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 |
|---|---|---|---|---|---|---|---|---|---|---|
2b0bcb5c-fa8f-44c4-8725-401a3815f947 | easy | DDanlov | 2026-08-10 10:02:53.089900+00:00 | succeeded | d213b8c1463d69d5c36c0bea653327c8a4dd971dcc1d0ed9c865b4334dfb0a6d | 16,157 | null | from __future__ import annotations
import sys
import os
import math
import time
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim.optimizer import Optimizer
VOCAB_SIZE = 17
DIGIT_OFFSET = 7
def init_cutting_the_skip(module):
"""SVD-based orthogonal initialization to ensure initia... | {
"id": "2b0bcb5c-fa8f-44c4-8725-401a3815f947",
"created_at": "2026-08-10 10:02:53.089900+00:00",
"db_md5": "0aba22a44da552196030465640ea3b33",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "cb63626e-6f9e-4f5d-82f8-1ce54a6aa9d0",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.3235897831883623,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.08333333373069764
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.3398524682913253,
"example_count": 100... |
2b0e137d-aa1d-4878-b3de-736f03b23667 | easy | shreyash-chonkie | 2026-08-23 19:10:39.723510+00:00 | failed | c3695b8bfc9cf1913b213b4c8e46e1f19e81aa64d832d4615d0df0679c365f3a | 8,233 | null | """Four-stage recurrent register Transformer for One Layer Deeper."""
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 = 1024
... | {
"id": "2b0e137d-aa1d-4878-b3de-736f03b23667",
"created_at": "2026-08-23 19:10:39.723510+00:00",
"db_md5": "089c7bc3558fdaef1af6b37e19105e43",
"submitter": "Shreyash",
"github_login": "shreyash-chonkie",
"run_id": "d73f2e48-00df-4d7a-93bf-a565efaa9c27",
"tier": "easy",
"dataset_id": "e6",
"status": "... | null
|
2b0e4b86-61a8-4482-b603-3e357e3c9cb8 | easy | KaustubhKumar05 | 2026-08-25 16:11:34.586377+00:00 | succeeded | 03d0c373e594aa5715323fd2ba901ecc5a0250f4deb9d9e32371e95dc8a36c27 | 9,417 | null | """abacus: faithful port of the Abacus recipe (McLeish et al., arXiv:2405.17399).
The load-bearing idea is the embedding. Each digit is indexed by its
SIGNIFICANCE WITHIN ITS OWN NUMBER (0 = least-significant), from a table
SHARED across N, x and T. Digit k of N and digit k of x therefore land on the
same embedding an... | {
"id": "2b0e4b86-61a8-4482-b603-3e357e3c9cb8",
"created_at": "2026-08-25 16:11:34.586377+00:00",
"db_md5": "80b03c6db927f2e12cf645f2453373df",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "d2c5a4de-7bd4-4477-98c8-f1f16a26961d",
"tier": "easy",
"dataset_id": "e6",
"status": "succee... | {
"score": {
"mean_loss": 5.862699031829834,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0335585605353117
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.780012607574463,
"example_count": 60,
... |
2b14c330-dccc-40e9-99d9-08ba01b89a89 | easy | liam-gb | 2026-08-12 11:58:25.478125+00:00 | succeeded | 63560835218f6cc8d2426c2dfe473ed6ed0b400ea97d7ccbdbf29d9eabbb74fc | 13,923 | null | """rns_pm_gather_ct_wide: the champion with a dataset-sized digit head.
Single change vs rns_pm_gather_ct: the readout carries one
distance-from-units slot per sequence position instead of three, so
answers longer than three digits are reachable. Behaviour on shorter
answers is unchanged — the extra slots are simply n... | {
"id": "2b14c330-dccc-40e9-99d9-08ba01b89a89",
"created_at": "2026-08-12 11:58:25.478125+00:00",
"db_md5": "6160238a93c003b99ce5e4e8e2555bad",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "bc385b2c-ec71-459d-976d-233e1053436d",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 4.428702086210251,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.4416666850447655
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.899816513061523,
"example_count": 100,
... |
2b157d78-4811-466c-882a-aa60f190c05e | easy | lpbb | 2026-08-07 15:24:08.521445+00:00 | succeeded | 94ba6f5ead7817b61d57c5097eca8fef3f8f7a1b594d8730ffa9c18c8a3a05b4 | 19,662 | null | """VARIANT S3 `smooth` — loss-shaping axis.
Base (submissions/looped) + three changes:
* LABEL SMOOTHING 0.1 via the token_training_loss hook (legal API
surface; eval always uses plain CE + exact match, and smoothing does
not move argmax). Memorization's signature in every run so far is
exploding confid... | {
"id": "2b157d78-4811-466c-882a-aa60f190c05e",
"created_at": "2026-08-07 15:24:08.521445+00:00",
"db_md5": "2324c98bdff56569b453f82975619ccb",
"submitter": "Lpbb",
"github_login": "lpbb",
"run_id": "4d562ecb-5a59-412f-8389-84f33882525c",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded",
"s... | {
"score": {
"mean_loss": 4.135654594562408,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008750000004656613
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.192871469883644,
"example_count": 800,... |
2b178c78-0a99-440f-98fc-f74e3b32c9ba | easy | 0Chris5R | 2026-08-02 17:08:18.129311+00:00 | succeeded | 5e67ee3441acbb200c18235c84d4a02f726d45a0fe1c631d57dadc31f215f36a | 25,154 | 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": "2b178c78-0a99-440f-98fc-f74e3b32c9ba",
"created_at": "2026-08-02 17:08:18.129311+00:00",
"db_md5": "d0b6cbcebe7434932cb9080749818ed3",
"submitter": "Chris ",
"github_login": "0Chris5R",
"run_id": "79906b27-4f86-4086-b1ca-6369af3aeed3",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.290615523481053,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.010833333358168603
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1894740930052614,
"example_count": 600... |
2b1995f2-bcce-44df-ac0d-25c60dce9b81 | easy | nikolageorgiev2000 | 2026-08-04 12:34:28.561938+00:00 | succeeded | bcb781bfdf7ba38a0a37eaa6ed064a66f4dfd339aab3b052a335c7c7073403a1 | 26,498 | null | """T-composed categorical-digit reasoner with fully learned pair/reduction maps."""
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,
asser... | {
"id": "2b1995f2-bcce-44df-ac0d-25c60dce9b81",
"created_at": "2026-08-04 12:34:28.561938+00:00",
"db_md5": "f86607b469fd9dd75773b8607f53025a",
"submitter": "Nikola Georgiev",
"github_login": "nikolageorgiev2000",
"run_id": "64555f4a-ecd4-47ea-9409-6d3f2350b692",
"tier": "easy",
"dataset_id": "e5",
"s... | {
"score": {
"mean_loss": 2.5893809822598803,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005833333333333334
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.262756957066968,
"example_count": 600... |
2b1e8e33-b6d0-4228-9205-ac05e9e0953a | easy | liam-gb | 2026-08-13 15:48:55.236496+00:00 | succeeded | b71629385c66fb4ca3d396a0c332954df4157ea6dfc0ad6b65e69b5f15a661b0 | 22,129 | null | """Bake-off template: one contract-legal file, three architecture families.
Generated variants live in submissions/bakeoff_*/submission.py. The CFG block
between the BEGIN_CFG/END_CFG markers is rewritten by scripts/mkvariant.py; the
rest of the file is identical across every run in the tournament, so a score
delta is... | {
"id": "2b1e8e33-b6d0-4228-9205-ac05e9e0953a",
"created_at": "2026-08-13 15:48:55.236496+00:00",
"db_md5": "707c410cf8c0f1babbd2432972791036",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "793dbe99-9eb7-4845-a18e-6b2d02b0657f",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 6.211209383902249,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.016666666728754838
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.313647270202637,
"example_count": 600,... |
2b2447a4-fb94-4576-94e5-7af67ceff68c | easy | viridale | 2026-08-15 04:25:48.739339+00:00 | succeeded | 36d227275dee4e35691ffdc6565348f375e09ca1f85c6b636be62c76f5ae9d67 | 16,227 | null | """Translation-shared bidirectional scan reducer.
hypothesis: a translation-shared bidirectional scan can normalize learned i+j coefficient features without the memorizing free endpoint head that failed P128.
axis: arch/loss
target: two distinct official Easy identities carrying a T1 ladder after exact routing and sca... | {
"id": "2b2447a4-fb94-4576-94e5-7af67ceff68c",
"created_at": "2026-08-15 04:25:48.739339+00:00",
"db_md5": "41de004b7b9d60b7f1313f529545cc7a",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "ffbe2060-a908-4ded-ad92-671c95652f54",
"tier": "easy",
"dataset_id": "e8",
"status": "succe... | {
"score": {
"mean_loss": 3.672257661819458,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.18163993209600449
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.821657657623291,
"example_count": 85,
... |
2b2f0bbe-79b3-48f4-8c16-98f6fe057016 | easy | oupadhyay | 2026-08-28 02:47:05.708412+00:00 | succeeded | bb13f70207127b88a0df56ae24f345934a0ca2412de654039ab3196d1e706a04 | 7,646 | null | """Attention-augmented spatial-gating categorical register for the Easy tier."""
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": "2b2f0bbe-79b3-48f4-8c16-98f6fe057016",
"created_at": "2026-08-28 02:47:05.708412+00:00",
"db_md5": "ba38ef28ffed748587800e2fdbcfdd9c",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "69d50895-e4cb-4084-84fb-e121ed407a23",
"tier": "easy",
"dataset_id": "e6",
"status": "s... | {
"score": {
"mean_loss": 2.2554112672805786,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.07680180575698614
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.166299819946289,
"example_count": 60,
... |
2b3863df-9768-4a9a-9570-64345ca0d799 | easy | AFeinfield | 2026-08-15 15:02:59.183801+00:00 | succeeded | 36b88d89175d1fe454da407cfcdbc1e4e7d04a30bc571258ba75052d7c2fe3d6 | 8,152 | null | """Looped (weight-tied) bidirectional Transformer, d=256, 8 iterations.
042: T-indexed arch on e2 (T in {1,2,4} coverage — every low iteration
directly graded by data; Alex's per-depth grading via trajectory sharing).
to amortize the evaluator's ~200ms/iteration overhead into 8x more steps.
prior. Same parameter count... | {
"id": "2b3863df-9768-4a9a-9570-64345ca0d799",
"created_at": "2026-08-15 15:02:59.183801+00:00",
"db_md5": "29cd821c13b7803d35f701ef422e4655",
"submitter": "AFeinfield",
"github_login": "AFeinfield",
"run_id": "e4e0b146-78bf-4a23-abb9-b491426be551",
"tier": "easy",
"dataset_id": "e2",
"status": "succ... | {
"score": {
"mean_loss": 2.248306393623352,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.012083333684131503
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.261932849884033,
"example_count": 300,... |
2b4c4fb0-a419-4aa5-9382-330e30fc9e31 | easy | KaustubhKumar05 | 2026-08-26 13:01:40.230728+00:00 | succeeded | d7147d242833e4afb54cb5521863a8d7e00a7dcd6a7d3c0160335937bf736542 | 12,761 | null | """Neural-GPU-style conv GRU, v2s: v2 + speedrun pack.
- LR: warmup, hold max to 85% of wall-clock, steep late cliff (2026 schedule theory)
- identity-at-init: cell output projection zero-initialized
- batch reuse x6 (multi-pass theory)
Base (v2):
Changes over v1 (each backed by measured results in the literature):
... | {
"id": "2b4c4fb0-a419-4aa5-9382-330e30fc9e31",
"created_at": "2026-08-26 13:01:40.230728+00:00",
"db_md5": "732b18ee025b92e16480548606faa9e5",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "41b1e13f-47ea-41d2-8d0b-f09b62f33701",
"tier": "easy",
"dataset_id": "e3",
"status": "succee... | {
"score": {
"mean_loss": 3.05742668954307,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004375
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.133051785692529,
"example_count": 800,
"e... |
2b4f3ab3-22c6-4af2-86e9-12604daecd11 | easy | jordanrubin | 2026-08-12 05:49:32.591925+00:00 | succeeded | 6072ddd859baaaadd512d39619839cdbe7ff402f38c47bd8ba7941d112bebe6f | 24,918 | 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": "2b4f3ab3-22c6-4af2-86e9-12604daecd11",
"created_at": "2026-08-12 05:49:32.591925+00:00",
"db_md5": "77ff777c577d8e7bf8ce9b05ac674fb7",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "b392ced4-0492-44b1-9b72-12cc925563b1",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.8505236804021785,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.022499999186644953
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2082617786700416,
"example_count": 60... |
2b505b33-8297-447b-b4e0-82b811363149 | easy | amirmeisami | 2026-08-17 19:42:06.331583+00:00 | succeeded | df188853833328b0f7cb6c95b8f18aa803dca6187cccb5c17f083172b749ce0f | 12,657 | null | """T021 Arm D: algebraic position-aware recurrent network."""
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_model_state
D_MODEL, POLY_WIDTH, IN... | {
"id": "2b505b33-8297-447b-b4e0-82b811363149",
"created_at": "2026-08-17 19:42:06.331583+00:00",
"db_md5": "d5bfe7143a3deb8a8807596c2ec61aa8",
"submitter": "Amir Meisami",
"github_login": "amirmeisami",
"run_id": "69ae74d3-11c0-4bf1-a6b8-3b377d3a7d79",
"tier": "easy",
"dataset_id": "e7",
"status": "s... | {
"score": {
"mean_loss": 6.233978985484792,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.061408199862628486
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.8527855454421625,
"example_count": 85,... |
2b551b62-0c12-44dc-b1fa-0be8f20db232 | easy | AFeinfield | 2026-08-28 02:24:29.861661+00:00 | succeeded | 46c757b1560bcf443488956fd3384bd069d12a0f46f1aba2d8e918562d5a0a2e | 8,680 | null | """180: 179 + soft-then-ST composition schedule (hard after 800 calls)
+ weight decay targeted at the MAP (1.0) not the head (0.01).
179 never memorized: hard argmax through a random-init map makes the
early landscape discontinuous. Literature trains the map directly; the
soft phase recovers that, the hard phase resto... | {
"id": "2b551b62-0c12-44dc-b1fa-0be8f20db232",
"created_at": "2026-08-28 02:24:29.861661+00:00",
"db_md5": "5bd2f7fa0ea09df195ea1dd780e3a628",
"submitter": "AFeinfield",
"github_login": "AFeinfield",
"run_id": "4da9213a-c40e-4740-8d3b-2cc1b7bae228",
"tier": "easy",
"dataset_id": "e1",
"status": "succ... | {
"score": {
"mean_loss": 2.0465433597564697,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04833333194255829
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.02018666267395,
"example_count": 100,
... |
2b5735cc-b49a-4c0d-ab69-85ae6b6e024c | easy | sapient-sapiens | 2026-08-20 17:19:55.012025+00:00 | succeeded | c8ac0e2bc843654b0c5a032f814997e31db4a05dd376c9f080511d36d2d02940 | 44,085 | null | """Persistent pair-indexed relational state around a categorical operator.
The recurrent answer state remains a bank of right-aligned categorical digit
slots. A separate continuous S-by-S grid retains every ordered digit-pair
cell across applications. Learned bilinear bindings and categorical modulus
context update ... | {
"id": "2b5735cc-b49a-4c0d-ab69-85ae6b6e024c",
"created_at": "2026-08-20 17:19:55.012025+00:00",
"db_md5": "081d657b94072d9174107b50bbc10a1c",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "3b195fcf-9a9f-4054-bfb2-363fafed327a",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 2.35282679315358,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.245436979516205,
"example_count": 600,
"exac... |
2b582c7a-374b-47da-997e-324019c40a46 | easy | briheet | 2026-08-07 03:51:32.671914+00:00 | succeeded | 51b26d1f12e1be5203babeab51158497e11b98785c46c011488be6fb41a2130f | 5,951 | null | """Transformer encoder with a small, recurrent learned latent state."""
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 = 32
... | {
"id": "2b582c7a-374b-47da-997e-324019c40a46",
"created_at": "2026-08-07 03:51:32.671914+00:00",
"db_md5": "198f5cda80f42c70765222013643427a",
"submitter": "Aizenn",
"github_login": "briheet",
"run_id": "8350ab20-7197-48a8-8666-ac464fb2c227",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",... | {
"score": {
"mean_loss": 2.3489257097244263,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05166666489094496
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.6745765209198,
"example_count": 100,
... |
2b6491ed-1f18-46a5-8517-60ded514f2dd | easy | blake-camp-surge | 2026-08-05 19:09:30.180919+00:00 | succeeded | 609bba7248e72cd8417039e4f83230e1d2d28ec2228938aa395d5e95d22e4334 | 10,518 | null | """Vector-state neural cellular automaton over attention edges."""
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_state,
)
TOKEN... | {
"id": "2b6491ed-1f18-46a5-8517-60ded514f2dd",
"created_at": "2026-08-05 19:09:30.180919+00:00",
"db_md5": "610972d54f50c4a0938dcc0f8ef37ced",
"submitter": "Blake Camp",
"github_login": "blake-camp-surge",
"run_id": "cf18bf67-3833-4f42-863a-c4a9a4fdc9ae",
"tier": "easy",
"dataset_id": "e3",
"status":... | {
"score": {
"mean_loss": 2.1346020396431857,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005625
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.133152095720855,
"example_count": 800,
... |
2b6a5d19-bf95-4e22-aaf4-ff932b5e427e | easy | oupadhyay | 2026-08-24 00:35:21.029534+00:00 | succeeded | 34377ca1828674623a4f4bbba1f04c829b2f8f9acea4c3df1dd5d0f448e44080 | 4,155 | null | """Pairwise digit-relation scan for the Easy tier."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import nn
from benchmark import ModelSpec,OptimizerSpec,OptimizerBundle,Submission,TokenLossBatch,assert_model_state
W,D,H=4,48,96
class C:
def __init__(s,vocab_size,max_seq_... | {
"id": "2b6a5d19-bf95-4e22-aaf4-ff932b5e427e",
"created_at": "2026-08-24 00:35:21.029534+00:00",
"db_md5": "a62caed1ad083b79d64e0b2e826e446f",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "7aedb75e-5776-47ff-8a19-552bc028663b",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.210372607941906,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004583333345750968
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1476309053566425,
"example_count": 600... |
2b6b0fde-0a80-46cb-a430-7d9f6376a055 | easy | 0Chris5R | 2026-08-07 09:50:01.932048+00:00 | succeeded | 25a190309838a06e73dae35cffa5ecc947527807f37da3c07cf78d5170022311 | 16,917 | null | """Learned convolutional-GRU over an aligned field-by-digit workspace.
The public field markers and decimal significance define only the layout of
the neural workspace. Every transition and output is learned; there is no
arithmetic routine, recurrence assumption, lookup table, or generated target.
"""
from __future__... | {
"id": "2b6b0fde-0a80-46cb-a430-7d9f6376a055",
"created_at": "2026-08-07 09:50:01.932048+00:00",
"db_md5": "2c734598c920311a1ac867127dc238c8",
"submitter": "Chris ",
"github_login": "0Chris5R",
"run_id": "41a0aa51-ff90-4fd9-ab8e-a9f5367efd49",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 7.615406660182555,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009583333345750967
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.779286934121307,
"example_count": 600,... |
2b775a08-9231-4711-86af-46779ebde333 | easy | khushidahi | 2026-08-21 22:28:37.189255+00:00 | succeeded | 4c9739766ea426413d2ead284095e245b75913ff80f7de721ab5f193bda5e49d | 39,821 | null | """R391: shallow-root / endpoint / shallow-root sandwich schedule.
R367's early short-operand phase produced the strongest seen shallow-root
Easy signal, while R376's late phase produced the strongest fresh shallow-root
Medium signal; each alone lost the other behavior at longer scale. R391 makes
that interaction fals... | {
"id": "2b775a08-9231-4711-86af-46779ebde333",
"created_at": "2026-08-21 22:28:37.189255+00:00",
"db_md5": "efe5f3c953dcaabf3d834d3b1a704272",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "d3271c4d-f630-4f4c-9869-75ed11c8ff25",
"tier": "easy",
"dataset_id": "e3",
"status": "succ... | {
"score": {
"mean_loss": 2.719834456721368,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007500000004656613
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.707177624062525,
"example_count": 800,... |
2b88b5b0-a649-404e-8eb8-e5e91b53c224 | easy | kseth | 2026-08-04 04:50:56.161929+00:00 | succeeded | a029a1aa92b9b0fe00962cb02a06e64b7a5824c3b3dbac65c1ff34ee43f02c2c | 7,127 | null | """Weight-tied recurrent Transformer baseline for One Layer Deeper.
This file is deliberately standalone: it only imports PyTorch and the public
``benchmark`` API, so it can be submitted directly to the official evaluator.
"""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch... | {
"id": "2b88b5b0-a649-404e-8eb8-e5e91b53c224",
"created_at": "2026-08-04 04:50:56.161929+00:00",
"db_md5": "8eea5881cb6c3e4ffb904db75b10b08e",
"submitter": "Karthik Sethuraman",
"github_login": "kseth",
"run_id": "00c7a76f-564c-4baa-9e9c-7286f4a145ff",
"tier": "easy",
"dataset_id": "e1",
"status": "s... | {
"score": {
"mean_loss": 6.050845623016357,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03500000108033419
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.336712837219238,
"example_count": 100,
... |
2b904535-3c7a-4e9b-8fbb-56c6d0b6fdae | easy | erdavis0 | 2026-08-22 02:23:59.024578+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": "2b904535-3c7a-4e9b-8fbb-56c6d0b6fdae",
"created_at": "2026-08-22 02:23:59.024578+00:00",
"db_md5": "464075811335119d26ad5014b80665e5",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "2072018b-32a6-4f19-b8af-f85c306904d3",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded",... | {
"score": {
"mean_loss": 3.919433355331421,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005624999990686774
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.8981120586395264,
"example_count": 800... |
2b9187fa-67ed-4aee-86bc-57c6c1ae6c10 | easy | yashkant | 2026-08-31 16:52:02.889731+00:00 | succeeded | 8aeadecab70131d8bd53b2dcce1a57b210bfc16ce50b36b983c2b4b75e15f801 | 104,475 | 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": "2b9187fa-67ed-4aee-86bc-57c6c1ae6c10",
"created_at": "2026-08-31 16:52:02.889731+00:00",
"db_md5": "4e9c65f4af368bd3681bb8695b99d2fb",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "bcd5c920-5ff4-479c-bf72-3494dc55dc78",
"tier": "easy",
"dataset_id": "e1",
"status": "succeed... | {
"score": {
"mean_loss": 4.552920579910278,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.019999999552965164
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.810990333557129,
"example_count": 100,... |
2b98dcf4-cd1a-46ed-868d-69fa4b7f5687 | easy | rteehas | 2026-08-19 09:34:18.616732+00:00 | succeeded | 67a121bc5064417c595c65f4606dd9dc4cbe2a449dbffb056559cf8932241f1f | 50,908 | 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": "2b98dcf4-cd1a-46ed-868d-69fa4b7f5687",
"created_at": "2026-08-19 09:34:18.616732+00:00",
"db_md5": "15a833c238504b9d53e3ecf1bc8d9e51",
"submitter": "Ryan Teehan",
"github_login": "rteehas",
"run_id": "289f0f33-33c2-456e-9e20-c0458db15d53",
"tier": "easy",
"dataset_id": "e2",
"status": "succee... | {
"score": {
"mean_loss": 2.833213448524475,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8332133293151855,
"example_count": 300,
"exac... |
2b9e1051-ef51-4028-b7a9-e89549313235 | easy | richardcepka | 2026-08-13 19:54:37.881110+00:00 | failed | 99db0d3da84d3ab8516ccbf6d744f2bfa71b6be166c29e031017f20dd6667b19 | 22,390 | null | """Recurrent digit-register model for One Layer Deeper.
A looped transformer whose recurrent state is a decimal digit register that is
re-quantised through a ten-entry codebook on every iteration. Nothing inside
the loop depends on the iteration index and no parameter is indexed by an
absolute slot position, so the e... | {
"id": "2b9e1051-ef51-4028-b7a9-e89549313235",
"created_at": "2026-08-13 19:54:37.881110+00:00",
"db_md5": "c115c878d2e09fff4c0bf635f0509f1b",
"submitter": "Richard Cepka",
"github_login": "richardcepka",
"run_id": "2b282494-4bbb-4a91-a60b-216f984be12b",
"tier": "easy",
"dataset_id": "e5",
"status": ... | null
|
2ba3c5a6-ee54-46f5-b117-4300060a17ba | easy | DDanlov | 2026-08-18 00:06:39.299668+00:00 | succeeded | 115d46ec00ec18546afcf1ba31d74ac788944592739c4e1a211477fa256dc823 | 23,102 | 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": "2ba3c5a6-ee54-46f5-b117-4300060a17ba",
"created_at": "2026-08-18 00:06:39.299668+00:00",
"db_md5": "ef8f2b4e45393e17f33e8d6b42c30e5d",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "ee5b6d6f-66d9-403f-87ac-2ca9ac0f3a5c",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 1.9827427733648642,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.033333333606521284
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.9421803648792095,
"example_count": 10... |
2baeb96d-521a-4a95-ba8a-c9712e99716f | easy | jordanrubin | 2026-08-03 21:02:10.091439+00:00 | succeeded | 09586f90ffa296e3bdbb25e6a0b059c7579344900d2f7d9eb12b0e902c4848f6 | 15,390 | null | """Bidirectional transformer, D=256 x 3 blocks, grokking-tuned AdamW.
Design notes:
- Inputs are padded to config.max_seq_len inside forward and logits sliced
back, so every training batch presents one static shape (compile/cudagraph
friendly, and uniform kernel shapes even in eager).
- Attention mask gets an iden... | {
"id": "2baeb96d-521a-4a95-ba8a-c9712e99716f",
"created_at": "2026-08-03 21:02:10.091439+00:00",
"db_md5": "ef7d67d872a65aa96e81ca53991ab27d",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "7806e045-7a22-477a-9b87-30c1ee456e07",
"tier": "easy",
"dataset_id": "e2",
"status": "s... | {
"score": {
"mean_loss": 2.9846639151307928,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.20937499442758659
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.685876167886626,
"example_count": 300,... |
2bb182dd-858f-4550-9956-e12d957cbe84 | easy | Ranoobaba | 2026-08-21 14:17:19.422133+00:00 | succeeded | ae49235ef72181017c6c2de58a2d2702d3633921926b25d38bc3f1e07bf2b53f | 6,667 | 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": "2bb182dd-858f-4550-9956-e12d957cbe84",
"created_at": "2026-08-21 14:17:19.422133+00:00",
"db_md5": "f2d534041dc915b4d57daa9ce8af5f86",
"submitter": "Syed Rayyan Ali",
"github_login": "Ranoobaba",
"run_id": "8236352a-0570-4c17-8e0a-eabdc5afe315",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 3.244180963520992,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03833333184321722
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.516923666000366,
"example_count": 100,
... |
2bb29635-90c2-4260-bf8b-2f5b14eedb43 | easy | alirezashirvani-jr | 2026-08-29 11:59:40.501984+00:00 | succeeded | d69d00a11e096efbd57498ece1064cb89bcfd2d20791931d41b7764a8df4fb6c | 32,203 | 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": "2bb29635-90c2-4260-bf8b-2f5b14eedb43",
"created_at": "2026-08-29 11:59:40.501984+00:00",
"db_md5": "752b3f410bfc2378b0a90c5626ec8257",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "1d940cbd-92b5-4223-8221-460596bf8505",
"tier": "easy",
"dataset_id": "e5",
... | {
"score": {
"mean_loss": 2.951478593152117,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03333333338300387
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.988334894180298,
"example_count": 600,
... |
2bb63753-99e3-4b83-aefc-f3a38cdf499d | easy | sapient-sapiens | 2026-08-19 11:37:41.511663+00:00 | succeeded | 74241046ae1f42c2fa06851819db68b2dbd7d5055cfd876eae9c1ab4e2f2a913 | 33,251 | 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": "2bb63753-99e3-4b83-aefc-f3a38cdf499d",
"created_at": "2026-08-19 11:37:41.511663+00:00",
"db_md5": "5eb894868492eb39a31b56404c76a71d",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "24c307f8-3772-4e9d-a995-54cc634529a3",
"tier": "easy",
"dataset_id": "e9",
"status": "su... | {
"score": {
"mean_loss": 5.088057994842529,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04555555712431669
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.012797832489014,
"example_count": 90,
... |
2bb73438-0e5a-47a3-85ae-c8a1e4990ca9 | easy | sapient-sapiens | 2026-08-17 19:46:49.904133+00:00 | succeeded | 73b24d001589dea7b46cf54d4671cee005cbfcf7bd92b0eb9133263887ecc018 | 28,660 | null | """v40 lineage: T-free operator, N memory, PonderNet mix, no T in the loss.
Field-local encoder, T-free core, countable halt, answer-slot readout.
N is seeded and restored every step. T is never parsed in forward or loss.
"""
from __future__ import annotations
import math
import time
import torch
import torch.nn.fu... | {
"id": "2bb73438-0e5a-47a3-85ae-c8a1e4990ca9",
"created_at": "2026-08-17 19:46:49.904133+00:00",
"db_md5": "ab0995ba57df7797ce934d12bf6df7e5",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "c0426356-3bd3-4edb-878f-9e57061d5b24",
"tier": "easy",
"dataset_id": "e4",
"status": "su... | {
"score": {
"mean_loss": 2.061567141750091,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007395833333333334
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.064910497960013,
"example_count": 1200... |
2bba0b15-ce2b-4b95-a045-ffdfac9ca6c3 | easy | oupadhyay | 2026-08-07 14:57:26.277234+00:00 | succeeded | 79694030b38024b641bdc729c71b2922a9a008a3bca5f9766fc017c0d920a3ca | 6,747 | null | """Easy E5 Neural-GPU with one homogeneous recurrent grid cell."""
import torch
from torch import nn
from benchmark import OptimizerBundle, Submission, assert_model_state
PAD, N, X, T, ANS, DIGIT = 0, 2, 3, 4, 5, 7
CHANNELS, MAX_PLACES, MAX_STEPS = 48, 16, 64
STATE_ELEMENTS = 64_330
class Config:
def __init__(s... | {
"id": "2bba0b15-ce2b-4b95-a045-ffdfac9ca6c3",
"created_at": "2026-08-07 14:57:26.277234+00:00",
"db_md5": "98192a45597baf6fe89b1f88d847e528",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "d3d207bc-68a2-417e-b841-3eaa30d8ecae",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.2775955893882203,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.002916666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.231324897218713,
"example_count": 600... |
2bba3b72-e71e-4767-9424-38cb94f13f73 | easy | viridale | 2026-08-13 11:29:42.495403+00:00 | succeeded | 24f3162f8be6f820a575af21bd12244a233e8168e367bdae4f75a12dbbbe325a | 23,723 | null | """Bit-canonical recurrent neural multiplier/reducer.
hypothesis: exact bit atoms plus a native learned bilinear product-column route and
learned N-conditioned associative carry/reduction scan make a small recurrence
edit constructively learnable instead of asking an unconstrained chart to invent
arithmeti... | {
"id": "2bba3b72-e71e-4767-9424-38cb94f13f73",
"created_at": "2026-08-13 11:29:42.495403+00:00",
"db_md5": "3ab791c19bd3b51ab53f046e3bd0137a",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "305a12a6-0705-44a5-ac09-84233d49acd6",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 16.587203979492188,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.011666666716337204
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 16.168136596679688,
"example_count": 10... |
2bbd72b4-016e-4c6e-ae10-a09095d20a7c | easy | Ranoobaba | 2026-08-16 09:06:45.566592+00:00 | succeeded | 1d644b8c0aac2cbda1d1336e25435a54cd7554a11c5d90cc1767e524d72f2074 | 5,723 | 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": "2bbd72b4-016e-4c6e-ae10-a09095d20a7c",
"created_at": "2026-08-16 09:06:45.566592+00:00",
"db_md5": "f8222fb387e6f8d20a07be54a2b9c442",
"submitter": "Syed Rayyan Ali",
"github_login": "Ranoobaba",
"run_id": "d7281f73-fe69-4048-980a-7e981d3d800b",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 7.2648771558319245,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.029999999776482583
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.568416595458984,
"example_count": 100... |
2bc9744f-9089-44a9-ab7a-c5796a8ba30d | easy | DDanlov | 2026-08-12 23:59:03.113913+00:00 | failed | 0e073fc3bf751cfba453f859d52a1c3a7ce6b4b8c203dd0e41959920ea689ad2 | 26,540 | null | from __future__ import annotations
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim.optimizer import Optimizer
try:
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
)
except ... | {
"id": "2bc9744f-9089-44a9-ab7a-c5796a8ba30d",
"created_at": "2026-08-12 23:59:03.113913+00:00",
"db_md5": "af401a0594a1953b6de234df2d54eb30",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "fbfd0a70-f5be-46df-8ba8-6bcbca77a231",
"tier": "easy",
"dataset_id": "e4",
"status": "failed",
... | null
|
2bd1e6d5-76fb-4c00-a449-b2d703bf6c02 | easy | DDanlov | 2026-08-21 20:04:07.321875+00:00 | failed | af97cb8f9bd15c3a9688b209276425101351b52c8ba61cef4562113128842639 | 25,971 | null | from __future__ import annotations
import math
import time
from typing import Any, Optional, Tuple, Dict
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Optimizer
try:
from benchmark.api import ModelSpec, OptimizerBundle, OptimizerSpec, Submission, assert_model_state
exce... | {
"id": "2bd1e6d5-76fb-4c00-a449-b2d703bf6c02",
"created_at": "2026-08-21 20:04:07.321875+00:00",
"db_md5": "23773e15d362a92c087118c0848256a3",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "68f2e6cb-807b-45e8-845d-26f2d1b1fa7c",
"tier": "easy",
"dataset_id": "e5",
"status": "failed",
... | null
|
2bd52087-1046-4a82-9cae-87374e24698f | easy | benjaminW2025 | 2026-08-18 00:30:27.413455+00:00 | succeeded | 61741a2431cfca70bd599f899f6dc649e03bc93b0fe5b862847db36ef1cc397f | 11,145 | 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,
assert_model_state,
)
D_MODEL = 256
NUM_HEADS = 4
... | {
"id": "2bd52087-1046-4a82-9cae-87374e24698f",
"created_at": "2026-08-18 00:30:27.413455+00:00",
"db_md5": "c6daaac25b0d550ef1ccbe9ea2b8335d",
"submitter": "benjawesome",
"github_login": "benjaminW2025",
"run_id": "0560dd3a-0b3a-45b8-9bc5-eb24394fb151",
"tier": "easy",
"dataset_id": "e5",
"status": "... | {
"score": {
"mean_loss": 2.234652792451052,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.011250000012417634
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2292453184256105,
"example_count": 600... |
2bd62422-e75d-4d83-8206-f6e5c185fbf8 | easy | adwhit | 2026-08-14 08:12:40.381042+00:00 | succeeded | c2996d589de9848f6aab76c6d44194e987db2461792834b829f992cf0499a924 | 17,687 | null | """GaussNet: additive-character recurrent model for One Layer Deeper.
Structure (all parameters learned from random init):
- Learned parse: cross-attention queries extract LSB-first digit registers
for x, copy-supervised from the prompt's own tokens via the auxiliary
channel (self-supervision, no task math).... | {
"id": "2bd62422-e75d-4d83-8206-f6e5c185fbf8",
"created_at": "2026-08-14 08:12:40.381042+00:00",
"db_md5": "959f3ce179cf0089ac39d8e1b150fd0a",
"submitter": "adwhit",
"github_login": "adwhit",
"run_id": "6def05cf-db50-41d2-aeef-698a07a1f4ae",
"tier": "easy",
"dataset_id": "e2",
"status": "succeeded",
... | {
"score": {
"mean_loss": 2.266539454460144,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.012916667386889458
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2618799209594727,
"example_count": 300... |
2becfef2-fd65-4c2b-bac8-dadbd113cace | easy | khushidahi | 2026-08-05 23:08:01.101769+00:00 | succeeded | 76e194987660fd964dc86ad7c5b660ed86b5e7db8fa70408ae44d6dfa6c634b6 | 25,718 | null | """R16 hard-transfer hybrid experiments for One Layer Deeper.
Motivated by complementary Hard behavior:
- the generic sequence recurrent model retained seen-N first-rung progress;
- the structured decimal-tape model retained OOD-N first-rung progress.
The dual variants co-train both learned representations, supervise... | {
"id": "2becfef2-fd65-4c2b-bac8-dadbd113cace",
"created_at": "2026-08-05 23:08:01.101769+00:00",
"db_md5": "d51b46b55c7a39d2ba4e500393a3cc73",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "4d87c8ed-2d26-44b4-b85e-cf82af54c94d",
"tier": "easy",
"dataset_id": "e4",
"status": "succ... | {
"score": {
"mean_loss": 2.0720999017553137,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004375000012417634
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.0710245658939135,
"example_count": 12... |
2bf525aa-1667-4acf-b68f-fc968b4f2508 | easy | karanganesan | 2026-08-06 03:35:19.599577+00:00 | succeeded | a584a26f8ec01901c8f1b277979f99723b0a970f2b1d206ca84d1095b8f3bb2e | 23,811 | 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": "2bf525aa-1667-4acf-b68f-fc968b4f2508",
"created_at": "2026-08-06 03:35:19.599577+00:00",
"db_md5": "3b9739c128f2ad545cdb73d36b935183",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "409efa64-0af5-4723-b2fe-e84bbd2c60fe",
"tier": "easy",
"dataset_id": "e5",
"status": ... | {
"score": {
"mean_loss": 5.369891405105591,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0054166666232049465
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.160238265991211,
"example_count": 600... |
2bfe8ade-9cfc-4a60-998c-1739d4659d31 | easy | ehonig | 2026-08-21 01:36:22.208621+00:00 | succeeded | a8101c7bf20f5fb52423225be6359acd59e36c1d9a8a08942c855acf40ae70ac | 31,882 | 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": "2bfe8ade-9cfc-4a60-998c-1739d4659d31",
"created_at": "2026-08-21 01:36:22.208621+00:00",
"db_md5": "75d8fa2efbdc026e18b4eec8692e50b8",
"submitter": "Edouardo Honig",
"github_login": "ehonig",
"run_id": "b6874020-5a37-42c9-b8c6-3fc44f8c36ea",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.2035625454139067,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.001666666679084301
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.213408946990967,
"example_count": 600... |
2c013312-6a07-4227-852e-65e83a1358ce | easy | sapient-sapiens | 2026-08-16 17:12:10.394761+00:00 | succeeded | de81b9daa3987f0cb073f020970e622a40641a52551492152823f0d3b2b58086 | 22,022 | null | """E6 generalization candidate: width-1024, d_h=8, 6-block ACT64.
Architecture changes relative to the hosted ACT max-4 source:
- selectable recurrence depths are 1 through 64;
- canonical cumulative-mass halting genuinely removes halted examples from
later recurrent applications;
- the T-field representation prese... | {
"id": "2c013312-6a07-4227-852e-65e83a1358ce",
"created_at": "2026-08-16 17:12:10.394761+00:00",
"db_md5": "6fe807fb286c2dc1a481d60a070535e5",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "6a86e074-7b25-4cc2-961b-31d7f7955403",
"tier": "easy",
"dataset_id": "e6",
"status": "su... | {
"score": {
"mean_loss": 6.052328998095369,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.136936945286957
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.238681793212891,
"example_count": 60,
... |
2c116577-cebe-485d-adbf-f284186e5253 | easy | christopherhwood | 2026-08-07 05:18:30.402178+00:00 | succeeded | 1642a38382429ed8cf111c10fe7fc73c1609694292786d3459984ace9e6c4b0c | 61,174 | null | """Learned-ISA tape machine: population-as-parameters program posterior.
A generic two-loop register machine (FROZEN ISA SPEC v1.1, two-address) whose
entire program -- per-slot routing, operands, finite-state-transducer tables,
head loops, commit predicates -- is a set of categorical parameters, learned
from endpoint... | {
"id": "2c116577-cebe-485d-adbf-f284186e5253",
"created_at": "2026-08-07 05:18:30.402178+00:00",
"db_md5": "b39054c5e1aef5b0a18bc226abb86676",
"submitter": "Chris Wood",
"github_login": "christopherhwood",
"run_id": "295fc778-6ede-4f4e-bfbf-4c1d1881908f",
"tier": "easy",
"dataset_id": "e1",
"status":... | {
"score": {
"mean_loss": 15.11289119720459,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 15.462688446044922,
"example_count": 100,
"exac... |
2c12d981-5b08-487f-9b87-0e722fbb6edd | easy | velocizapkar | 2026-08-13 10:06:21.484551+00:00 | succeeded | e10ebfdf1cde2148a53354e3a77e0ef5068d674264ec074743a566469b3acd34 | 13,370 | null | """Stabilized residual-free E-SPA/DKS Transformer control.
This preserves the non-causal E-SPA construction of He et al. (ICLR 2023):
each untied attention layer is initialized to L_out L_in^-1, where L is the
symmetric positive square root of a less rank-collapsed exponential position
kernel. Q/K retain the near-zero... | {
"id": "2c12d981-5b08-487f-9b87-0e722fbb6edd",
"created_at": "2026-08-13 10:06:21.484551+00:00",
"db_md5": "cef4f5c12cf0df0a6170818493dc08a8",
"submitter": "Aakanksh Zarapkar",
"github_login": "velocizapkar",
"run_id": "202f2476-1b20-4293-a8e8-3cf0e059ab9d",
"tier": "easy",
"dataset_id": "e2",
"statu... | {
"score": {
"mean_loss": 2.2654157876968384,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.010000000474974513
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.28839111328125,
"example_count": 300,... |
2c1be9cd-cc62-4ead-8b9d-eef63c1051a8 | easy | oupadhyay | 2026-08-27 08:17:38.224420+00:00 | succeeded | ba0f593cadf4304540a82340085f20d2d42f5addc245f1645a8f0fb561c64ff3 | 3,720 | null | """Persistent globally mixed polynomial recurrence over a whole-number latent state."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import nn
from benchmark import OptimizerBundle,Submission,TokenLossBatch,assert_model_state
W,D,R=4,32,128
class C:
def __init__(s,vocab_si... | {
"id": "2c1be9cd-cc62-4ead-8b9d-eef63c1051a8",
"created_at": "2026-08-27 08:17:38.224420+00:00",
"db_md5": "a676523432d4022873f4617cbadbfee9",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "aa43ac9c-52a8-449b-b560-1c80d0fd5d95",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 5.207215717739974,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005416666691501935
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.36001451774563,
"example_count": 600,
... |
2c1e14b4-d132-45a2-8234-f140a666b527 | easy | sapient-sapiens | 2026-08-17 11:59:27.089215+00:00 | succeeded | 2c81a075b20053268cec22c72deface05595d31c759ee2bcd2cc116301c4e4dd | 17,309 | null | """Parameterized fixed-K recurrent candidate for the additive sweep.
Knobs at the top are the only intended experimental variables. T remains an
ordinary prompt field and is never parsed into a loop count.
"""
from __future__ import annotations
import math
import time
import torch
import torch.nn.functional as F
fr... | {
"id": "2c1e14b4-d132-45a2-8234-f140a666b527",
"created_at": "2026-08-17 11:59:27.089215+00:00",
"db_md5": "b9ccb0a726cb5c7eb0fb4e1005c7bcb7",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "3ba169c4-6636-46d3-98a0-2e15127342ef",
"tier": "easy",
"dataset_id": "e3",
"status": "su... | {
"score": {
"mean_loss": 3.362554336507399,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.3749095725571854,
"example_count": 800,
"ex... |
2c314110-e44c-4d88-b4ca-920291ed90a2 | easy | DDanlov | 2026-08-29 14:52:13.083209+00:00 | succeeded | 62f1bcf5c36c89bd56eaed81f749b302fe103db99c27efd974e00f58e6fbea59 | 28,657 | null | """
submission_latent_kronecker.py
2nd-Order Latent Kronecker Curvature-Preconditioned Continuous Hopfield Deliberator.
Features full GPU-inverted Kronecker factor preconditioning (Token Covariance L_seq x L_seq, Channel Covariance D x D),
trace damping, positive eigenvalue masking, and Preconditioned Heavy-Ball latent... | {
"id": "2c314110-e44c-4d88-b4ca-920291ed90a2",
"created_at": "2026-08-29 14:52:13.083209+00:00",
"db_md5": "a121ee075f188c504ce77256c1c8f9c2",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "62231912-df30-436c-a400-ab8017b405fd",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.7042305424465716,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.011250000024835269
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.706457528832782,
"example_count": 600... |
2c3d04d8-6f2a-4c88-aa50-df0c2b11ea01 | easy | lpbb | 2026-08-19 09:36:08.498703+00:00 | succeeded | f649bd901efa894492c08a7a18ad1cbd816f401ef1f37ca1e2fb346cfecb7e67 | 8,044 | null | """S0 + RoPE + abacus embeddings: the narrow-deep rope variant (36 untied
layers, positions entering only through rotary q/k so attention sees relative
offsets), plus a per-digit significance tag (units=1, tens=2, ...) embedded
additively, with random per-row offsets at train time so tag rows beyond the
trained digit c... | {
"id": "2c3d04d8-6f2a-4c88-aa50-df0c2b11ea01",
"created_at": "2026-08-19 09:36:08.498703+00:00",
"db_md5": "9ebd61e5dddfb29c9d46d8a9cddfe013",
"submitter": "Lpbb",
"github_login": "lpbb",
"run_id": "bb5398d8-c46c-4599-8977-eb4fab23779b",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
"s... | {
"score": {
"mean_loss": 3.836822986602783,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05166666489094496
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.289954662322998,
"example_count": 100,
... |
2c4b0d31-ffdc-4797-9806-2da04ff35519 | easy | YM2132 | 2026-08-05 06:53:59.280859+00:00 | succeeded | 95bd43c77c18fc28497828c1e0f9b823e72b46bd960a23b3d74a93b737023f0b | 8,057 | null | """Weight-tied looped transformer with per-row adaptive depth.
v1 for the squaring-mod benchmark: one encoder block builds an initial state
from the embedded prompt, then a single weight-tied core block is applied
k = K_BASE + K_PER_T * T times per row, where T is parsed from the prompt's
digit tokens. Each row's sta... | {
"id": "2c4b0d31-ffdc-4797-9806-2da04ff35519",
"created_at": "2026-08-05 06:53:59.280859+00:00",
"db_md5": "dfe58049bffc8dfefaabeecdd2520ab0",
"submitter": "Yusuf Mohammad",
"github_login": "YM2132",
"run_id": "29a52ed0-075a-4e49-b704-2e706c871e35",
"tier": "easy",
"dataset_id": "e1",
"status": "succ... | {
"score": {
"mean_loss": 6.512671947479248,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.016666666604578495
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.774292945861816,
"example_count": 100,... |
2c4f9eb2-7410-4892-9ddf-4dfab17e675b | easy | shreyash-chonkie | 2026-08-24 23:34:00.133468+00:00 | succeeded | 6a21a00f92da6a591f6afb7417ebbf85c4879620e795a6332379c8493b935d7b | 10,468 | null | """Sequential residual actors with localized improvement feedback."""
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,... | {
"id": "2c4f9eb2-7410-4892-9ddf-4dfab17e675b",
"created_at": "2026-08-24 23:34:00.133468+00:00",
"db_md5": "a55fa90d118fcdaee4e7707828e43674",
"submitter": "Shreyash",
"github_login": "shreyash-chonkie",
"run_id": "c9086bdf-4ac6-4dc7-ab46-8c11a110e674",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 7.1946423053741455,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.014999999664723873
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.330679416656494,
"example_count": 100... |
2c532ff3-058c-45d0-9e58-353315a32940 | easy | sapient-sapiens | 2026-08-20 20:00:30.723046+00:00 | succeeded | cceef680a3cc6ae3f8b41fc2966b9734c862bfea9c4cd10d820068f6fdd0ca96 | 38,475 | null | """Hierarchical learned digit-mixing recurrence with an answer-only 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. T... | {
"id": "2c532ff3-058c-45d0-9e58-353315a32940",
"created_at": "2026-08-20 20:00:30.723046+00:00",
"db_md5": "f1efdf842c7a7b4a87cf560f07a92d55",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "3ef10940-6984-46ee-91e3-98813b7ca7dc",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 2.1645346720776804,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007916666691501935
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1686683044572583,
"example_count": 60... |
2c572598-29e3-46c9-9fc8-3eba0b188a08 | easy | yashkant | 2026-08-17 00:05:06.234309+00:00 | succeeded | d29f9668ee99bafa7f0ec9c57909e7242e513104db06c3c407bf41343bdf0ef6 | 52,170 | null | """Round2152 SAM4 active-Dykstra antisymmetric twin expansion.
The complete SAM-every-four Round2098 parent remains active. One byte-identical
copy of its compact recurrent Dykstra correction is appended without advancing
the parent's RNG stream. If the inherited branch is A and the clone is B, the
attachment become... | {
"id": "2c572598-29e3-46c9-9fc8-3eba0b188a08",
"created_at": "2026-08-17 00:05:06.234309+00:00",
"db_md5": "4d3be06210b135bae6e6446869e50507",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "47a40e64-45a3-46a7-a9c0-e5aabc0e9543",
"tier": "easy",
"dataset_id": "e1",
"status": "succeed... | {
"score": {
"mean_loss": 2.803783893585205,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05166666768491268
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8453681468963623,
"example_count": 100,... |
2c5e8357-9753-4707-b736-4ba2ba678f2b | easy | khushidahi | 2026-08-02 06:24:30.745171+00:00 | succeeded | d00bbc64f820d6804a006ac58901548f1445eec55891ee3e9ea92388f58105fb | 8,466 | null | """Prelude -> recalled shared core -> coda model for One Layer Deeper.
R2 architecture reset:
- use a real one-block prelude and one-block coda;
- apply one shared recurrent Transformer core four times;
- inject immutable prelude context at every recurrent update;
- use sandwich RMS normalization for stable repeated a... | {
"id": "2c5e8357-9753-4707-b736-4ba2ba678f2b",
"created_at": "2026-08-02 06:24:30.745171+00:00",
"db_md5": "3d42389dfcecc4b2976228efce0dc7eb",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "90d51718-5c08-4515-852f-df83a61dc189",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.1638060808181763,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004166666767559946
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1708147525787354,
"example_count": 60... |
2c628ee0-b539-4ce2-9a11-5ed3f9d3e418 | easy | TheYkk | 2026-08-04 12:40:10.646316+00:00 | succeeded | 05de0c46a2a6b9bbb268fc55c24888fd961d07a37ceae73e1ac3e1e5b7eb10a9 | 17,824 | null | """Encoder / iterated-operator / positional-decoder composition model.
Optimizer ablation
------------------
The architecture, loss, batch constants, and batch-reuse policy are byte-for-byte
the same as ``narrow_compose_v9``; the *only* change is the optimizer. Instead of
a single AdamW over every tensor, this submis... | {
"id": "2c628ee0-b539-4ce2-9a11-5ed3f9d3e418",
"created_at": "2026-08-04 12:40:10.646316+00:00",
"db_md5": "034762e6b87bfe6f7a050247546681a5",
"submitter": "Kaan Karakaya",
"github_login": "TheYkk",
"run_id": "c6c84aaf-81d9-4846-8af3-dd530137c749",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 9.065346717834473,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.06333333067595959
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.843230247497559,
"example_count": 100,
... |
2c6b31d6-e017-47fc-a32e-0ef2154c6bb1 | easy | alirezashirvani-jr | 2026-08-29 05:12:27.886593+00:00 | succeeded | 8b730148f27170686b596bd5fe974dbf358905a4b5346438bfeca2b81edcd2af | 36,975 | null | """Adaptive learned discrete recurrent fabric for One Layer Deeper.
A gradient-trained, operation-free recurrent computational machine. The
public prompt grammar only routes literal digit tokens into canonical,
tail-aligned categorical registers. A reusable learned transition circuit
is applied serially, the public ... | {
"id": "2c6b31d6-e017-47fc-a32e-0ef2154c6bb1",
"created_at": "2026-08-29 05:12:27.886593+00:00",
"db_md5": "d21e694a3484c8f7e5b389d51abb8c59",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "44520d94-15fe-4e43-abdc-9c58f0119e4a",
"tier": "easy",
"dataset_id": "e5",
... | {
"score": {
"mean_loss": 2.2308616698929318,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004166666686845323
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2211174964904785,
"example_count": 60... |
2c6e24f2-d115-4942-ab46-9f305b43f4b9 | easy | chad-atexpedient | 2026-08-08 06:39:54.619542+00:00 | succeeded | 3fb09b40dbe9de65469474cc646c3d20b7f113929f4d940063258783c50cf7fc | 15,003 | null | """S01: E04 with same-example short/long rollout distillation."""
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,
)
... | {
"id": "2c6e24f2-d115-4942-ab46-9f305b43f4b9",
"created_at": "2026-08-08 06:39:54.619542+00:00",
"db_md5": "6f36da855891143bc62873903577eb35",
"submitter": "chad-atexpedient",
"github_login": "chad-atexpedient",
"run_id": "b1ba7aab-b7ad-49a8-bcd0-261ea27431d8",
"tier": "easy",
"dataset_id": "e5",
"st... | {
"score": {
"mean_loss": 2.3012646051631744,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.012500000012417634
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.350969960336728,
"example_count": 600... |
2c6ea542-008f-49f5-9bc2-85c26cce4001 | easy | DDanlov | 2026-08-13 19:48:20.903028+00:00 | succeeded | 939f58d818c3f953053dfcbc068030b31142e50f323a87b6d7f84ab5df0f2dd5 | 26,904 | null | from __future__ import annotations
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim.optimizer import Optimizer
try:
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
)
except ... | {
"id": "2c6ea542-008f-49f5-9bc2-85c26cce4001",
"created_at": "2026-08-13 19:48:20.903028+00:00",
"db_md5": "e3de2d9e8ccead933e2b8372c4ddba6b",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "fd528f84-02c5-494d-8c20-c402551759d8",
"tier": "easy",
"dataset_id": "e4",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.262225118913082,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008541666666666666
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.26572224591345,
"example_count": 1200,... |
2c786222-16c8-4b16-9ade-57638115c586 | easy | 0Chris5R | 2026-08-07 12:23:53.223837+00:00 | succeeded | 46d9c5df786be5f745e609908387b148fa74f5772fbe177e42e6f77979dcc116 | 20,841 | 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": "2c786222-16c8-4b16-9ade-57638115c586",
"created_at": "2026-08-07 12:23:53.223837+00:00",
"db_md5": "418fb965ebdffa6048badd271b5272f7",
"submitter": "Chris ",
"github_login": "0Chris5R",
"run_id": "a9bc5f08-7d5d-402c-be85-74b03893d950",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.7061406723360015,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.032916666666666664
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.6733783627839367,
"example_count": 60... |
2c78e341-1228-444f-8767-d7f623a8d45e | easy | jordanrubin | 2026-08-11 01:41:11.186112+00:00 | succeeded | aa0c446a01b1c50a3739ac906b125fbe05d6b1cd9cf0a23d1a564e062285e8d6 | 21,670 | 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": "2c78e341-1228-444f-8767-d7f623a8d45e",
"created_at": "2026-08-11 01:41:11.186112+00:00",
"db_md5": "344e901468d7fc4aff5006fca563aa6e",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "698619ce-bebc-48d5-a1b9-377672f77e3b",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.566089617847476,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0037499999161809683
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.160022048404933,
"example_count": 600... |
2c7bdfcf-144f-46ca-a438-0d30c9f89e60 | easy | oupadhyay | 2026-08-11 08:54:56.325057+00:00 | succeeded | 337ed0cbd8652f29f1cbc270924e460b57b47d8e90b55d6f8927a036175036ea | 7,707 | 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... | {
"id": "2c7bdfcf-144f-46ca-a438-0d30c9f89e60",
"created_at": "2026-08-11 08:54:56.325057+00:00",
"db_md5": "d3e47ee4fe81ebed4cfc36b1d1d8b590",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "18ff8d86-6893-4638-9945-5be561c243b0",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.2019131847300244,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0037500000000000003
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.155863962900478,
"example_count": 60... |
2c7bf05f-a76f-4213-a875-9573dbf03ab2 | easy | johnowhitaker | 2026-08-03 21:32:03.824974+00:00 | succeeded | 2133a04bd1dec91e9cc609e79a885a1d75e3b03cc09943b060c9e65c2fe1425d | 11,600 | null | """Interleaved residue-state recurrent Transformer.
Four possible residue positions at the right edge of each prompt receive five
internal spacing slots apiece. A shared Transformer body is rerun six times.
After each pass, the state at the current prediction position is inserted into
the next spacing slot, creating ... | {
"id": "2c7bf05f-a76f-4213-a875-9573dbf03ab2",
"created_at": "2026-08-03 21:32:03.824974+00:00",
"db_md5": "c81b8a9388a5b18520bc4eeb3b5586fd",
"submitter": "Jonathan Whitaker",
"github_login": "johnowhitaker",
"run_id": "4c60c55a-fef1-497b-9279-e27147d15727",
"tier": "easy",
"dataset_id": "e1",
"stat... | {
"score": {
"mean_loss": 2.3189460039138794,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.023333333432674408
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.7073490619659424,
"example_count": 10... |
2c846e6c-fc87-4148-b59d-c136d851f65f | easy | viridale | 2026-08-28 22:35:07.167561+00:00 | succeeded | 0743eb9b689efbd6adc934d6667143be416bc3a88d7b2ad5d5fa96b9fce05a95 | 63,851 | null | """Sparse lifted-quadratic action with coherent final-answer evidence.
hypothesis: H1 has rejected a fixed small visible-state quadratic, while winner evidence
still favors a bilinear law. A source-small phase, lag, original-input coupling, or
terminal wrapper becomes a sparse quadratic edit after lifting stat... | {
"id": "2c846e6c-fc87-4148-b59d-c136d851f65f",
"created_at": "2026-08-28 22:35:07.167561+00:00",
"db_md5": "cd6cfd310e68ce9b7d3b9089c5e2dd02",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "41ac8518-4590-4488-93a5-7bf0c07f8017",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 0.0,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.0,
"example_count": 800,
"exact_accuracy": 1.0,
"... |
2c877fde-5b3b-4dcf-94a8-ec071c8fed74 | easy | alirezashirvani-jr | 2026-08-30 21:28:07.477937+00:00 | succeeded | 7f01b8a4c66f7682440fae7e83a7c3cf1cea0e9a5ca575aa504804576e005fb8 | 22,897 | 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_WIDTH = 128
ATTENTION_HEADS =... | {
"id": "2c877fde-5b3b-4dcf-94a8-ec071c8fed74",
"created_at": "2026-08-30 21:28:07.477937+00:00",
"db_md5": "6c457d9f6d633abd15c3c281ba921761",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "59987837-867e-4f19-bd84-66bb29e4e906",
"tier": "easy",
"dataset_id": "e3",
... | {
"score": {
"mean_loss": 2.29495632648468,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0037499999161809683
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.297003984451294,
"example_count": 800,... |
2c92e506-4724-4416-b272-7ebea2f9ad10 | easy | erdavis0 | 2026-08-26 04:06:14.301966+00:00 | succeeded | 6fe6fdf46bcdc9c76ffef64fbdc6b302f2262f8879ac4d7daea4e62dccf8b390 | 30,354 | null | """Latent-role continuous REV C5: hybrid token CE and sequence mixture.
Public delimiter syntax only routes opaque token categories into distinct,
right-aligned N, X, and T tapes. Learned local cells compile separate
interaction, carry, and modulus-conditioned workspaces. N is isolated from
interaction and carry and... | {
"id": "2c92e506-4724-4416-b272-7ebea2f9ad10",
"created_at": "2026-08-26 04:06:14.301966+00:00",
"db_md5": "14ff631f83062f36483f24b1bd94c4c9",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "a428334e-d722-4c8d-9cae-cce3335824b4",
"tier": "easy",
"dataset_id": "e2",
"status": "succeeded",... | {
"score": {
"mean_loss": 5.034527540206909,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03395833447575569
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.1980414390563965,
"example_count": 300,... |
2c951d19-f54f-4e41-bf10-fe352b795c33 | easy | jordanrubin | 2026-08-07 01:10:29.780877+00:00 | succeeded | 0eb249f609a0101d70cb07bbf9b0f2aa4cc7c5f7465d3e5e29bc4c3e68abecc6 | 14,786 | 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": "2c951d19-f54f-4e41-bf10-fe352b795c33",
"created_at": "2026-08-07 01:10:29.780877+00:00",
"db_md5": "497d08edf33fa52921753c6fa8942084",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "42190dbf-dedb-43ec-81ad-2ef62c2ee2fa",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 3.9113000959962507,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008333333147068819
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.9777883261843114,
"example_count": 60... |
2c9845a3-9c27-4fd3-9032-5cd126969a11 | easy | khushidahi | 2026-08-02 06:22:28.336047+00:00 | succeeded | 641b0ee10051d247a9154b7d13b89c4c538f18742dec345edeb3121b70cef6b5 | 8,485 | null | """Prelude -> recalled shared core -> coda model for One Layer Deeper.
R2 architecture reset:
- use a real one-block prelude and one-block coda;
- apply one shared recurrent Transformer core four times;
- inject immutable prelude context at every recurrent update;
- use sandwich RMS normalization for stable repeated a... | {
"id": "2c9845a3-9c27-4fd3-9032-5cd126969a11",
"created_at": "2026-08-02 06:22:28.336047+00:00",
"db_md5": "b630891d79fb2d1be2b060b8aa144e01",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "7dc72d6d-7dfb-406a-bce0-8a140380e1d1",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.1773288249969482,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008333333535119891
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.182541847229004,
"example_count": 600... |
2c9d4a88-3ef2-440a-a1f0-798333809a6c | easy | DDanlov | 2026-08-28 21:27:45.571389+00:00 | succeeded | c6c4f6552bb6e4cde025ca0dbcc200471c873717a02cbf276ffc1b3f928f87e3 | 15,439 | null | """
Trial 01: Model A: D=256, H=4, Trec=32, lr=0.0035
"""
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from dataclasses import dataclass
from typing import Optional, Tuple, Dict, Any, List, Union
try:
from benchmark.api import ModelSpec, OptimizerBundle, OptimizerSpec, Submission... | {
"id": "2c9d4a88-3ef2-440a-a1f0-798333809a6c",
"created_at": "2026-08-28 21:27:45.571389+00:00",
"db_md5": "8a30f500c2a7037d7444d15712c1c59a",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "611191ae-382d-42e2-a80f-2031d73eb34c",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.277832381124485,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008333333333333333
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.28485460612806,
"example_count": 600,
... |
2c9dbb03-10e2-48da-b96d-2539c093d4df | easy | chad-atexpedient | 2026-08-13 01:57:17.943128+00:00 | succeeded | 24df2c9e24a9670aa39902becd4aa3b000a64fe407a1aefc4efac0018a5a9346 | 13,865 | null | """X17D: pair behavior with ordinal loss."""
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,
)
LA... | {
"id": "2c9dbb03-10e2-48da-b96d-2539c093d4df",
"created_at": "2026-08-13 01:57:17.943128+00:00",
"db_md5": "ea4414d92b99a89367fbe2ce3f9d06bf",
"submitter": "chad-atexpedient",
"github_login": "chad-atexpedient",
"run_id": "342f4806-5be7-4afb-a523-f81082951a0a",
"tier": "easy",
"dataset_id": "e3",
"st... | {
"score": {
"mean_loss": 2.1648099812802735,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.01312500006519258
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1683727447155094,
"example_count": 800... |
2ca26be2-67ed-4a32-b475-b57f42b523b9 | easy | chuk-uzowihe | 2026-08-06 23:45:43.900585+00:00 | succeeded | ab4becf976aa6ac2b96eb336cf34c2445278e99cf6ba2ee7a5e5d030f271bc8f | 33,676 | null | """One Layer Deeper submission: FPRM over a digit lattice with a Fourier T latent.
Structure (v4)
--------------
1. N and x are *digit slots*: each digit is gathered into a dense per-number
grid (column j = place value 10^j, short rows left-padded with explicit
zero digits) and embedded as digit token + learned ... | {
"id": "2ca26be2-67ed-4a32-b475-b57f42b523b9",
"created_at": "2026-08-06 23:45:43.900585+00:00",
"db_md5": "e37330168f88d1b4011ceb538e11abf6",
"submitter": "Chuk Uzowihe",
"github_login": "chuk-uzowihe",
"run_id": "4fe14021-e4a6-450e-8c4d-0faf7dfe3337",
"tier": "easy",
"dataset_id": "e4",
"status": "... | {
"score": {
"mean_loss": 9.214919120535704,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006041666691501935
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.305688433939387,
"example_count": 1200... |
2caa36f2-c91a-4bf4-833d-a2c2702e9f18 | easy | KaustubhKumar05 | 2026-08-27 02:56:05.745333+00:00 | succeeded | 636758f777a2fb1d57a71238e31cd3036c039385ec13c3e14fe6e3bfb1a67d1e | 12,222 | null | """abacus_modsquare — candidate B: digit-ALIGNED arithmetic transformer.
FINDINGS §32-§38. Every recurrent/operator architecture we have tried encodes
the whole integer into one dense latent and then iterates it. §35 showed that
family scrambles the state past the trained depth, and §37 showed it scores
0.000% on 9,00... | {
"id": "2caa36f2-c91a-4bf4-833d-a2c2702e9f18",
"created_at": "2026-08-27 02:56:05.745333+00:00",
"db_md5": "9079215d75ad77148ade1b09cd89c571",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "db4e13f9-dd89-43e3-8d67-a7ad24360d3f",
"tier": "easy",
"dataset_id": "e3",
"status": "succee... | {
"score": {
"mean_loss": 2.5131873543894097,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00375
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.515554110744093,
"example_count": 800,
"... |
2caaea09-6aea-48c5-ba27-80577120ef2c | easy | erdavis0 | 2026-08-23 11:50:43.133577+00:00 | succeeded | a088f356590cfba5ad21d2d47f6b52ef6ea3fd6c700f05b01303b3ce92032250 | 23,907 | null | """Reversible augmented-state transducer over aligned learned digit tapes.
Public delimiter syntax only routes opaque token categories into distinct,
right-aligned N, X, and T tapes. Learned local cells compile separate
interaction, carry, and modulus-conditioned workspaces. N is isolated from
interaction and carry ... | {
"id": "2caaea09-6aea-48c5-ba27-80577120ef2c",
"created_at": "2026-08-23 11:50:43.133577+00:00",
"db_md5": "3e958839de80efbf59a22b9e0b2a18fd",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "12388153-0bcb-4262-b376-e20bfc4e0b37",
"tier": "easy",
"dataset_id": "e10",
"status": "succeeded"... | {
"score": {
"mean_loss": 5.853435754776001,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05705050611868501
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.731305122375488,
"example_count": 125,
... |
2cabdc06-d5e6-49d7-9105-126b65878769 | easy | sapient-sapiens | 2026-08-15 08:36:48.885631+00:00 | succeeded | 81bd484faaa911113c4531bbfe6ba68cc26928ef4bda463eeaae6c311b1d169a | 13,985 | null | """Narrow/deep looped Transformer with optional latent workspace tokens.
Research candidate for sampled-modulus fixed-T=2. The encoder builds immutable
prompt memory. A stack of learned operator blocks updates a mutable latent
stream, and the whole stack is reused for each task recurrence. Optional blank
workspace ... | {
"id": "2cabdc06-d5e6-49d7-9105-126b65878769",
"created_at": "2026-08-15 08:36:48.885631+00:00",
"db_md5": "445cef06f96c03c148fe5edffd94ee00",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "0e2802ac-e19e-45b4-a06c-416c73c5d559",
"tier": "easy",
"dataset_id": "e4",
"status": "su... | {
"score": {
"mean_loss": 2.0638278138691284,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0028125
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.066425144523902,
"example_count": 1200,
... |
2cacfd6d-6e23-46ac-a899-9910021c42ee | easy | MarcoF1 | 2026-08-13 08:27:47.514830+00:00 | succeeded | 21c61eade1a0dcdabc86d3b928dd04577b15591a9f754b3b43451f8601c2bd6c | 14,690 | null | """Looped variant of the v9 stack: prelude, tied 2-sublayer core run T+2
times with per-example halting, postlude. Same residue/cyclic-T features
and Muon/perp-grad/EMA/fp64-loss stack as v9_clock; no fixed-point aux
readouts (they reward the wrong invariance for a time-homogeneous map).
"""
from __future__ import ann... | {
"id": "2cacfd6d-6e23-46ac-a899-9910021c42ee",
"created_at": "2026-08-13 08:27:47.514830+00:00",
"db_md5": "59bd17b6ee3339f04bda7e965f358779",
"submitter": "Marco Fleming",
"github_login": "MarcoF1",
"run_id": "faaf2799-9c84-4d13-ac01-ca0f97d3bbe0",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.4420712123753394,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004166666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.4329893572982653,
"example_count": 60... |
2cb34df9-1362-4879-b357-f9bb8c4c6bf5 | easy | viridale | 2026-08-24 03:47:58.883807+00:00 | succeeded | 0c45115e2fa4d194504e7877ebcf5ad038e996e44e6aab608f349340e60bb964 | 232,638 | null | """NV095: cross-subsystem learned-program joins.
hypothesis: v816 exhausts its word-program schedule but can discard two learned
partial programs that live in different subsystems because no bank ever
executes their compatible union. Archive evidence-half specialists from
the complete inherited search and... | {
"id": "2cb34df9-1362-4879-b357-f9bb8c4c6bf5",
"created_at": "2026-08-24 03:47:58.883807+00:00",
"db_md5": "b28d3713aa0b61ecf23f3202ab968130",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "c10494af-64ca-4110-9069-2d5c546bdae0",
"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,
... |
2cbac0ad-a42d-46b7-849b-70457b5ade85 | easy | Ranoobaba | 2026-08-20 15:45:27.181600+00:00 | succeeded | 97aaf829333110f0ab20c9491fd64c975a215b662d5d1a6f74d4154478e73098 | 7,101 | 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": "2cbac0ad-a42d-46b7-849b-70457b5ade85",
"created_at": "2026-08-20 15:45:27.181600+00:00",
"db_md5": "9ea83f2ef6069967f31c958930067fc2",
"submitter": "Syed Rayyan Ali",
"github_login": "Ranoobaba",
"run_id": "f758ab11-097b-4f23-bfab-0b436e2dd83b",
"tier": "easy",
"dataset_id": "e3",
"status": "... | {
"score": {
"mean_loss": 5.775462397183216,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005625
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.982564235866998,
"example_count": 800,
"... |
2cc2c015-25ac-4d6b-81a3-350dc91b1fe4 | easy | liam-gb | 2026-08-14 22:27:19.235508+00:00 | succeeded | ee8e3602e45d4a044761dc84e5d6b53ee9f63b01800876d8c48a95b4ec2aca98 | 15,754 | null | """rns_pm_gather_ct_widet_ptnore: widet + per-prime temperature (16-prime bank).
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": "2cc2c015-25ac-4d6b-81a3-350dc91b1fe4",
"created_at": "2026-08-14 22:27:19.235508+00:00",
"db_md5": "5051d831398b6c8922c27ddf8e5e8b36",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "96dd4db0-d54d-4f45-8d31-08a77e289b7d",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.5309442281723022,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.18166666477918625
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.563542604446411,
"example_count": 100,... |
2ccd4294-4c85-4d68-ae1b-dd5898c9c49b | easy | KaustubhKumar05 | 2026-08-26 05:40:37.291675+00:00 | succeeded | 9865a53036bec8416eccf2a7a8d295f95617274e98089ef7c578b5af40204359 | 12,762 | null | """Neural-GPU-style conv GRU, v2s: v2 + speedrun pack.
- LR: warmup, hold max to 85% of wall-clock, steep late cliff (2026 schedule theory)
- identity-at-init: cell output projection zero-initialized
- batch reuse x6 (multi-pass theory)
Base (v2):
Changes over v1 (each backed by measured results in the literature):
... | {
"id": "2ccd4294-4c85-4d68-ae1b-dd5898c9c49b",
"created_at": "2026-08-26 05:40:37.291675+00:00",
"db_md5": "a573e724a777e03d1fe72ec17e11405d",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "095c35b0-4071-40f9-a091-465b73c4f1a5",
"tier": "easy",
"dataset_id": "e8",
"status": "succee... | {
"score": {
"mean_loss": 3.721069812774658,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.3599821776151657
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.987016201019287,
"example_count": 85,
... |
2cd7e375-e9ad-4377-a6f8-e67951f19e1d | easy | jordanrubin | 2026-08-07 01:26:55.697390+00:00 | succeeded | 0eb249f609a0101d70cb07bbf9b0f2aa4cc7c5f7465d3e5e29bc4c3e68abecc6 | 14,786 | 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": "2cd7e375-e9ad-4377-a6f8-e67951f19e1d",
"created_at": "2026-08-07 01:26:55.697390+00:00",
"db_md5": "497d08edf33fa52921753c6fa8942084",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "0b288d90-17d0-4524-99e0-8cef41bad7a1",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 4.738736252386113,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007083333097398281
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.733007672388992,
"example_count": 600,... |
2cd9ad61-aec7-4bed-b0c0-cc5a682e61b5 | easy | yashkant | 2026-08-12 03:56:58.500809+00:00 | succeeded | 04c833153a8d6ef1d91e5e7e531770bac78721e4c9929474846c4f379fbc4de6 | 31,389 | null | """Round1702 one-projection-per-T Dykstra recursion on the D416 carrier.
Each parsed public-T cycle applies one shared learned smooth projection and
alternates the A/B set code and correction memory across cycles. The immutable
anchor is reinjected on every call; all active cycle outputs are averaged, and
terminal CE... | {
"id": "2cd9ad61-aec7-4bed-b0c0-cc5a682e61b5",
"created_at": "2026-08-12 03:56:58.500809+00:00",
"db_md5": "b03b2ca17a1b6408d02aa0f7d50b8038",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "606b187b-71bb-42e5-a845-11505ddf430c",
"tier": "easy",
"dataset_id": "e2",
"status": "succeed... | {
"score": {
"mean_loss": 3.3980451822280884,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.14937500655651093
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.44537878036499,
"example_count": 300,
... |
2cd9d597-8d78-4b07-ba24-89a6d5b6c5a0 | easy | oupadhyay | 2026-08-23 19:50:13.763809+00:00 | failed | 98307a9b765fe4fa7015466c6f4eb079d223c8d9c14d98a1a247373d758059be | 7,504 | null | """BiScan with one order-aware pooled-N FiLM branch before readout."""
import torch
import torch.nn.functional as F
from torch import nn
from benchmark import OptimizerBundle, Submission, assert_model_state
PAD, N, X, T, ANS, DIGIT = 0, 2, 3, 4, 5, 7
MAX_STEPS = 64
class Config:
def __init__(self, spec):
... | {
"id": "2cd9d597-8d78-4b07-ba24-89a6d5b6c5a0",
"created_at": "2026-08-23 19:50:13.763809+00:00",
"db_md5": "aa19a18cc57b0df74b01c940521a9688",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "a261087f-cdc6-45a9-b5a5-2e530b903f56",
"tier": "easy",
"dataset_id": "e5",
"status": "f... | null
|
2cdc6905-9dc3-440f-8191-60fe9f8535c1 | easy | lpbb | 2026-08-07 18:52:31.934710+00:00 | succeeded | 98556529fadd2e0c1d843680ab972281929828829bf682de4b26ea44edc4236d | 24,042 | null | """VARIANT R2b `fourier` — representation-prior axis (phase bank).
Where the campaign stands: memorization cannot be blocked (d=64 with 106k
params still memorizes 3200 rows in 45 s), delayed into irrelevance
(wd 1.0 at ~1000 full-batch epochs), or taxed away (softcap+smoothing
bounds confidence, map unchanged). The ... | {
"id": "2cdc6905-9dc3-440f-8191-60fe9f8535c1",
"created_at": "2026-08-07 18:52:31.934710+00:00",
"db_md5": "957561b78131c4868751936af7afcbca",
"submitter": "Lpbb",
"github_login": "lpbb",
"run_id": "1d8d4455-30c1-481d-a765-b8d12bd7a7f3",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded",
"s... | {
"score": {
"mean_loss": 4.154027829267003,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.02812500011175871
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.249011114358802,
"example_count": 800,
... |
2cdcb7a1-0743-4446-9148-102103ca5eda | easy | shan23chen | 2026-08-08 02:10:01.921319+00:00 | succeeded | 36962cef1e93992728f9de65590e8bc2a39cb45d64a9614608619fa7702710c7 | 20,447 | null | """Generic recurrent Transformer with normalized numeric and CRF objectives.
The model is deliberately task-generic: it learns a transition over prompt and
workspace tokens instead of embedding the public modular-squaring algorithm.
A learned linear-chain CRF models dependencies among adjacent output digits;
its endpo... | {
"id": "2cdcb7a1-0743-4446-9148-102103ca5eda",
"created_at": "2026-08-08 02:10:01.921319+00:00",
"db_md5": "6c18b6c933b36e87a0b3d4ac05c44412",
"submitter": "Shan Chen",
"github_login": "shan23chen",
"run_id": "3144f81b-5a42-4f01-8bb1-943611173e9c",
"tier": "easy",
"dataset_id": "e5",
"status": "succe... | {
"score": {
"mean_loss": 2.9567915201187134,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.014166667126119137
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.9125521183013916,
"example_count": 60... |
2cdf18a8-e578-463d-872e-56e264c81ef3 | easy | EyimofeA | 2026-08-06 23:51:40.592666+00:00 | succeeded | df552728b3013033ad6fab41fa797afced1f076f74eadce9ab492449467be7e8 | 9,471 | null | """Card 2: T=1 Assassin — one-step specialist with weak multi-T fallback.
Learned pairwise x-digit interactions cross-attend to N; iterative refinement;
T-conditioned router between T=1 specialist and generic head. No explicit
digit-product arithmetic.
Source design: GPT-5 Pro (2026-08-07 five-card suite).
"""
from _... | {
"id": "2cdf18a8-e578-463d-872e-56e264c81ef3",
"created_at": "2026-08-06 23:51:40.592666+00:00",
"db_md5": "da8faead62936dc6cfe9331be671ecdb",
"submitter": "mof",
"github_login": "EyimofeA",
"run_id": "2e78eda6-2998-4f07-a119-111e7747a005",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded",
... | {
"score": {
"mean_loss": 2.170519994914114,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004166666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.177875091142184,
"example_count": 600,... |
2ce8a896-3c79-4724-8966-95cdc0d66b6e | easy | sapient-sapiens | 2026-08-17 02:41:19.545834+00:00 | succeeded | 96d137d90e52e00d4bfe7084329085f05efee0981345708cdf647be0670dae02 | 15,120 | null | """Fixed-4 recurrence candidate: v70 source without adaptive halt.
Parent: act64_e6_w1024_dh4_L6_wd1_v70 (width 1024, d_h=4, L=6, WD=1.0).
Primary variable: adaptive ACT over 1--64 (typically hard-halt 2) versus a
fixed four applications of the shared core. The halt controller is removed.
T remains an ordinary prompt ... | {
"id": "2ce8a896-3c79-4724-8966-95cdc0d66b6e",
"created_at": "2026-08-17 02:41:19.545834+00:00",
"db_md5": "a360270b1bd62e5c960d2cc497021af4",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "867a671a-60fa-45ea-be10-8ef15fc801d1",
"tier": "easy",
"dataset_id": "e2",
"status": "su... | {
"score": {
"mean_loss": 3.6094544855754678,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.013541666697710751
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.6398087616719987,
"example_count": 30... |
2ceb4810-f624-4752-8935-e79e9ca5b7d7 | easy | DDanlov | 2026-08-07 09:54:33.588679+00:00 | succeeded | 13e2558ca0f102d0cd5183619c06189af68862732fd0a8bf09bed2fe842a0bab | 15,241 | null | """
Ultra-Scaled 20-Layer Official Competition Submission:
Architecture Specs:
- Depth: 20 Transformer Decoder Layers (NUM_LAYERS = 20)
- Width: D_MODEL = 1024
- Attention Heads: 128 Heads (NUM_HEADS = 128)
- Recurrent Unrolled Steps: K = 120 PLER Steps (NUM_STEPS = 120)
- Total Parameters: ~257.3 Million (Safely withi... | {
"id": "2ceb4810-f624-4752-8935-e79e9ca5b7d7",
"created_at": "2026-08-07 09:54:33.588679+00:00",
"db_md5": "b72b8dffa171437d362e40f00094b101",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "ff2c25aa-fb37-4093-a3e5-92e3dc9d66f8",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.2463878393173218,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.221022605895996,
"example_count": 100,
"exac... |
2cffac9e-7d11-48cc-800f-1f9578800f84 | easy | viridale | 2026-08-03 09:01:49.041931+00:00 | succeeded | d6feb67d395c0959fc188fd0ba7f0fb6265ef7372aa49c806c4ade20e3747c8e | 133,578 | null | """Dataset-global source evidence with an alias-safe boundary latch.
hypothesis: v131's exact source support is broad enough, but pruning all
364,608 programs to 512 after one minibatch can discard the globally
correct deep alias or a distinct complete T=1 boundary class.
axis: opt
target: revised h1 Max T at ... | {
"id": "2cffac9e-7d11-48cc-800f-1f9578800f84",
"created_at": "2026-08-03 09:01:49.041931+00:00",
"db_md5": "5d3ba980c5fafb337e39ded276fe084a",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "33c1e0d9-9dd4-4e2a-ae01-f0fe9ddfa23c",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 0.003014546020381487,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.0030145461205393076,
"example_count": 100,
... |
2d030574-42ad-4616-b7b1-a23ad1c696a8 | easy | sapient-sapiens | 2026-08-14 17:58:24.629064+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": "2d030574-42ad-4616-b7b1-a23ad1c696a8",
"created_at": "2026-08-14 17:58:24.629064+00:00",
"db_md5": "1efadda9e01ae84ec2857432a666c542",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "5bf1495e-97cd-4537-9ef3-1a0579329dc2",
"tier": "easy",
"dataset_id": "e4",
"status": "su... | {
"score": {
"mean_loss": 2.125286244844646,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0007291666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.130324468246787,
"example_count": 120... |
2d047840-64e2-469a-a58c-cf65d0d6ecf5 | easy | TZZheng | 2026-08-06 17:49:44.581100+00:00 | succeeded | 77050784dfede550f938da52acd3bcd94173e3e9aaee1895e73a1797a1033b18 | 7,302 | null | """Self-contained standard Transformer scaling-sweep submission."""
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,
)
VARIANT = {'block... | {
"id": "2d047840-64e2-469a-a58c-cf65d0d6ecf5",
"created_at": "2026-08-06 17:49:44.581100+00:00",
"db_md5": "19e399e0db4c898faa44216cc0e6770c",
"submitter": "TZZheng",
"github_login": "TZZheng",
"run_id": "f0d86858-3097-49c3-9d6e-eb950b67fd95",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.8967952132225037,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.044999998062849045
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.8587276935577393,
"example_count": 10... |
2d0e2e5b-6b49-4a97-b3ad-0c3ce0347fa7 | easy | MarcoF1 | 2026-08-13 06:32:01.310200+00:00 | succeeded | 426c87c5ec923d783097b973f6d0ad3ce3cdedcf3cc0f77f0aff4470a585ac05 | 5,887 | null | """Weight-tied looped transformer; latent depth scales with the prompt's T.
Hypothesis treatment arm: one tied block learns "one modular-squaring step" and
is re-applied 2T+2 times, where T is parsed from the prompt's digit tokens.
Input re-injection and per-iteration RMSNorm keep the state bounded so the
recurrence c... | {
"id": "2d0e2e5b-6b49-4a97-b3ad-0c3ce0347fa7",
"created_at": "2026-08-13 06:32:01.310200+00:00",
"db_md5": "bb142a51fe7ccd7eb8a972efd9767188",
"submitter": "Marco Fleming",
"github_login": "MarcoF1",
"run_id": "390cd248-859a-43b5-bcd4-e3640b249e6c",
"tier": "easy",
"dataset_id": "e1",
"status": "succ... | {
"score": {
"mean_loss": 3.689873021439721,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03166666643073161
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.9168827533721924,
"example_count": 100,... |
2d139725-8e1d-4e68-bb9e-55471fda70bd | easy | DESU-CLUB | 2026-08-30 18:43:53.448617+00:00 | succeeded | b76551433ca464f807900f58503352b456867d65a3323cbb35f0afb40c624055 | 13,947 | null | """Discrete-state loop over a place-value tape, trained with orthogonalize-only Muon.
The recurrent state carried between squarings is the *digit string*, not a
residual stream. Each outer iteration reads (N, x) off an LSB-first place-value
tape, emits digit logits for one squaring, hard-argmaxes them (straight-throu... | {
"id": "2d139725-8e1d-4e68-bb9e-55471fda70bd",
"created_at": "2026-08-30 18:43:53.448617+00:00",
"db_md5": "e7ead3cbb3a1c810b23a7d0f03b90169",
"submitter": "Low Keng Hoong, Warren",
"github_login": "DESU-CLUB",
"run_id": "ad2d0806-dc14-4846-90ef-7267b2b2d949",
"tier": "easy",
"dataset_id": "e6",
"sta... | {
"score": {
"mean_loss": 1.0185516921624744,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.6891892181860434
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.7969363331794739,
"example_count": 60,
... |
2d157f64-d84e-40d6-9da3-5e4b362a58cc | easy | ehonig | 2026-08-14 01:14:28.605466+00:00 | succeeded | 6cc7f0ea1126c8f8a210569bdbe0d322f6256813c1a3e5cecb857c62b7684268 | 54,754 | 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": "2d157f64-d84e-40d6-9da3-5e4b362a58cc",
"created_at": "2026-08-14 01:14:28.605466+00:00",
"db_md5": "5e22f84b86ad97e9c17d31e7481f6320",
"submitter": "Edouardo Honig",
"github_login": "ehonig",
"run_id": "b875c0a0-9e1a-41bf-90c5-f492856710ee",
"tier": "easy",
"dataset_id": "e1",
"status": "succ... | {
"score": {
"mean_loss": 5.224117755889893,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.1366666667163372
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.490714073181152,
"example_count": 100,
... |
2d16ac04-b39d-4f1b-8eb5-87824e748ac7 | easy | erdavis0 | 2026-08-29 01:14:04.384513+00:00 | succeeded | c30171170d2368108bbd1f8de018dfbd1d387e0e28dec2d543ae91ea35b7ad59 | 20,136 | null | """Curvature table with a GPU-resident cached modulus-relative residual."""
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,
TokenLos... | {
"id": "2d16ac04-b39d-4f1b-8eb5-87824e748ac7",
"created_at": "2026-08-29 01:14:04.384513+00:00",
"db_md5": "efb4da441ead29fd5826581673cba669",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "886d9516-bf5d-48d6-8b08-adea9ff4c8d5",
"tier": "easy",
"dataset_id": "e10",
"status": "succeeded"... | {
"score": {
"mean_loss": 0.12815570808288612,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.9823232293128967
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.080156082229223e-05,
"example_count": ... |
2d1c5f41-0355-447e-ac19-fc802806b45b | easy | erdavis0 | 2026-08-27 01:12:31.372773+00:00 | succeeded | f696b4a0dbdab40d944b3bb47ef7b8e716ff6e0849099f9cf030611313221c72 | 31,312 | null | """Typed-prefix reversible transition B3: sequence mixture plus balance.
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": "2d1c5f41-0355-447e-ac19-fc802806b45b",
"created_at": "2026-08-27 01:12:31.372773+00:00",
"db_md5": "aa01dfb95948365c687b75f02f03cd86",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "f87fa10a-5cbe-4b09-94ba-605e966f46f0",
"tier": "easy",
"dataset_id": "e8",
"status": "succeeded",... | {
"score": {
"mean_loss": 7.131964921951294,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.09300357149913907
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.019749641418457,
"example_count": 85,
... |
2d225f3c-21fd-4e2e-a5e5-cf85c398bf9d | easy | benjaminW2025 | 2026-08-20 23:19:45.488965+00:00 | succeeded | 5db8da585d49776e364482bd964eeabdeb62d09358da91f767ca94009230ae49 | 14,141 | null | """Phase 1 S2x4 scratchpad with exponent-0.5 hard-sequence CE."""
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,
)
... | {
"id": "2d225f3c-21fd-4e2e-a5e5-cf85c398bf9d",
"created_at": "2026-08-20 23:19:45.488965+00:00",
"db_md5": "58f269b326463d4dbf0e84874c6c595d",
"submitter": "benjawesome",
"github_login": "benjaminW2025",
"run_id": "62f63582-66a2-43e8-bfe4-752f31415996",
"tier": "easy",
"dataset_id": "e3",
"status": "... | {
"score": {
"mean_loss": 2.267544660307095,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.014375000046566128
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2702013877544474,
"example_count": 800... |
2d23b7c8-53aa-4a3b-8127-b625d4283922 | easy | 0Chris5R | 2026-08-07 12:30:52.614705+00:00 | succeeded | a1c69104254c4d6e3feafd648bba486bdf766aaa08968b0f5397cf9e895aadf6 | 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": "2d23b7c8-53aa-4a3b-8127-b625d4283922",
"created_at": "2026-08-07 12:30:52.614705+00:00",
"db_md5": "7bf457d4646c39e4bd9ad7fc70a4f1f7",
"submitter": "Chris ",
"github_login": "0Chris5R",
"run_id": "af3d0c58-061d-4044-a169-e1562eb287e8",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.831068298577346,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0483333333581686
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.777678679992265,
"example_count": 600,
... |
2d2c5ff7-d51e-4028-93f2-695032419f6d | easy | karanganesan | 2026-08-05 10:58:26.692447+00:00 | succeeded | 7d02c8c5da4773ccf5e1dfdc89c48fdcf33e7736d7cef4531e90a51a9bcef450 | 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": "2d2c5ff7-d51e-4028-93f2-695032419f6d",
"created_at": "2026-08-05 10:58:26.692447+00:00",
"db_md5": "c3c44d28bb5fd98c8e6bdb7fc4fda49b",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "f455a840-895e-47a7-8d33-391e05f1a2e2",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 11.965276002883911,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.054999999701976776
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 16.726831436157227,
"example_count": 10... |
2d2d136d-c97e-47b1-8350-8c5f0a9bfa9f | easy | shreyash-chonkie | 2026-09-01 01:07:39.617005+00:00 | succeeded | fe255b7a22f9ffc0a2604643616d8f387ac2ba0bd49d32e00ab6522cafd80ad4 | 10,316 | null | """Diagnostic recurrent bottleneck with semantic input training wheels."""
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,
a... | {
"id": "2d2d136d-c97e-47b1-8350-8c5f0a9bfa9f",
"created_at": "2026-09-01 01:07:39.617005+00:00",
"db_md5": "2065358c28acaa9c1967a8d38f04bd4b",
"submitter": "Shreyash",
"github_login": "shreyash-chonkie",
"run_id": "cedbc109-e54a-4897-8f61-b7956c9c55bb",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 3.8241971731185913,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05499999783933163
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.9485018253326416,
"example_count": 100... |
2d330903-f63b-40ef-85a8-e80cf9dc9cae | easy | chuk-uzowihe | 2026-08-06 23:29:28.887527+00:00 | succeeded | 6fbe22bed2df69d3f434489fe778a9759e98bbc997f686ad4d4aeebeec32e22d | 31,911 | null | """One Layer Deeper submission: FPRM over a digit lattice with a Fourier T latent.
Structure (v4)
--------------
1. N and x are *digit slots*: each digit is gathered into a dense per-number
grid (column j = place value 10^j, short rows left-padded with explicit
zero digits) and embedded as digit token + learned ... | {
"id": "2d330903-f63b-40ef-85a8-e80cf9dc9cae",
"created_at": "2026-08-06 23:29:28.887527+00:00",
"db_md5": "9df3970cd2e1bbf17160245609af0baa",
"submitter": "Chuk Uzowihe",
"github_login": "chuk-uzowihe",
"run_id": "4a13a97f-4578-42c4-a3e5-eb810b125609",
"tier": "easy",
"dataset_id": "e4",
"status": "... | {
"score": {
"mean_loss": 11.57301090632379,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0036458333333333334
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 11.65764838852096,
"example_count": 120... |
2d362f0d-edf5-4447-a085-26e8b003f877 | easy | KaustubhKumar05 | 2026-08-27 17:09:32.071957+00:00 | succeeded | 8cbcae5effbcce803a28e428aaf162be5e247f6a5b079fdd8e4c4d73ee955bbb | 24,094 | null | """cell_red3 -- cell_best plus QUOTIENT-BIT auxiliary supervision for the reduction half.
The carry-sum auxiliary (cell_aux) supervises the N-INDEPENDENT half: squaring is the same
function for every modulus, so all 48k examples reinforce it. It measured as our best arm.
Reduction is the N-DEPENDENT half, and it is wh... | {
"id": "2d362f0d-edf5-4447-a085-26e8b003f877",
"created_at": "2026-08-27 17:09:32.071957+00:00",
"db_md5": "8b68f736afc3e8cb5eb08dff6cc7171c",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "70931519-0b4e-4e24-89e6-7800c47d18cb",
"tier": "easy",
"dataset_id": "e3",
"status": "succee... | {
"score": {
"mean_loss": 6.059721318526755,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.016875
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.616459068953113,
"example_count": 800,
"... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.