": "6af45014-a4f7-48a1-a2f3-1164e24101bc",
"tier": "easy",
"dataset_id": "e3",
"status": "... | {
"score": {
"mean_loss": 2.4295702166902986,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0037500000046566126
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.723390179418645,
"example_count": 80... |
26288438-640d-4ac3-b490-e5923baaea4a | easy | khushidahi | 2026-08-16 14:50:56.794134+00:00 | succeeded | be0845006226563a3ad335aadc7f3bf2ddf95a2cb15d95202f742eac0206ad22 | 25,008 | null | """R123: clean factorized digit-pair transition ablation.
This candidate is deliberately diagnostic. It trains on evaluator-owned final
labels at several depths so that one shared transition must compose. The model
never converts N or X to native numbers and never implements multiplication or
modular reduction.
The... | {
"id": "26288438-640d-4ac3-b490-e5923baaea4a",
"created_at": "2026-08-16 14:50:56.794134+00:00",
"db_md5": "1d3aac24eb42c8bc378cee0b993f56ea",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "745d85ba-3a82-49fa-9be4-b01ddcd2f0a3",
"tier": "easy",
"dataset_id": "e10",
"status": "suc... | {
"score": {
"mean_loss": 1.059126690030098,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.8019596338272095
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.40500202775001526,
"example_count": 125,... |
262c8d70-5f26-4070-9286-7d8da88e4af8 | easy | karanganesan | 2026-08-05 11:54:18.201901+00:00 | succeeded | 583255644591423dc68784c8b9de477c91a077e872a59848fd5764f7261ce90e | 22,081 | null | """Parametric looped-transformer family (P1).
One weight-tied transformer block applied k times in latent space. Config
flags cover four P1 families with one file:
looped recall=0 gated=0 tfilm=0 plain weight-tied loop
looped-recall recall=1 re-inject the input embedding each
... | {
"id": "262c8d70-5f26-4070-9286-7d8da88e4af8",
"created_at": "2026-08-05 11:54:18.201901+00:00",
"db_md5": "063e3524c812ff4e70a81897d5db6f97",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "3b5ca395-71cd-4fbc-ab25-a44595aa541c",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 13.621500730514526,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.036666667088866234
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 19.618684768676758,
"example_count": 10... |
262e9d1e-0f1d-4ef1-b691-a39c9e1a9f1d | easy | erdavis0 | 2026-08-22 02:45:50.731384+00:00 | succeeded | ad3cfa4ed3a309cd88e9fbab62f457059a8a4be249920a3faee4d638b54d7bbd | 5,712 | null | """Padding-invariant bidirectional fused GRU with global context."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
)
WIDTH = 48
HIDDE... | {
"id": "262e9d1e-0f1d-4ef1-b691-a39c9e1a9f1d",
"created_at": "2026-08-22 02:45:50.731384+00:00",
"db_md5": "f6574666cb73e6fedd6ea419febb213d",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "71116f9b-d0bb-4eaa-8a80-b94d0c88b6c0",
"tier": "easy",
"dataset_id": "e7",
"status": "succeeded",... | {
"score": {
"mean_loss": 6.558822393417358,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.029010696336627007
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.067569732666016,
"example_count": 85,
... |
263793e8-88ef-4a0d-8eef-bc5d2178a998 | easy | nikhilkumar26 | 2026-08-30 02:56:52.805079+00:00 | succeeded | e18ede76098bfe349b22cc05155eaa594ffc2e579f740d3ea4a1af7dcbb76b86 | 10,944 | null | """T-adaptive residue TRM: Fourier MLP square-map, detached unroll.
T is the unroll count only. Each step classifies the next residue in 0..N-1.
Digit logits and residue logits come from the last active step. Recurrence
state is detached so T=1 can train a stable one-step map that then composes.
"""
from __future__ i... | {
"id": "263793e8-88ef-4a0d-8eef-bc5d2178a998",
"created_at": "2026-08-30 02:56:52.805079+00:00",
"db_md5": "0706cedaf4f7d41e42617d55a9afa639",
"submitter": "Nikhil Kumar",
"github_login": "nikhilkumar26",
"run_id": "ad557d2e-7a10-493a-95e0-f659aa5ab4e0",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 0.2834542468190193,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.9483333230018616
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.12361232936382294,
"example_count": 100... |
26444ce9-d0e3-4222-85cd-bd294833e6b2 | easy | khushidahi | 2026-08-16 13:28:12.392839+00:00 | succeeded | 69259bf87c3127d5825c83ab56194692b9042bd88b9bbbd75b323c9dedd598cf | 24,432 | null | """R111: small-input curriculum for a quadratic recurrent transition.
This candidate is deliberately diagnostic. It trains on evaluator-owned final
labels at several depths so that one shared transition must compose. The model
never converts N or X to native numbers and never implements multiplication or
modular red... | {
"id": "26444ce9-d0e3-4222-85cd-bd294833e6b2",
"created_at": "2026-08-16 13:28:12.392839+00:00",
"db_md5": "e703335d4083308259a0ac4cc6534eaf",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "ee6f2e71-a19a-4c87-8079-ffa7eb5552cc",
"tier": "easy",
"dataset_id": "e3",
"status": "succ... | {
"score": {
"mean_loss": 2.537668757684714,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0056250000605359675
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.561662455792682,
"example_count": 800... |
26459a55-69cd-4070-9b16-61039d38b855 | easy | sankalp1999 | 2026-08-04 11:59:02.673744+00:00 | succeeded | c8616e80a03395bbb6b5e4098500af8004719cabe619a451c468d1ea365afea6 | 8,284 | null | """Bidirectional transformer for the squaring task, looped or plain.
This is the general form: attention over whatever tokens arrive, so a larger
modulus with more digits needs no code change, unlike a flattened-grid MLP
that ties every absolute slot to its own weights.
LOOPED=1 applies ONE weight-shared block T time... | {
"id": "26459a55-69cd-4070-9b16-61039d38b855",
"created_at": "2026-08-04 11:59:02.673744+00:00",
"db_md5": "34baac31831bb7aa5ffc5bd81c7c8a47",
"submitter": "Sankalp ",
"github_login": "sankalp1999",
"run_id": "278d5343-8176-46d1-a0b1-6d16d0a64100",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 7.8387651443481445,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.029999999329447746
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.089754104614258,
"example_count": 100... |
264742f4-5f35-4f36-945b-9b8bb5e69fa9 | easy | DDanlov | 2026-08-24 04:24:39.400365+00:00 | succeeded | 47d23b38d324c9d1c7067f357ec89e1ce45195551892568829c0e312d7758c5b | 40,843 | null | """
Unified Hybrid Masked Deliberation Model (44.0M Parameters) - Pure Skipless, No Cross-Attention
Architecture:
- Scale: dim=1536, num_heads=24, d_ff=6144 (Exactly 43,978,752 parameters)
- Hybrid Prefix-Bidirectional + Causal Target Attention Mask
- Token Layout: [Prompt (L_p), 4 Pause Registers, Target Tokens (L_tgt... | {
"id": "264742f4-5f35-4f36-945b-9b8bb5e69fa9",
"created_at": "2026-08-24 04:24:39.400365+00:00",
"db_md5": "ab3b58571e452f5b99987fcc144dc4b9",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "a28828af-4c3e-4bf9-a9ce-2cee3c54f1cd",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 3.2698811945215995,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0008333333333333334
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.2918704090631596,
"example_count": 6... |
264bc2a5-3d54-4875-89f1-dcaeb34aebd3 | easy | jordanrubin | 2026-08-01 20:40:55.418508+00:00 | succeeded | 2c6adbc91db77091eaa1fdb53b2b15fdc6e05306608a2440be5e6a2aa2591b8a | 18,183 | null | """Looped bidirectional selective-SSM (Mamba-style), pure PyTorch.
The statefulness bet: input-dependent SSM dynamics along the token axis,
with the block weight-tied and applied NUM_LOOPS times in depth with input
injection — recurrence in depth is where the task's T-iteration lives.
The selective scan is computed i... | {
"id": "264bc2a5-3d54-4875-89f1-dcaeb34aebd3",
"created_at": "2026-08-01 20:40:55.418508+00:00",
"db_md5": "35fec302dbad6ab1de27b5092ede0b6a",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "99037a14-38ef-4090-8abd-dfa746297544",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 4.2026439594794995,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007499999832361937
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.27298256447497,
"example_count": 600,... |
2653b065-9ad6-445a-9d60-89db599ad53b | easy | liam-gb | 2026-08-13 01:52:48.486333+00:00 | succeeded | a922a24f6d897ff79c7df36202bf5adaf215adc204bf4e9750ac5097b36dc04f | 17,291 | null | """rns_pm_gather_ct_widet_rank: prime-level state with a CRT rank channel.
Base:
The wide readout is what makes long answers reachable; the trim is what pays
for it, by not running loop iterations whose update is multiplied by zero.
Base docstring:
Single change vs rns_pm_gather_ct: the readout carries one
distance... | {
"id": "2653b065-9ad6-445a-9d60-89db599ad53b",
"created_at": "2026-08-13 01:52:48.486333+00:00",
"db_md5": "38bc09c3be37222b2dcdfacc053d0c51",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "10ab06ea-f6d2-42ce-b03c-401a8fc08c1a",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 3.044544082705502,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00999999995653828
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.0696303844451904,
"example_count": 600,... |
2656b5ef-7214-4644-bdc3-ae90143c6586 | easy | shreyash-chonkie | 2026-09-01 00:12:26.253659+00:00 | succeeded | e6ecb8ff9107399e69c186bf5f260721f42b1e1b9ede153e5411b59b18b2a7a6 | 8,886 | null | """Through-time recurrent Transformer with checkpoint supervision."""
from __future__ import annotations
import math
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert... | {
"id": "2656b5ef-7214-4644-bdc3-ae90143c6586",
"created_at": "2026-09-01 00:12:26.253659+00:00",
"db_md5": "39cd3825fa6de000cdd3185ee940891e",
"submitter": "Shreyash",
"github_login": "shreyash-chonkie",
"run_id": "1dc78e15-2871-478b-a822-330a2d7e6a3c",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 10.19569444656372,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05166666768491268
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 10.040533065795898,
"example_count": 100,... |
265b32e3-41a0-4044-9ef8-34d80cd1b2d1 | easy | DDanlov | 2026-08-18 10:35:41.232205+00:00 | succeeded | 9e2e886a7f72218c6f6788250c79c4edcba3908cfb09a02021dbdad2941008ba | 23,584 | null | """
Official Submission for One Layer Deeper Challenge
Architecture: 4-Layer Recurrent Core Pure No-Norm No-Skip DKS Reasoning Model:
- 4 Distinct Pure No-Norm No-Skip Transformer Layers in the Recurrent Core: L1 -> L2 -> L3 -> L4
- Unrolled T times for serial modular squaring
- dim=512, 16 heads, d_ff=1024, RohanShamp... | {
"id": "265b32e3-41a0-4044-9ef8-34d80cd1b2d1",
"created_at": "2026-08-18 10:35:41.232205+00:00",
"db_md5": "d6174d2a9211082098f18de894eca987",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "c2a8c7ea-c898-44f6-a656-cf4de91533bb",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.038882458996896,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.011666666703919571
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.0024302370512665,
"example_count": 100... |
26618db5-bac7-4271-988f-8f750e44f66a | easy | DDanlov | 2026-08-17 12:16:06.020713+00:00 | succeeded | 210eee0835be106b219294fa26fa38c7b2dbe27025c042c8e2ecf9e31a3fea8e | 31,729 | null | """
Scalable Recurrent Deliberation Transformer (ScalableDEPHNet)
============================================================
Architecture:
- Pure No-Norm: Single unit RMS normalization strictly isolated to the input token embedding layer.
Zero LayerNorm / RMSNorm inside recurrent deliberation blocks.
- Pure No-Skip... | {
"id": "26618db5-bac7-4271-988f-8f750e44f66a",
"created_at": "2026-08-17 12:16:06.020713+00:00",
"db_md5": "706c104bb8c2928331e3cda0d6b3b70a",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "1b824370-c395-44cb-be6b-6cecc9de330c",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.3351317213063227,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.059999999726812046
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.679548978805542,
"example_count": 100... |
267133ca-8ff8-46c8-8f14-e7bf6fe58e9a | easy | himalalps | 2026-08-06 16:11:35.181072+00:00 | succeeded | ac9442f4cbc0e698a66ddc4e398351b821c576fa4075200064072d78a0f99d60 | 63,514 | null | """Algorithmically constrained latent recurrence for One Layer Deeper.
One masked cross-attention pass creates two independent representations:
``state_0`` from X only and an immutable context from N only. T is excluded
from both representations and is used solely to choose how many times the
learned recurrent MLP tr... | {
"id": "267133ca-8ff8-46c8-8f14-e7bf6fe58e9a",
"created_at": "2026-08-06 16:11:35.181072+00:00",
"db_md5": "d59cf47a66a85a7d46aa9e17ddcd3959",
"submitter": "Haoyu Tang",
"github_login": "himalalps",
"run_id": "4fc99bcc-6010-4d01-8f6d-0a6ce933881b",
"tier": "easy",
"dataset_id": "e5",
"status": "succe... | {
"score": {
"mean_loss": 3.1399665863865667,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.010833333438883225
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.110630512237549,
"example_count": 600... |
2673fa09-dac1-47a6-8447-eb8ebe578420 | easy | yunjiangster | 2026-08-28 16:48:36.667470+00:00 | succeeded | 6fc226e19eb49e01e092e3952cedd3494d1d4cd4523523f0c4bb37845e10b5e0 | 15,403 | null | """Generic arithmetic-circuit probe with a low-degree polynomial prior.
Two hypothesis tiers share one categorical selector, so neither is privileged by
construction beyond its share of the bank:
Tier A (prior) -- every degree-<=4 integer polynomial in a tapered coefficient
box, evaluated exac... | {
"id": "2673fa09-dac1-47a6-8447-eb8ebe578420",
"created_at": "2026-08-28 16:48:36.667470+00:00",
"db_md5": "f70cc08199b028a0773f73df6a7bdf03",
"submitter": "Yunjiang Jiang",
"github_login": "yunjiangster",
"run_id": "81b4a450-393f-41a2-b273-05e77d699c93",
"tier": "easy",
"dataset_id": "e1",
"status":... | {
"score": {
"mean_loss": 0.004424780840054154,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.004424781072884798,
"example_count": 100,
... |
267411b2-415c-4127-ab2b-22efb5564318 | easy | jordanrubin | 2026-08-13 17:57:02.700489+00:00 | succeeded | e2245c74dfcfe7d00f8aeee61890a07c07a0d89977c245e3844890ce7f77b424 | 27,175 | null | """Weight-tied MLP-cell loop over soft digit states (mlploop).
Cell = MLP over [expected digits, pairwise digit products, RNS residue
simplices (fixed differentiable mixing over Z_p), N digits]. The cell that
learns one-step modular squaring from direct pairs (52% unseen at 14k rows,
day-20 screen) inside the exact-T ... | {
"id": "267411b2-415c-4127-ab2b-22efb5564318",
"created_at": "2026-08-13 17:57:02.700489+00:00",
"db_md5": "3f710550312ea469d13b008d83980242",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "06e67e6f-c8d7-4a70-9b34-44508abb8aa8",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.909317314024749,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.028333332234372694
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2734933626491394,
"example_count": 600... |
267e0429-6d97-4258-9a54-5a5bd8f82ec0 | easy | AFeinfield | 2026-08-27 21:40:32.459500+00:00 | succeeded | 9528eabf8952b1d49b72ee2128a5b6939c5f1ff8526b376523933792fe4a8ed4 | 10,124 | null | """148: STRUCTURED-LEARNED ARITHMETIC MACHINE.
The wiring encodes that the task is iterated digit arithmetic (place
grid, i+j=k pair products, carry sweeps, conditional reduction attempts,
T iterations) — inductive bias in the same category as convolution. All
VALUES are learned end-to-end by small shared operators wh... | {
"id": "267e0429-6d97-4258-9a54-5a5bd8f82ec0",
"created_at": "2026-08-27 21:40:32.459500+00:00",
"db_md5": "f5725a25b6993d810748ccd36c1495c5",
"submitter": "AFeinfield",
"github_login": "AFeinfield",
"run_id": "6107c501-200e-40b3-9a23-5a826f8075f0",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.662691521607359,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0020833333457509675
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.6595235662075436,
"example_count": 60... |
2681add8-8f46-4c13-a90b-6c3348b77685 | easy | AFeinfield | 2026-08-13 00:08:50.233475+00:00 | succeeded | a28ef4dd07383f63d8be5b88a5ca1170be806f78e6c78aa7eb4889e53dd872a5 | 7,223 | null | """Looped (weight-tied) bidirectional Transformer, d=256, 8 iterations.
013: 012 + GrokFast-EMA (a=.98, l=2) — revived: now targets the actual
post-memorization transition (012: 98.8% train / 0% test @ 918 steps)
to amortize the evaluator's ~200ms/iteration overhead into 8x more steps.
prior. Same parameter count orde... | {
"id": "2681add8-8f46-4c13-a90b-6c3348b77685",
"created_at": "2026-08-13 00:08:50.233475+00:00",
"db_md5": "e0d1ffcb34a1e72cdf53439b11e836f4",
"submitter": "AFeinfield",
"github_login": "AFeinfield",
"run_id": "3f3b0aa7-5d53-4f8a-94f6-b19a182c9771",
"tier": "easy",
"dataset_id": "e1",
"status": "succ... | {
"score": {
"mean_loss": 8.746458530426025,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04166666604578495
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.358805656433105,
"example_count": 100,
... |
269fcf71-43a0-4dbb-b224-a808dcd8f7bb | easy | 0Chris5R | 2026-08-02 16:37:44.413407+00:00 | succeeded | c9c67ebc93e5cf924b538d3d3c8d66620939007a5cdae5ef506252de990ae74d | 24,379 | null | """Shared-digit Neural GPU with late full-objective EMA evaluation.
The model aligns decimal digits by significance, forms a learned low-rank
second-order feature map of the current state, and evolves a two-dimensional
workspace with the original convolutional-GRU equations. Each arithmetic
transition applies the sam... | {
"id": "269fcf71-43a0-4dbb-b224-a808dcd8f7bb",
"created_at": "2026-08-02 16:37:44.413407+00:00",
"db_md5": "65cfb2f4109673b3a7f402928e914e44",
"submitter": "Chris ",
"github_login": "0Chris5R",
"run_id": "7444efc6-af5c-4db6-bd57-efec8ccf0969",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 3.101562497590067,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.028333333345750965
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.821115565406902,
"example_count": 600,... |
26a085a9-75e5-4ec7-a7c1-d68fcaa0b2a3 | easy | shirvani-jr | 2026-08-29 04:18:52.489467+00:00 | succeeded | 894e7191c097b72299ae3dcbde61e7e64a45f94fddd519110020addcac824d46 | 46,351 | null | """E5 research arm: categorical-pair neural cellular automaton.
One modular-composition unit starts from the complete 4x4 grid of ordered
value-digit pairs. A position-shared learned map embeds each 100-way pair
identity, local modulus context, and generic grid geometry. One learned 3x3
gated cell then evolves a con... | {
"id": "26a085a9-75e5-4ec7-a7c1-d68fcaa0b2a3",
"created_at": "2026-08-29 04:18:52.489467+00:00",
"db_md5": "b2a36685d6eba2f6f328756893c0aecd",
"submitter": "Ali",
"github_login": "shirvani-jr",
"run_id": "d0d99eb4-fce9-42a9-a6dd-18d61a26b142",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 3.016747247897558,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0004166666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.008693218231201,
"example_count": 600... |
26a6d0f6-4276-4798-ab3f-8bd27799554f | easy | rteehas | 2026-08-19 08:23:00.230020+00:00 | succeeded | 89938d6e70952c6215586e8bebe56eb591140b567b9149d5ca797584302788db | 41,409 | null | """Anonymous marginal-program submission.
Mechanism
---------
The model is a differentiable probabilistic program. Its forward pass
computes, exactly, the answer-token predictions of a small structured latent
hypothesis space:
(prompt segmentation = which token types head the three fields)
x (field-role permutat... | {
"id": "26a6d0f6-4276-4798-ab3f-8bd27799554f",
"created_at": "2026-08-19 08:23:00.230020+00:00",
"db_md5": "f8da87d40944b0aee64044f85b4c7e51",
"submitter": "Ryan Teehan",
"github_login": "rteehas",
"run_id": "70905175-7953-4d2b-bd63-03016e4f8485",
"tier": "easy",
"dataset_id": "e4",
"status": "succee... | {
"score": {
"mean_loss": 6.757862335637462,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0022916666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.8135250970149555,
"example_count": 12... |
26ac7973-3737-4824-9a27-e276e8d740be | easy | erdavis0 | 2026-08-24 16:14:57.564612+00:00 | succeeded | 51c1ac6f1c1b63d62814de1d34409c49ecd688b4e0533ed3562b45bbe4363796 | 9,114 | null | """Iterative latent workspace for soft prompt-role discovery and slot readout."""
from __future__ import annotations
import math
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_s... | {
"id": "26ac7973-3737-4824-9a27-e276e8d740be",
"created_at": "2026-08-24 16:14:57.564612+00:00",
"db_md5": "8c1df270c6f26cdb1236360125a41b1a",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "fd7e11dd-8bb4-48f8-b418-9b2433c54788",
"tier": "easy",
"dataset_id": "e2",
"status": "succeeded",... | {
"score": {
"mean_loss": 9.041173934936523,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008541666902601719
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.934642791748047,
"example_count": 300,... |
26bba0a4-51f2-4d66-8e28-f01520dc3baa | easy | erdavis0 | 2026-08-22 02:26:17.810943+00:00 | succeeded | dd717b6474fda04ebdb5466cb2037b63cbf06831fdcb8eb294b90d300288916b | 5,797 | null | """Soft field factorization with multiplicative field interactions."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
)
WIDTH = 64
FIE... | {
"id": "26bba0a4-51f2-4d66-8e28-f01520dc3baa",
"created_at": "2026-08-22 02:26:17.810943+00:00",
"db_md5": "464075811335119d26ad5014b80665e5",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "b0bb6ef5-b681-48b8-8e67-a72a1c88a19b",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",... | {
"score": {
"mean_loss": 4.322443246841431,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05166666768491268
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.172218322753906,
"example_count": 100,
... |
26c2b7f6-f0ca-49be-9ac0-653e272a6c00 | easy | jordanrubin | 2026-08-08 10:57:17.580770+00:00 | succeeded | 19e434c7e9fc9d9d601292a0f5c0a4083813bc7790caf87c0582fc1bbcc2cad8 | 16,909 | null | """Autonomous digit-state recurrent Transformer for repeated modular squaring.
The model is deliberately organized around one learned transition:
p_0 = right_aligned_decimal_digits(x)
p_{k+1} = F_theta(p_k, decimal_digits(N))
The same two-block Transformer cell is applied exactly T times. T is used only
as ... | {
"id": "26c2b7f6-f0ca-49be-9ac0-653e272a6c00",
"created_at": "2026-08-08 10:57:17.580770+00:00",
"db_md5": "671c15c2db39c821999bae96a7defa51",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "61307b0c-c5ec-49df-bf29-b11a343e72dd",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 4.5262259922321775,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009999999698872367
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.499489360860645,
"example_count": 600... |
26c34d20-6706-416d-9260-c065fdeb8cc5 | easy | sapient-sapiens | 2026-08-14 18:40:43.581371+00:00 | succeeded | 28ba469aae3e42ca679270105f219b6bd47433fa80f83dc154b3c3e55e44c7c3 | 14,299 | null | """Digit-space recurrent operator for repeated modular squaring.
Nothing here converts a token into a decimal number. Digits stay as embeddings,
so what a digit is worth, and every carry and reduction, is learned. That also
sidesteps a hard numeric wall: at Hard's modulus scale `x^2` runs past the 53-bit
exact-integer... | {
"id": "26c34d20-6706-416d-9260-c065fdeb8cc5",
"created_at": "2026-08-14 18:40:43.581371+00:00",
"db_md5": "1efadda9e01ae84ec2857432a666c542",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "b4c195b9-a52c-41cd-b1f3-9d67efbf9afc",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 2.2929393479915317,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.002916666691501935
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.352475570322687,
"example_count": 600... |
26c80ccd-50a0-4aea-9e0b-ca9b7b260c3b | easy | yashkant | 2026-08-31 19:08:30.989024+00:00 | succeeded | 14f97b18c077e852b4493ec4a90d0373ae0cff1013abc900d8ba09dcf50cdd0c | 106,610 | null | """One Layer Deeper submission -- a learned register machine.
Architecture (all outputs produced by learned parameters; no task arithmetic on the answering path):
a bank of R=6 digit registers (W=16 positions each) carried as DISCRETE tokens between micro-steps: five
data registers and one COUNTDOWN register holding t... | {
"id": "26c80ccd-50a0-4aea-9e0b-ca9b7b260c3b",
"created_at": "2026-08-31 19:08:30.989024+00:00",
"db_md5": "2266578d8daa341d5e4286574e240b40",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "1e315976-b1a8-49ae-a738-ad94dd187cad",
"tier": "easy",
"dataset_id": "e1",
"status": "succeed... | {
"score": {
"mean_loss": 3.5465598106384277,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.3218367099761963,
"example_count": 100,
"exa... |
26ca4176-8d21-4e3f-a9d2-4e4c20820682 | easy | khushidahi | 2026-08-15 14:46:59.861920+00:00 | succeeded | 22d63ef11024cc85fcdb2a846ead5dcf820ddcaa7abbba4ff79f42efbc0a47fc | 16,528 | null | """R6 structured decimal-workspace model for One Layer Deeper.
The model does not implement multiplication, modular reduction, or the public
recurrence. It uses the public tokenizer structure to build learned, right-
aligned decimal tapes for N, X, and T, then applies a shared neural transition
to a mutable work tape.... | {
"id": "26ca4176-8d21-4e3f-a9d2-4e4c20820682",
"created_at": "2026-08-15 14:46:59.861920+00:00",
"db_md5": "f77e8c3873a1258f8bb9051c502f4837",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "e040e341-eff3-4ccf-a7cb-aba72c5e596f",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.1604453325271606,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007916666800156236
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1608076095581055,
"example_count": 60... |
26d03967-5cba-4f25-81ad-73d5a0314d98 | easy | jordanrubin | 2026-08-07 01:15:34.381018+00:00 | succeeded | f0055280d0601984f0e1916287c802b30a5f9748471523e732881bed2e665ab7 | 14,787 | null | """Autonomous digit-state recurrent Transformer for repeated modular squaring.
The model is deliberately organized around one learned transition:
p_0 = right_aligned_decimal_digits(x)
p_{k+1} = F_theta(p_k, decimal_digits(N))
The same two-block Transformer cell is applied exactly T times. T is used only
as ... | {
"id": "26d03967-5cba-4f25-81ad-73d5a0314d98",
"created_at": "2026-08-07 01:15:34.381018+00:00",
"db_md5": "a2278c00d16203bc738f562ba8b269c9",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "cc0c3f3f-9171-410a-a1d1-220c9240bad3",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.1761966807834647,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006249999860301614
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.178716673712025,
"example_count": 600... |
26d0c03b-c10d-4928-907b-9e02d11ce935 | easy | viridale | 2026-08-24 22:38:15.022008+00:00 | succeeded | 092919da1f2c8fa6ac30cd7a0f2e74faa9b39d15bb751d68d4de05003df3e048 | 16,081 | null | """Factorized-semiring Transformer digit operator.
hypothesis: v864's unrestricted 10x10x96 pair table can interpolate digit-pair
identities; forcing every pair feature to factor as a product of two shared
learned digit codes makes endpoint depth constraints identify a reusable algebra
axis: arch
target: Easy ... | {
"id": "26d0c03b-c10d-4928-907b-9e02d11ce935",
"created_at": "2026-08-24 22:38:15.022008+00:00",
"db_md5": "835fc1092b6502d011f52d143b2effee",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "495b401e-45c3-4b06-817e-7cfa5a0f7271",
"tier": "easy",
"dataset_id": "e6",
"status": "succe... | {
"score": {
"mean_loss": 2.038322329521179,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.37409912049770355
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.7061128616333008,
"example_count": 60,
... |
26d59941-b053-468a-a0ff-524ce64fa5d1 | easy | sapient-sapiens | 2026-08-09 21:28:33.585497+00:00 | succeeded | 3aa9f5b2bd9148ddf88f7915587d336a010cf47de178f313ed16c7b05a557dc3 | 15,567 | null | """T-independent categorical recurrence with learned latent-depth selection."""
from __future__ import annotations
import math
import time
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
Toke... | {
"id": "26d59941-b053-468a-a0ff-524ce64fa5d1",
"created_at": "2026-08-09 21:28:33.585497+00:00",
"db_md5": "150888dbdeeac3dfffad4ef02d444831",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "e335efe6-6f54-41d5-b65e-7f27e66cceb3",
"tier": "easy",
"dataset_id": "e1",
"status": "su... | {
"score": {
"mean_loss": 2.11162742622959,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006666666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.028711439068638,
"example_count": 100,
... |
26d62916-74d8-4d78-98be-7eda39b0473c | easy | DDanlov | 2026-08-29 12:00:27.171344+00:00 | succeeded | 6b50a2a36e57dc7ef157759a6e0b72549bdc6570596fee0cd178dc783385b5a8 | 20,526 | null | """
submission_eqprop_deep_v2.py
Exact, faithful implementation of the Modern Continuous Hopfield Energy Deliberator (eqprop_deep)
with 2 coupled physical layers (L=2), Mimetic Gram Attention Initialization,
Conformal Symplectic Velocity Verlet integration, and full RohanShampoo optimizer.
"""
import math
import time
... | {
"id": "26d62916-74d8-4d78-98be-7eda39b0473c",
"created_at": "2026-08-29 12:00:27.171344+00:00",
"db_md5": "1b7f4a7b89241b85f2c4a8e55303022c",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "51840c1a-af63-47c2-8797-b0b4797e952f",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.2463026907390855,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007083333333333334
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2537483217470315,
"example_count": 60... |
26d84ffe-2d47-4af4-a733-1231fd050fc7 | easy | k-penchev | 2026-08-22 22:17:43.223043+00:00 | succeeded | 6de120e7e059d02f3c9a5537e58372cbb8356ceadfb680139ec43b4bed5cc01a | 4,580 | null | """Two-loop tied Transformer with digit places and immutable input recall."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
)
D_MODEL... | {
"id": "26d84ffe-2d47-4af4-a733-1231fd050fc7",
"created_at": "2026-08-22 22:17:43.223043+00:00",
"db_md5": "714cdba1d1a87e2cac5bdbc2bd5db2a4",
"submitter": "Kaloyan Penchev",
"github_login": "k-penchev",
"run_id": "d08cea50-8ddc-44bb-8f6d-a8d1a3a86647",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 2.634168565273285,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04833333380520344
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.2704975605010986,
"example_count": 100,... |
26dbe306-c656-4d7c-8a36-28ba6a3bae21 | easy | karanganesan | 2026-08-04 02:00:52.741595+00:00 | succeeded | a9f40f8a9044651de96df9da960df48e0e547f70ca238a374ae59371c7c514b0 | 11,979 | null | """Parametric looped-transformer family (P1).
One weight-tied transformer block applied k times in latent space. Config
flags cover four P1 families with one file:
looped recall=0 gated=0 tfilm=0 plain weight-tied loop
looped-recall recall=1 re-inject the input embedding each
... | {
"id": "26dbe306-c656-4d7c-8a36-28ba6a3bae21",
"created_at": "2026-08-04 02:00:52.741595+00:00",
"db_md5": "e6071d7335a82dfd09d1a16f8191eeb3",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "b2652bf5-ac44-4f7b-9c8a-616aa32c7110",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 2.530966281890869,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.06833333149552345
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.273712158203125,
"example_count": 100,
... |
26e64599-9838-4720-8fbd-6e7fc54a75cc | easy | khushidahi | 2026-08-06 00:35:12.481010+00:00 | succeeded | 1949e5fca9b57703b8a4a958e04055e3d7b33766b22f4a99374c979a8fb25977 | 17,452 | null | """R6 structured decimal-workspace model for One Layer Deeper.
The model does not implement multiplication, modular reduction, or the public
recurrence. It uses the public tokenizer structure to build learned, right-
aligned decimal tapes for N, X, and T, then applies a shared neural transition
to a mutable work tape.... | {
"id": "26e64599-9838-4720-8fbd-6e7fc54a75cc",
"created_at": "2026-08-06 00:35:12.481010+00:00",
"db_md5": "c42572817607a8407413fc7bf84cd271",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "3f55c128-774c-4492-bf9c-845290162b64",
"tier": "easy",
"dataset_id": "e4",
"status": "succ... | {
"score": {
"mean_loss": 2.077978253364563,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005520833423361182
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.076930046081543,
"example_count": 1200... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.