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 |
|---|---|---|---|---|---|---|---|---|---|---|
229bf357-8204-4e21-ab03-806e22bb354a | easy | yashkant | 2026-08-23 22:01:59.715457+00:00 | succeeded | 717945598734f9a1276a505f5b87aa0891891ab4b1dbe7e4211ba80baee1a6cf | 1,086 | null | from __future__ import annotations
import torch, torch.nn as nn, torch.nn.functional as F
from torch import Tensor
from benchmark import ModelSpec, OptimizerBundle, OptimizerSpec, Submission, TokenLossBatch, assert_model_state
class Config:
def __init__(self, v, m): self.vocab_size=v; self.max_seq_len=m
class Min... | {
"id": "229bf357-8204-4e21-ab03-806e22bb354a",
"created_at": "2026-08-23 22:01:59.715457+00:00",
"db_md5": "d5f7a973bc1ce012ee1e6c76e62756b9",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "2ba532ea-50a2-4fbf-a549-879800708aba",
"tier": "easy",
"dataset_id": "e1",
"status": "succeed... | {
"score": {
"mean_loss": 2.2866389751434326,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.030000001192092896
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.427333116531372,
"example_count": 100... |
22a52c33-d996-4268-8822-1c5a73bc583b | easy | TZZheng | 2026-08-11 14:48:15.373753+00:00 | succeeded | 43806d0e88866ea440123c87a69ef372b31844b20e573542fdbf73cc2db5a9c4 | 8,480 | null | """End-to-end learned-position Transformer specialized toward the first depth rung.
This is a deliberately small change to the tested A6 Transformer: each of the
ten blocks is applied once instead of twice, and a boundary-preserving loss
gives T=1 examples extra weight. All examples retain nonzero weight. The
model ... | {
"id": "22a52c33-d996-4268-8822-1c5a73bc583b",
"created_at": "2026-08-11 14:48:15.373753+00:00",
"db_md5": "0cf13df9db0a7739e75d753426790e77",
"submitter": "TZZheng",
"github_login": "TZZheng",
"run_id": "dae17654-0aa5-4a3d-a9d7-5d7979650ae4",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.197590015236849,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0008333333333333334
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.200479949001774,
"example_count": 600... |
22ae8915-6181-483f-a66d-37fdaa1909d1 | easy | 0Chris5R | 2026-08-01 06:31:12.186221+00:00 | succeeded | 5c1056b3309469e194b702ac60ce57094578047a48fc659b08e19517a5fd9346 | 19,851 | null | """Post-normalized, input-recalling shared Neural GPU with progressive loss.
The model aligns decimal digits by significance, forms a learned low-rank
second-order feature map of the current state, and evolves a two-dimensional
workspace with the original convolutional-GRU equations. Each arithmetic
transition applie... | {
"id": "22ae8915-6181-483f-a66d-37fdaa1909d1",
"created_at": "2026-08-01 06:31:12.186221+00:00",
"db_md5": "95adbb03a19996b098ac1fefb5cab484",
"submitter": "Chris ",
"github_login": "0Chris5R",
"run_id": "7ad0dd57-74a5-4ce6-abb3-e1b4d521a780",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.1694814117824537,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0050000000248352685
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.167537210233543,
"example_count": 60... |
22b7baeb-1626-49c9-a7a8-d6b27fab6d5f | easy | liam-gb | 2026-08-17 09:32:56.896498+00:00 | succeeded | 24fc878b19987564e88eccf1d98937ae5735fa8ab620fdef470b9eacfec14e2f | 16,740 | null | """rns_pm_gather_ct_widet_l32nore_r8: single change vs
rns_pm_gather_ct_widet_l32nore — the batch-reuse policy returns True, so the
evaluator reuses each fetched batch for the maximum eight optimizer updates.
WHY. On the granular Easy suites one batch is 85-97% of the training split,
so every step ends an epoch and th... | {
"id": "22b7baeb-1626-49c9-a7a8-d6b27fab6d5f",
"created_at": "2026-08-17 09:32:56.896498+00:00",
"db_md5": "46c5decfff7d97a850fb297981e9d310",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "1f7611f8-498d-4808-be3e-d8192195d1e8",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 3.297567665576935,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.5550000071525574
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.741147994995117,
"example_count": 100,
... |
22b8c2b5-9011-434f-b7be-c623edd08770 | easy | DDanlov | 2026-08-12 20:18:00.104370+00:00 | failed | 5e2371dff926bd6839dee3b4b9860f9b884a2dfd5b6030fe6572b7b9278ceb45 | 25,988 | 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": "22b8c2b5-9011-434f-b7be-c623edd08770",
"created_at": "2026-08-12 20:18:00.104370+00:00",
"db_md5": "9071fd4645ee4c014d5cbdb3e9321ead",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "fced2771-e797-4549-a9fd-a0888595bc89",
"tier": "easy",
"dataset_id": "e5",
"status": "failed",
... | null
|
22bc2917-b343-4b16-a1ed-f1213715b3c8 | easy | DDanlov | 2026-08-09 00:37:20.680997+00:00 | succeeded | 336dab7c7f688d23eb45ba40127ce89d509695692d183039e0957407b4d8d841 | 15,978 | 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
class RohanShampoo(Optimizer):
"""Self-contained RohanShampoo optimizer with eigenvalue ma... | {
"id": "22bc2917-b343-4b16-a1ed-f1213715b3c8",
"created_at": "2026-08-09 00:37:20.680997+00:00",
"db_md5": "1c1481c8c5835d8fc0f7de55a4a49de8",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "3d23d68f-f246-481a-8ee3-092c3209c45a",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 44.88976916172855,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.024999999664723875
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 53.24393844604492,
"example_count": 100,... |
22bce7fc-25c7-4fee-b805-d8e72b7a4611 | easy | Yalyenea | 2026-08-11 21:27:24.088257+00:00 | succeeded | 361ea030488ad892482d14d59ca084f5788c7177ac814744807088b427c0afa3 | 14,150 | null | """R4 full-bandwidth candidate with a causal 64-bin T counter register."""
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,
)... | {
"id": "22bce7fc-25c7-4fee-b805-d8e72b7a4611",
"created_at": "2026-08-11 21:27:24.088257+00:00",
"db_md5": "01839307451452081cbdaf4359e0f0e1",
"submitter": "yfff",
"github_login": "Yalyenea",
"run_id": "17a7c624-eeb3-401a-b607-f20a13e2bf7b",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded",
... | {
"score": {
"mean_loss": 6.36755691764054,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009583333693444728
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.140053749084473,
"example_count": 600,
... |
22c10f86-f7ba-4212-b31c-c5ea4626d543 | easy | khushidahi | 2026-08-23 16:35:01.834621+00:00 | succeeded | 832aa947b0d379e04d73a83bd9ad4fa724419bd37ce2c13584505365247858c8 | 17,660 | null | """R426: matched Abacus arm of the graph-safe Deep Thinking baseline.
R424 is the frozen no-Abacus control. R426 changes only the input
representation: every active decimal place receives a learned Abacus
significance vector. During training a single batch-shared start offset is
drawn uniformly from beta in {1, ...,... | {
"id": "22c10f86-f7ba-4212-b31c-c5ea4626d543",
"created_at": "2026-08-23 16:35:01.834621+00:00",
"db_md5": "34922516efcbcea77939dbb989d1843e",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "94a8c7e0-d823-4eb9-8c0e-f785980d26d7",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.1549326559593496,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004166666679084301
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1502299779199165,
"example_count": 60... |
22cd3342-8ea2-43bc-b539-895119d3bb06 | easy | jordanrubin | 2026-08-05 21:05:26.796341+00:00 | succeeded | ceed1ecd6fc743a94287b5316bf6ec8b2b96897c427bc663728a832f98579f51 | 24,859 | null | """Adaptive exploration template for the exact-T looped Transformer.
The T=1 computation is the proven value-Fourier baseline cell: a D=256,
three-block bidirectional Transformer with abacus digit positions and learned
Fourier features over the parsed integer x. For T>1 the *same* three-block
cell is reused exactly T... | {
"id": "22cd3342-8ea2-43bc-b539-895119d3bb06",
"created_at": "2026-08-05 21:05:26.796341+00:00",
"db_md5": "7b0556917b9cc13f77bb6e3a8df7cd14",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "ae3e6b48-0e4b-49b5-ad22-c15631ef73fa",
"tier": "easy",
"dataset_id": "e1",
"status": "s... | {
"score": {
"mean_loss": 6.5383303230837395,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.2533333313961824
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.334372520446777,
"example_count": 100,
... |
22e084be-e8ae-4f6f-ba9d-581c31d7ab71 | easy | viridale | 2026-08-16 03:56:01.820772+00:00 | succeeded | 19d07d9030dcba5927ae45e13f087d19935191a93e686c7680777deaeed611fd | 18,061 | null | """Integer-polynomial boundary quotient with clip-invariant acquisition.
hypothesis: quotienting the T1 coordinate fiber to one autonomous polynomial neuron
and projecting its learned coefficients to integers makes the acquired boundary
compose exactly and transfer across modulus identities.
axis: arch
target:... | {
"id": "22e084be-e8ae-4f6f-ba9d-581c31d7ab71",
"created_at": "2026-08-16 03:56:01.820772+00:00",
"db_md5": "aa64c42e392b1cd8e58ee153383a4ab4",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "748f444f-d398-4751-949c-b103e3ea62ad",
"tier": "easy",
"dataset_id": "e10",
"status": "succ... | {
"score": {
"mean_loss": 0.13438325375318527,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.13522446155548096,
"example_count": 125,
"e... |
22e7c557-4093-4a71-8681-1ffb96e294c5 | easy | kushasareen | 2026-08-10 20:58:45.531019+00:00 | succeeded | 7e6e8eb0b9608d01132b480daca93fdf12b78a4794d08f1c803f82deae021aa6 | 7,846 | null | """Variant B: variant A plus Attention Residuals over loop iterations.
Kimi K3's AttnRes lets each layer attend over all preceding layers' outputs
instead of only receiving the summed residual stream. Here the recurrent
step attends over the states of all previous loop iterations through a
learned pseudo-query. The re... | {
"id": "22e7c557-4093-4a71-8681-1ffb96e294c5",
"created_at": "2026-08-10 20:58:45.531019+00:00",
"db_md5": "b5ed0bafac45eb876c8ac1a892fbf658",
"submitter": "Kusha Sareen",
"github_login": "kushasareen",
"run_id": "a03d8749-e095-4ddd-8eb6-5f8c3f2d5852",
"tier": "easy",
"dataset_id": "e1",
"status": "s... | {
"score": {
"mean_loss": 2.1651912927627563,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04999999888241291
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1087915897369385,
"example_count": 100... |
22f08b25-7b58-4bb7-9abf-dfc4e398d10c | easy | sapient-sapiens | 2026-08-16 20:17:53.493605+00:00 | succeeded | 4e41982ddf4c99f495d070f162da2a676903291242ca515083c2ee2969e16a4d | 21,254 | null | """E6 early-stop probe: frozen v55 with MAX_STEPS=250.
The only change relative to act64_e6_w1024_dh4_L6_v55 is the evaluator step
cap (20000 -> 250), stopping in the 30-94% train-exact window. T is
diagnostic-only.
"""
from __future__ import annotations
import math
import time
import torch
import torch.nn.function... | {
"id": "22f08b25-7b58-4bb7-9abf-dfc4e398d10c",
"created_at": "2026-08-16 20:17:53.493605+00:00",
"db_md5": "708e959dd2eebcd4fa893695f4a8374e",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "017fc565-1eca-4eec-b6a0-941d5d764905",
"tier": "easy",
"dataset_id": "e6",
"status": "su... | {
"score": {
"mean_loss": 4.77086223641487,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.06351351808454539
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.9812374114990234,
"example_count": 60,
... |
22f5aa0b-97a4-4545-821f-bc30b30ab3ac | easy | briheet | 2026-08-07 03:19:23.780801+00:00 | succeeded | 8361aae532f50ee9f49623f5e9e7f2c5d3512b888852ab97711d83a76bd80bbc | 4,097 | 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 = 32
NUM_HEADS = 4
NU... | {
"id": "22f5aa0b-97a4-4545-821f-bc30b30ab3ac",
"created_at": "2026-08-07 03:19:23.780801+00:00",
"db_md5": "eef568cd5d53d93ec3cb21c75af50c8a",
"submitter": "Aizenn",
"github_login": "briheet",
"run_id": "2908cef6-dcf5-49eb-9aa0-0daa8bac9f99",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",... | {
"score": {
"mean_loss": 2.4852306246757507,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04999999701976776
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.975520372390747,
"example_count": 100,... |
22f740f8-0675-4b64-98a0-f73255abb700 | easy | adambrzosko | 2026-08-30 14:19:20.341022+00:00 | succeeded | 7bf30e19d90e7f6cadb4ad67d0b2cd51634a4fd2d29a507be6f840a0dd78f8c1 | 3,628 | 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 = 128
NUM_HEADS = 4
... | {
"id": "22f740f8-0675-4b64-98a0-f73255abb700",
"created_at": "2026-08-30 14:19:20.341022+00:00",
"db_md5": "eaac69fe345f82973fcf392c504a40ff",
"submitter": "Adam Brzosko",
"github_login": "adambrzosko",
"run_id": "3368189e-56fb-4a1d-a3cd-ef2e609e3d53",
"tier": "easy",
"dataset_id": "e6",
"status": "s... | {
"score": {
"mean_loss": 3.024079918861389,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006756756920367479
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.0479207038879395,
"example_count": 60,... |
22fbcdb7-4912-43a7-b4c6-b7838313be01 | easy | oupadhyay | 2026-08-29 18:07:10.114364+00:00 | succeeded | 2f117d6dcce73a53253aeb7b599eac0444feea9ca195c0d449d632a5ebe36b59 | 3,988 | null | """Conditional optimizer screen: lr3."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import nn
from benchmark import ModelSpec,OptimizerSpec,OptimizerBundle,Submission,assert_model_state
W,D,H=4,64,32
class C:
def __init__(s,vocab_size,max_seq_len):s.vocab_size,s.... | {
"id": "22fbcdb7-4912-43a7-b4c6-b7838313be01",
"created_at": "2026-08-29 18:07:10.114364+00:00",
"db_md5": "5875b6ee33ec06641c91ae76c8ce6b0c",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "0b6b2e95-a518-44d0-b5aa-53d89a8aa7fa",
"tier": "easy",
"dataset_id": "e7",
"status": "s... | {
"score": {
"mean_loss": 1.9166486859321594,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.08155080303549767
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.8357034921646118,
"example_count": 85,... |
2305ffd5-8cdc-44e8-95ae-7b381e696d84 | easy | liam-gb | 2026-08-16 16:21:33.892796+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": "2305ffd5-8cdc-44e8-95ae-7b381e696d84",
"created_at": "2026-08-16 16:21:33.892796+00:00",
"db_md5": "5051d831398b6c8922c27ddf8e5e8b36",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "a1df1ee3-3787-4999-8de5-e4384191c209",
"tier": "easy",
"dataset_id": "e2",
"status": "succeeded"... | {
"score": {
"mean_loss": 5.231051743030548,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.4145833421498537
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.236486434936523,
"example_count": 300,
... |
230e3828-7bb0-46fa-adc4-a5cf9f0f616d | easy | DESU-CLUB | 2026-08-30 18:46:29.873016+00:00 | succeeded | dc13c8800f5a14114e56a4defc371e9d5c45291cbd5c91e0c21787830214a868 | 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": "230e3828-7bb0-46fa-adc4-a5cf9f0f616d",
"created_at": "2026-08-30 18:46:29.873016+00:00",
"db_md5": "9e5d793767851c14274e9e1f21709964",
"submitter": "Low Keng Hoong, Warren",
"github_login": "DESU-CLUB",
"run_id": "afcb23ed-89cb-4d40-91b0-8bdafda4cec1",
"tier": "easy",
"dataset_id": "e6",
"sta... | {
"score": {
"mean_loss": 1.3705623827568472,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.5549549605395343
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.1502190828323364,
"example_count": 60,
... |
230e8d66-05c6-46c0-adad-fcc63c7f16aa | easy | DDanlov | 2026-08-24 22:10:15.509053+00:00 | succeeded | ab80873f8352b85bc58df0834fc2d407f082cbf11329c1dc14d7eb4726662a88 | 49,569 | null | """
Unified Pure Skipless Prefix Deliberation Model + Transformer Readout Decoder
Architecture:
- Deliberation Core: dim=1536, num_heads=24, d_ff=6144 (Pure Skipless SO(2)^{D/2} Givens Recurrence)
- Token Layout: Recurrent loop processes ONLY [Prompt (L_p), 4 Pause Registers] (Zero target tokens in loop)
- Recurrent At... | {
"id": "230e8d66-05c6-46c0-adad-fcc63c7f16aa",
"created_at": "2026-08-24 22:10:15.509053+00:00",
"db_md5": "26884e19ba52d2d6a06696577a1d9939",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "f917575c-cc0a-4cae-a550-e861cef195cd",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.9379832052278028,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00666666670391957
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.922371400311389,
"example_count": 600,... |
2314e67b-9619-4732-b66d-8bf467e64ed3 | easy | sapient-sapiens | 2026-08-26 11:05:11.686739+00:00 | succeeded | dd78b2739f614ecc31682b9759ca4d8a6f558407d27b9e184dd1559575cf450e | 20,121 | null | """Standalone nested inner/outer recurrence candidate.
Representation is categorical only: ``E_token + E_group + E_leftrel``. No
field is decoded into a number, no field value sets depth, halting, or any
mixing weight, and the only supervised target is the final answer.
Structure
---------
One prompt encoder produce... | {
"id": "2314e67b-9619-4732-b66d-8bf467e64ed3",
"created_at": "2026-08-26 11:05:11.686739+00:00",
"db_md5": "3e72cf7ff731dcb801e5e985bd871029",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "03ee4992-7994-4ba9-892e-b6af6b990620",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 5.435762192565292,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008333333345750968
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.277600728877457,
"example_count": 600,... |
231c4362-837b-4789-890d-4afa536fd407 | easy | EyimofeA | 2026-08-16 23:52:21.491449+00:00 | succeeded | 13ffe9fdfabe1efc89168256403117190e89221946c04064137dff2b61fcc297 | 10,907 | null | """Binary fused work-state recurrence: width 256 + tuned flattened Muon.
The model receives exact binary representations of evaluator-provided x and N,
then learns the complete transition from final labels only. It has no explicit
squarer, reducer, arithmetic trace, or task oracle.
"""
from __future__ import annotati... | {
"id": "231c4362-837b-4789-890d-4afa536fd407",
"created_at": "2026-08-16 23:52:21.491449+00:00",
"db_md5": "e9e889e80217192f6af2c2618a69829e",
"submitter": "mof",
"github_login": "EyimofeA",
"run_id": "9835c545-b047-4e44-87e7-25cacf460e8d",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded",
... | {
"score": {
"mean_loss": 36.21353958155157,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0020833333333333333
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 36.188340914089046,
"example_count": 60... |
231d7266-e627-40a1-8883-4adeadd6d8c3 | easy | velocizapkar | 2026-08-06 21:51:54.310599+00:00 | succeeded | b50aa41473495876db78f8a1dcc879691866e5c45d407f44806f0b89b36bd9d2 | 53,087 | null | """Fast packed residual-free Attention-Residual transformer with Muon.
The model is a generic learned sequence model. It contains no parser,
arithmetic primitive, task constant, generated data, label feedback, or answer
construction. Every macro-state is produced by a serial pack of attention and
MLP transformations... | {
"id": "231d7266-e627-40a1-8883-4adeadd6d8c3",
"created_at": "2026-08-06 21:51:54.310599+00:00",
"db_md5": "b7893852659777072a8913ec1ca4406f",
"submitter": "Aakanksh Zarapkar",
"github_login": "velocizapkar",
"run_id": "bf64a655-6ed2-47e7-a9d8-9563575df469",
"tier": "easy",
"dataset_id": "e3",
"statu... | {
"score": {
"mean_loss": 2.603618025779724,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0068749996135011315
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.6639316082000732,
"example_count": 80... |
23248ebe-a993-4d84-ab34-1d6dbb15a96b | easy | jordanrubin | 2026-08-03 19:18:42.755419+00:00 | succeeded | 7abd506f00e20667c1cad715722c124645f9b4b0e649120eea838b0ec4aaf2b6 | 15,389 | 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": "23248ebe-a993-4d84-ab34-1d6dbb15a96b",
"created_at": "2026-08-03 19:18:42.755419+00:00",
"db_md5": "6f7ba13277a1be43ba72155c2f865e97",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "a736bf6d-fb9a-4a10-92ff-1a131838b076",
"tier": "easy",
"dataset_id": "e2",
"status": "s... | {
"score": {
"mean_loss": 3.9868681502098857,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05874999888086071
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.908548830278244,
"example_count": 300,... |
2324d22a-284e-4bc7-bd4c-8922b92403c3 | easy | yashkant | 2026-08-14 17:19:38.281346+00:00 | succeeded | 6ac22679d58dac876674ee189a0a98d69ee5a8b2938b55e9439e7d4adc0ce4ef | 33,286 | null | """Round1786 evolutionary crossover of two controlled survivors.
The Round1780 W8/D2 carrier keeps its winning contrastive, exactly
antisymmetric learned set codes. The candidate additionally removes each
initial and recurrent Dykstra correction residual's coordinate mean before
the unchanged tanh bound, combining Ro... | {
"id": "2324d22a-284e-4bc7-bd4c-8922b92403c3",
"created_at": "2026-08-14 17:19:38.281346+00:00",
"db_md5": "826965a26e17d76ebb94eb10b94b8659",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "ceab0609-a25b-4143-90f2-246fa4f82e0b",
"tier": "easy",
"dataset_id": "e2",
"status": "succeed... | {
"score": {
"mean_loss": 4.307932376861572,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.33770834282040596
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.067376613616943,
"example_count": 300,
... |
232d9946-0a8a-4d82-bce9-7e9b4138d6a4 | easy | khushidahi | 2026-08-23 16:17:51.961700+00:00 | failed | a4976399730fc34d1fe021950a310171e59c619a1cfd0150a471ed66c8cd56f7 | 16,365 | null | """R423: AdamW compatibility arm of the equal-example Deep Thinking model.
This is the contest adapter for the from-scratch Bansal et al. (2022) model,
not an extension of an older Transformer candidate. A single learned 1-D
transition receives LSD-aligned categorical X and N tapes, repeatedly recalls
that immutable ... | {
"id": "232d9946-0a8a-4d82-bce9-7e9b4138d6a4",
"created_at": "2026-08-23 16:17:51.961700+00:00",
"db_md5": "dcca418bc60558397f28772e5c2fdfdd",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "5290cb8a-4ac0-4109-9274-01dd732a3c2e",
"tier": "easy",
"dataset_id": "e5",
"status": "fail... | null
|
23328a5f-f4dc-452d-a8e8-57161ef83160 | easy | KaustubhKumar05 | 2026-08-25 16:26:19.100623+00:00 | succeeded | 0dd780da0f2643e57765b3d6e0e2a43ad3d3d1878ba500bfc9e4abda1f9875d1 | 17,167 | null | """conv_abacus: final_v1 + the CORRECT Abacus digit embedding.
final_v1 indexed positions from the end of the whole padded sequence, so the
units digit of N and the units digit of x landed on DIFFERENT embeddings --
the exact failure McLeish et al. (arXiv:2405.17399) identify. This uses a
table indexed by significance... | {
"id": "23328a5f-f4dc-452d-a8e8-57161ef83160",
"created_at": "2026-08-25 16:26:19.100623+00:00",
"db_md5": "3009ed5dc9d1fec3670f0014eba79c15",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "bba97f10-0ca0-4d64-a13f-71911017f014",
"tier": "easy",
"dataset_id": "e3",
"status": "succee... | {
"score": {
"mean_loss": 6.561740316206553,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.01375
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.029930717161738,
"example_count": 800,
"e... |
233666fc-a05a-455b-9948-c795f44394cc | easy | jordanrubin | 2026-08-06 19:56:32.931990+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": "233666fc-a05a-455b-9948-c795f44394cc",
"created_at": "2026-08-06 19:56:32.931990+00:00",
"db_md5": "ef7d67d872a65aa96e81ca53991ab27d",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "39769b02-f3de-47a7-a206-f4832c3e2ad2",
"tier": "easy",
"dataset_id": "e2",
"status": "s... | {
"score": {
"mean_loss": 3.424706823591202,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.2329166610725224
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.571666068454791,
"example_count": 300,
... |
233711cc-4e09-4052-8ee8-bc632f88e641 | easy | yashkant | 2026-08-23 21:49:52.491941+00:00 | failed | 4744b59be15011db5802a99458982729d592849e3e406b66b98b73c7a4096ca8 | 11,547 | null | """One Layer Deeper submission: CRT residue-register machine with a learned squaring
transition and a Fourier "clock" bilinear reconstruction readout.
Idea (end-to-end learned; nothing about the task is hard-coded):
* State = per-modulus residue registers over a fixed bank of small primes. The input
integer x is... | {
"id": "233711cc-4e09-4052-8ee8-bc632f88e641",
"created_at": "2026-08-23 21:49:52.491941+00:00",
"db_md5": "b346d7c1e564f7694de3ebf9fded2683",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "71fec71e-ad7c-4ba6-9827-887212684a51",
"tier": "easy",
"dataset_id": "e1",
"status": "failed"... | null
|
2338f544-144d-4954-8f6d-72fb8661129c | easy | viridale | 2026-08-02 15:15:49.841522+00:00 | succeeded | e9a687910975e81e9c2058c04329436cf50c7a4a8ef6b8d83bcd5f31eb7ca03e | 55,786 | null | """Physically isolated evidence-triggered five-monomial expansion.
hypothesis: v108 correctly skips family-7 computation on practice, but sharing
one 16.9M rule parameter still materializes the unused tail's gradient and
yields only 2,862 official steps. Store the v102 pool and family 7 in two
learned par... | {
"id": "2338f544-144d-4954-8f6d-72fb8661129c",
"created_at": "2026-08-02 15:15:49.841522+00:00",
"db_md5": "1ba1b7bf2246a8d9fc1b9a3d48fdfed2",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "f3067e3b-c66d-4158-b41a-ca3bd612d40e",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 0.003014546214626436,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.00301454615599774,
"example_count": 800,
"... |
23430c01-abc0-4b86-b2eb-d2ff85373f64 | easy | adwhit | 2026-08-14 15:45:57.119761+00:00 | succeeded | edcf27a35c29ddd9d99db06d52278983b8b899f973e0b806e234b52f8fa068dd | 20,984 | 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": "23430c01-abc0-4b86-b2eb-d2ff85373f64",
"created_at": "2026-08-14 15:45:57.119761+00:00",
"db_md5": "d49727f000c4f5e9c1618c01423bbc0c",
"submitter": "adwhit",
"github_login": "adwhit",
"run_id": "3b9f60f7-6776-48c5-ac01-5711d7024f04",
"tier": "easy",
"dataset_id": "e2",
"status": "succeeded",
... | {
"score": {
"mean_loss": 2.864920139312744,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8659191131591797,
"example_count": 300,
"exac... |
23577287-9e4a-4047-a42e-486c84209ff6 | easy | EyimofeA | 2026-08-15 03:21:45.519775+00:00 | succeeded | cbf327b36f8306a10ccfcb52bef481c1ee2d7b22bf1de93013ab8fb818a88268 | 13,707 | null | """Final-label-only Neural GPU square phase plus learned reduction phase.
The model uses generic learned scratch lanes and tied local updates. It has no
arithmetic trace, carry target, task solver, or hard-coded numeric transition.
Training uses only evaluator-provided final labels.
"""
from __future__ import annotati... | {
"id": "23577287-9e4a-4047-a42e-486c84209ff6",
"created_at": "2026-08-15 03:21:45.519775+00:00",
"db_md5": "0e073167a13091b2d07cd3317bef757a",
"submitter": "mof",
"github_login": "EyimofeA",
"run_id": "ab0a93cd-a88a-44bd-9ba5-140419cf556d",
"tier": "easy",
"dataset_id": "e10",
"status": "succeeded",
... | {
"score": {
"mean_loss": 1.9735602736473083,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.3536161631345749
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.876900315284729,
"example_count": 125,
... |
235773df-8f23-4b7f-9b12-73ea5d11e97a | easy | oupadhyay | 2026-08-24 00:29:48.769693+00:00 | succeeded | 3ad0766509e2b8747c870e7564ab87e67285a804af8589e5c21952fac99fd96b | 4,080 | 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": "235773df-8f23-4b7f-9b12-73ea5d11e97a",
"created_at": "2026-08-24 00:29:48.769693+00:00",
"db_md5": "f15a50c064876ee68bb2ccd247ba7256",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "1b57fcd8-b51c-4186-91f2-6ae18f7ed123",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.168242682012164,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00791666670391957
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1546544109224737,
"example_count": 600,... |
235969cd-49fb-42fe-a1e7-9337a1acccad | easy | karanganesan | 2026-08-06 04:25:59.685901+00:00 | succeeded | 2d5be1d82b063a121f9734409172fd1af2d7d36788b40293d97995cb43e999af | 23,810 | 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": "235969cd-49fb-42fe-a1e7-9337a1acccad",
"created_at": "2026-08-06 04:25:59.685901+00:00",
"db_md5": "da04c573398909f0dd442a18c20ed23c",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "708efce2-c404-45a5-affb-e21f4fdc1d68",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 10.371922492980957,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.023333333432674408
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 10.68427848815918,
"example_count": 100... |
235dce4d-27cb-4a85-8a2d-e8f99b5f8b7c | easy | lpbb | 2026-08-20 11:50:02.929529+00:00 | succeeded | 8f10b6a419a7f7d562471f953248811bc9f0579fe64e3b89c0834aa9433b007f | 15,613 | null | """FORBIDDEN PROBE — local experiment only, never submit to the leaderboard.
This variant breaks the competition rules on purpose: it *computes the
intermediate states* x^(2^t) mod N inside the model (parsing N and x from the
prompt tokens) and deep-supervises every loop iteration against them. The
point is to measure... | {
"id": "235dce4d-27cb-4a85-8a2d-e8f99b5f8b7c",
"created_at": "2026-08-20 11:50:02.929529+00:00",
"db_md5": "ba6db70807945bcdc109b9361cff2dc8",
"submitter": "Lpbb",
"github_login": "lpbb",
"run_id": "65d55ace-80bd-40ef-865a-53d6e9a311a2",
"tier": "easy",
"dataset_id": "e2",
"status": "succeeded",
"s... | {
"score": {
"mean_loss": 2.348598365784634,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005208333364377419
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.4647620225478826,
"example_count": 300... |
2364aa08-e1ed-4266-a90a-e1002a43546a | easy | liam-gb | 2026-08-17 01:53:29.588835+00:00 | succeeded | 24fc878b19987564e88eccf1d98937ae5735fa8ab620fdef470b9eacfec14e2f | 16,740 | null | """rns_pm_gather_ct_widet_l32nore_r8: single change vs
rns_pm_gather_ct_widet_l32nore — the batch-reuse policy returns True, so the
evaluator reuses each fetched batch for the maximum eight optimizer updates.
WHY. On the granular Easy suites one batch is 85-97% of the training split,
so every step ends an epoch and th... | {
"id": "2364aa08-e1ed-4266-a90a-e1002a43546a",
"created_at": "2026-08-17 01:53:29.588835+00:00",
"db_md5": "46c5decfff7d97a850fb297981e9d310",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "81e1a635-d1a3-4cef-9e9a-31d5b4f4150c",
"tier": "easy",
"dataset_id": "e10",
"status": "succeeded... | {
"score": {
"mean_loss": 5.020414710044861,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.30830303207039833
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.032134056091309,
"example_count": 125,
... |
236fd78b-b27a-4a16-bc5a-5d9cacd09a74 | easy | karanganesan | 2026-08-08 05:21:53.294810+00:00 | succeeded | 8d9eecdc8607c91b190336e24b49597738decc69ce7d4e24ee5930ae97cfcad7 | 27,127 | 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": "236fd78b-b27a-4a16-bc5a-5d9cacd09a74",
"created_at": "2026-08-08 05:21:53.294810+00:00",
"db_md5": "3e8439ba8ab561733e7dc883fab4ecb2",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "633ba3ba-bf7f-43e5-8a4b-392e9ebd37d8",
"tier": "easy",
"dataset_id": "e5",
"status": ... | {
"score": {
"mean_loss": 6.98724365234375,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.002916666679084301
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.318107604980469,
"example_count": 600,
... |
23702d7d-4b4a-47cb-a42b-dfebd3b28160 | easy | ddoan | 2026-08-14 10:29:40.415981+00:00 | succeeded | 80273c7c56fcef8d2a2f227640aaf495082d1f1081af734c77f83ef37fc70186 | 10,222 | null | """Place-aligned active memory with hard-gated diagonal CGRU updates."""
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_sta... | {
"id": "23702d7d-4b4a-47cb-a42b-dfebd3b28160",
"created_at": "2026-08-14 10:29:40.415981+00:00",
"db_md5": "70147871ff133625ee65be389217986d",
"submitter": "Doug Doan",
"github_login": "ddoan",
"run_id": "2ba0ff29-aaf4-45da-aed2-0ea39fa8e5bd",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.14412968851298,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00625
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.145586385213381,
"example_count": 800,
"ex... |
23735645-ca43-420c-adfe-f853d7d7a947 | easy | DDanlov | 2026-08-28 22:07:59.775770+00:00 | succeeded | 68b82569c8fac1f599d999e732ac382fa66701c11d5bec1c69ec77db36993491 | 15,078 | null | """
Trial 33: Model C: D=1024, H=16, M=4, u=2 (T=8)
"""
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": "23735645-ca43-420c-adfe-f853d7d7a947",
"created_at": "2026-08-28 22:07:59.775770+00:00",
"db_md5": "11119624c38f9b24eb4eea20877bf191",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "54ef1212-4016-46f2-b323-cf77f4a9eca6",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 183.27368866291084,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.002500000024835269
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 181.33560910246297,
"example_count": 60... |
2380e7cf-b523-4527-b059-c6161f76161c | easy | sankalp1999 | 2026-08-04 12:18:40.392790+00:00 | succeeded | a3cca4a00c52d5b0cd64193ece0179c9279dc19826195ac6af5084a13338c5e7 | 7,511 | null | """Latent recurrence: encode once, apply a shared squaring map T times, decode.
v4 put the loop inside a transformer and underfit badly (train acc 4% after
4,432 steps) because attention cost ~7ms/step at 13 tokens and bought nothing.
v2 fit the train set perfectly with a cheap flattened encoder but memorized,
because... | {
"id": "2380e7cf-b523-4527-b059-c6161f76161c",
"created_at": "2026-08-04 12:18:40.392790+00:00",
"db_md5": "8f8dc8bc0d1c8c5ff09b9db259ae64e8",
"submitter": "Sankalp ",
"github_login": "sankalp1999",
"run_id": "2b917978-e08e-44e2-ab78-c19ca5486acc",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 6.6024370193481445,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.125
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.184398651123047,
"example_count": 100,
"ex... |
238c3841-5ad5-4d54-9408-fafbb7c718f2 | easy | karanganesan | 2026-08-04 02:10:49.241550+00:00 | succeeded | a2a6a52febad15170cc09f132592c0b881e3662c6d1acc0808786b9f15e5e47f | 11,993 | 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": "238c3841-5ad5-4d54-9408-fafbb7c718f2",
"created_at": "2026-08-04 02:10:49.241550+00:00",
"db_md5": "59a0513e316bcd4bf0f0a6b66c911b28",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "7e5de517-33eb-4a9e-99f8-ac71f410eb67",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 2.1435985565185547,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0033333334140479565
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.9177846908569336,
"example_count": 1... |
238dd231-df2a-449c-bb9e-b1eb00b95bcb | easy | erdavis0 | 2026-08-21 14:12:16.262906+00:00 | succeeded | e37358a21622100ab57ead40071bc75817c1cd70ff5e1eca862fdf0ea3de1faf | 4,764 | null | """Single fused forward GRU augmented by an invariant global prompt summary."""
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... | {
"id": "238dd231-df2a-449c-bb9e-b1eb00b95bcb",
"created_at": "2026-08-21 14:12:16.262906+00:00",
"db_md5": "2ddc044b08096c26eb27e0bacf9f9213",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "568f658f-8b3b-4913-8ab6-b9ba085b4f6d",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",... | {
"score": {
"mean_loss": 8.49101209640503,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03999999910593033
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.937400817871094,
"example_count": 100,
... |
2390e8e7-d175-435d-bc44-dd1ab1d9c62a | easy | DDanlov | 2026-08-09 00:13:08.796543+00:00 | failed | 02768d0b0db4f8c8423d74abb4a0ceb98ab15b665f725b0b78ac23a29bf7cf40 | 15,978 | 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
class RohanShampoo(Optimizer):
"""Self-contained RohanShampoo optimizer with eigenvalue ma... | {
"id": "2390e8e7-d175-435d-bc44-dd1ab1d9c62a",
"created_at": "2026-08-09 00:13:08.796543+00:00",
"db_md5": "54f00bb7f2e86084aad8574abdf4e26e",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "b5c34aa2-c8f1-41c1-84f3-0ce2135c9abd",
"tier": "easy",
"dataset_id": "e5",
"status": "failed",
... | null
|
23982aa2-5e2f-43d3-a282-b2a70c0ba208 | easy | 0xDEADD06 | 2026-08-12 14:25:56.521895+00:00 | succeeded | 652de6f6dbeae63b555d612964c39672a74dde859906419c5ba7375d33f43ba3 | 5,741 | null | """ParaLSTM: recurrent LSTM with weight-tied passes for algorithmic tasks."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import nn, Tensor
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
)
D_MODEL... | {
"id": "23982aa2-5e2f-43d3-a282-b2a70c0ba208",
"created_at": "2026-08-12 14:25:56.521895+00:00",
"db_md5": "185fbfdf4a87a60b00ef4dc3a0f39ca6",
"submitter": "0xDEADD06",
"github_login": "0xDEADD06",
"run_id": "35b1349f-45da-4918-bd97-658052724097",
"tier": "easy",
"dataset_id": "e5",
"status": "succee... | {
"score": {
"mean_loss": 2.7220749616227415,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.002083333395421505
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.943257444776227,
"example_count": 600... |
239972f1-c75b-44fe-9170-3f4defb32ea4 | easy | alirezashirvani-jr | 2026-08-29 05:04:33.682014+00:00 | succeeded | 102318215ba49333fe0733d02b30f4d5c680e7cd69ca25d7f33af3af554abe9d | 37,468 | 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": "239972f1-c75b-44fe-9170-3f4defb32ea4",
"created_at": "2026-08-29 05:04:33.682014+00:00",
"db_md5": "a1b092bde5c37f7c474450df07abd76b",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "915373fd-aaca-4d3d-82b2-67e330165fc2",
"tier": "easy",
"dataset_id": "e5",
... | {
"score": {
"mean_loss": 7.101005212076613,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0079166666790843
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.126977920532227,
"example_count": 600,
... |
239e76f9-e621-4f13-8033-940cfb348875 | easy | erdavis0 | 2026-08-18 03:46:00.003140+00:00 | succeeded | 8f4c29ffe858078f302e4246a7e0a5849de2651e5abdcee42cbc78d6df517992 | 34,559 | null | """Canonical token-register transition learner, version 9.
The model deliberately learns token semantics and field roles. Its recurrent
cell receives only a canonical categorical state and static context, while the
wall-clock curriculum controls how many copies of that same cell are composed.
Version 9 retains the c... | {
"id": "239e76f9-e621-4f13-8033-940cfb348875",
"created_at": "2026-08-18 03:46:00.003140+00:00",
"db_md5": "482b05aabe8cc4061015eba869534a87",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "2430741e-fbf5-499c-80a6-2b1ce9334aee",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded",... | {
"score": {
"mean_loss": 2.225641130655944,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007500000012417634
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2298805892734785,
"example_count": 600... |
23a5a030-8108-4683-85b3-ca233467a664 | easy | ArpBansal | 2026-08-15 16:40:35.540850+00:00 | succeeded | dcb9c700e0273c0bec01ef898d054fb22e96691164e8566871c1bdc328ff45d1 | 13,963 | null | """Digit-bottleneck recurrent model for repeated modular squaring.
The learned arithmetic path is constrained to
state[0] = encode_digits(x)
proposal[k + 1] = f_theta(state[k], encode_digits(N))
state[k + 1] = digit_bottleneck(proposal[k + 1])
T is used only as the recurrence counter. Between task steps,... | {
"id": "23a5a030-8108-4683-85b3-ca233467a664",
"created_at": "2026-08-15 16:40:35.540850+00:00",
"db_md5": "b2f686ad839b4f04c1050640df96aa32",
"submitter": "Arpit Bansal",
"github_login": "ArpBansal",
"run_id": "9b90a1b7-fb73-4676-ad65-b0c3972f6d4a",
"tier": "easy",
"dataset_id": "e5",
"status": "suc... | {
"score": {
"mean_loss": 2.8200512885793323,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006666666679084302
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1628612750314278,
"example_count": 60... |
23ad1a14-d2bd-4858-930e-5c73d7daa2fe | easy | DDanlov | 2026-08-24 20:05:19.011210+00:00 | succeeded | f7433efc11eacf437a22780da11479746e8dfed3728fff45fc88add9315e4c66 | 49,566 | null | """
Unified Pure Skipless Prefix Deliberation Model + Transformer Readout Decoder
Architecture:
- Deliberation Core: dim=1536, num_heads=24, d_ff=6144 (Pure Skipless SO(2)^{D/2} Givens Recurrence)
- Token Layout: Recurrent loop processes ONLY [Prompt (L_p), 4 Pause Registers] (Zero target tokens in loop)
- Recurrent At... | {
"id": "23ad1a14-d2bd-4858-930e-5c73d7daa2fe",
"created_at": "2026-08-24 20:05:19.011210+00:00",
"db_md5": "119b9428393d593beb5ad02936c8d970",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "aaf5cd2c-51c2-458b-9071-4386da32d49e",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.3838722026638157,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006250000024835268
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.4071259167162293,
"example_count": 60... |
23ae911f-00e5-45e9-9138-735ebe37cbed | easy | sapient-sapiens | 2026-08-30 21:50:11.467486+00:00 | succeeded | e3f7597be5da38cd8042c1ff8f09538a09b805731a6a03ebbd8374a4dd981da8 | 17,155 | null | """Eligible E5 direct-Horner model with two latent S writes before every A write.
One approximately 30M-parameter Transformer block is tied across all calls.
At each of four Horner positions it advances the latent workspace twice while
holding A fixed, then consumes the resulting S to commit A. N and X are
immutable,... | {
"id": "23ae911f-00e5-45e9-9138-735ebe37cbed",
"created_at": "2026-08-30 21:50:11.467486+00:00",
"db_md5": "38834e710d1f57c4d73c97f97dfe7403",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "7487a907-5204-4035-a30f-14028eb802cf",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 2.138376599569541,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004166666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.137136352436425,
"example_count": 600,... |
23b3943a-431b-4b35-afb3-1d81fb29814a | easy | erdavis0 | 2026-08-21 14:30:00.729195+00:00 | succeeded | e37358a21622100ab57ead40071bc75817c1cd70ff5e1eca862fdf0ea3de1faf | 4,764 | null | """Single fused forward GRU augmented by an invariant global prompt summary."""
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... | {
"id": "23b3943a-431b-4b35-afb3-1d81fb29814a",
"created_at": "2026-08-21 14:30:00.729195+00:00",
"db_md5": "2ddc044b08096c26eb27e0bacf9f9213",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "4e9c7f27-da36-4424-b8e2-7b3733854d24",
"tier": "easy",
"dataset_id": "e4",
"status": "succeeded",... | {
"score": {
"mean_loss": 2.66018283367157,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004583333386108279
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.7544050216674805,
"example_count": 1200... |
23b4bfb2-6db1-4448-9012-19795ff2512d | easy | DDanlov | 2026-08-10 09:59:12.880887+00:00 | succeeded | 17b9b3f0539b476189b7aefe44b1fe6d215c2d37b8f5728655abba7af9a79627 | 16,079 | 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": "23b4bfb2-6db1-4448-9012-19795ff2512d",
"created_at": "2026-08-10 09:59:12.880887+00:00",
"db_md5": "98d2e87f6ed3cef29de5a4048b5ab0bb",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "2082f577-4b2a-4a28-89cd-d1bd5c27f948",
"tier": "easy",
"dataset_id": "e2",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.520103540088063,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.519028925591091,
"example_count": 300,
"exact... |
23b68599-7890-4cf9-82da-8ec619a3ddfa | easy | DDanlov | 2026-08-23 21:37:35.106033+00:00 | succeeded | 6846bff4e5d0e8547e2eeae086051f1cb6a2c5d4b967e59842808f3235865561 | 31,831 | null | """
Unified Hybrid Masked Deliberation Model (71.9M Parameters) - Tier 1 (Pure Skipless, No Cross-Attention)
Architecture:
- Scale: dim=2048, num_heads=32, d_ff=8192 (Exactly 71,889,920 parameters)
- Hybrid Prefix-Bidirectional + Causal Target Attention Mask
- Token Layout: [Prompt (L_p), 4 Pause Registers, Target Toke... | {
"id": "23b68599-7890-4cf9-82da-8ec619a3ddfa",
"created_at": "2026-08-23 21:37:35.106033+00:00",
"db_md5": "4b6623e0e1aaa5d218483bfc3f7410ad",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "e140d042-b3b6-4841-88e8-79281f90307a",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.355640501042928,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005833333333333334
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.3649525535480858,
"example_count": 600... |
23ba7d21-d0c8-4a78-87f7-873813f3065f | easy | christopherhwood | 2026-08-05 14:44:02.984170+00:00 | failed | ecd975fc2dd5844218091c85db3837b953bbf456830da205af067eed0985a9e6 | 12,938 | null | """Weight-tied scratchpad reasoner for repeated modular squaring.
This file is intentionally standalone: the official evaluator copies only this
file. The model is a learned iterative computer. It appends a trainable
scratch tape to the immutable prompt and applies the same learned transition
once per requested time... | {
"id": "23ba7d21-d0c8-4a78-87f7-873813f3065f",
"created_at": "2026-08-05 14:44:02.984170+00:00",
"db_md5": "94b3db12a3b5c21a57ba832f8f34245c",
"submitter": "Chris Wood",
"github_login": "christopherhwood",
"run_id": "060cb212-aab4-4766-a64f-3fbfbc902cae",
"tier": "easy",
"dataset_id": "e5",
"status":... | null
|
23bc6e70-091d-452c-af3c-fcee2ceca072 | easy | lpbb | 2026-08-12 08:21:12.651615+00:00 | succeeded | 82c5fcfcb061c1d63946be944ca6c9237a590679f41a8ba9519b9e87068197b6 | 24,149 | null | """DIGIT-STATE DISCRETE FLOW (v2) — diffusion-style quantized recurrence, e1.
v1 (submissions/agents/diffusion) put the recurrent state in a learned
codebook. Flaw found on paper: nothing ties step(enc(x)) to enc(f(x)), so
the code space can fragment into per-chain private paths and the coverage
argument (98-99% of e... | {
"id": "23bc6e70-091d-452c-af3c-fcee2ceca072",
"created_at": "2026-08-12 08:21:12.651615+00:00",
"db_md5": "d5450fefde068717ec0cc35f4da0b63c",
"submitter": "Lpbb",
"github_login": "lpbb",
"run_id": "b6b3a5ca-bf5a-4a6c-91b7-9d7cd63f1ff7",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
"s... | {
"score": {
"mean_loss": 1.4298076629638672,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.648333340883255
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.0205473899841309,
"example_count": 100,
... |
23c69221-9a13-4829-80ad-18c5b66c69af | easy | nikolageorgiev2000 | 2026-08-04 02:00:58.244492+00:00 | succeeded | 4fbc32e34da472bb8b82eab6477d6e20777ad18b6f61c1ed846753e5e96896dd | 8,615 | 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": "23c69221-9a13-4829-80ad-18c5b66c69af",
"created_at": "2026-08-04 02:00:58.244492+00:00",
"db_md5": "f5f08bd2e1da3b5b1fc6996586bf650c",
"submitter": "Nikola Georgiev",
"github_login": "nikolageorgiev2000",
"run_id": "f369ef80-3a43-4e48-9fd2-1671ed3fc9f0",
"tier": "easy",
"dataset_id": "e1",
"s... | {
"score": {
"mean_loss": 3.0706549882888794,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.2966666668653488
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.411609649658203,
"example_count": 100,
... |
23caa770-ff3b-4250-a6a7-0dfd04b6fa73 | easy | isaac0804 | 2026-08-06 15:42:39.950709+00:00 | succeeded | af65d70b03b96b0d25ed3ce19dd713901f0c98d21184752b8ec88c34f42a0ae0 | 9,267 | null | """E3 candidate #7: same-modulus batch cross-attention -- an in-context /
retrieval mechanism, structurally different from every N-conditioning
attempt tried so far.
Every prior attempt this session (e3_hyper_lora, e3_hard_routing,
e3_vq_bottleneck, e3_narrow_bottleneck) conditioned each row on N using only
that row's... | {
"id": "23caa770-ff3b-4250-a6a7-0dfd04b6fa73",
"created_at": "2026-08-06 15:42:39.950709+00:00",
"db_md5": "1940badf501ffeaad17ff7ffb6a410bf",
"submitter": "Isaac Yong",
"github_login": "isaac0804",
"run_id": "916b8b81-ccef-452d-bd8d-a23b59ec0c90",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 12.452964653836723,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009375000013969839
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 12.662464684280568,
"example_count": 80... |
23cb3099-8ffb-4866-bffa-fc402f0704f4 | easy | karanganesan | 2026-08-07 07:16:20.600667+00:00 | succeeded | 4b2ee265962c2f920dc807cc7d58afec28f8de3af9cb03d8fa6ede8e2504be6f | 25,600 | 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": "23cb3099-8ffb-4866-bffa-fc402f0704f4",
"created_at": "2026-08-07 07:16:20.600667+00:00",
"db_md5": "58c192efb47c66fbf11d58dfb4f0420a",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "589d589b-ae67-4ea2-98fe-6e9c1e06bdcf",
"tier": "easy",
"dataset_id": "e5",
"status": ... | {
"score": {
"mean_loss": 2.347777843475342,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004583333386108279
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.348510503768921,
"example_count": 600,... |
23d47ff9-a4ba-4775-8870-b9b0c924c16d | easy | c-nnr | 2026-08-04 12:50:17.775708+00:00 | succeeded | ffde187a65bea31693bd380d2dab0f746d491cbde86c37c451f3585859b3559b | 4,618 | null | """One Layer Deeper E2 round-two experiment: Small x/z/y TRM: two train cycles and eight evaluation cycles."""
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
... | {
"id": "23d47ff9-a4ba-4775-8870-b9b0c924c16d",
"created_at": "2026-08-04 12:50:17.775708+00:00",
"db_md5": "1f38fbf2a0e1dc6dfc5b090d3a5494d5",
"submitter": "c-nnr",
"github_login": "c-nnr",
"run_id": "f09ba5af-4e99-48e5-8615-a489a0759c43",
"tier": "easy",
"dataset_id": "e2",
"status": "succeeded",
... | {
"score": {
"mean_loss": 3.59578800201416,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0010416667209938169
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.902723789215088,
"example_count": 300,... |
23e06263-2a62-4677-8985-56b2d3237c04 | easy | khushidahi | 2026-08-03 05:36:46.813454+00:00 | succeeded | 106e286a7f40da358b0477b427ce31be6b3832ee968b7b3611ff6986936c3260 | 14,356 | null | """R6 structured decimal-workspace model for One Layer Deeper.
The model does not implement multiplication, modular reduction, or the public
recurrence. It uses the public tokenizer structure to build learned, right-
aligned decimal tapes for N, X, and T, then applies a shared neural transition
to a mutable work tape.... | {
"id": "23e06263-2a62-4677-8985-56b2d3237c04",
"created_at": "2026-08-03 05:36:46.813454+00:00",
"db_md5": "134cae990c04ff64ea905c79e1dfdde8",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "7777cc80-2315-429e-a62b-e772a7a3ef3c",
"tier": "easy",
"dataset_id": "e4",
"status": "succ... | {
"score": {
"mean_loss": 2.07889461517334,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005625000107102096
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.076965093612671,
"example_count": 1200,... |
23e18605-63e2-40fa-b7fe-8c8a8b6c0071 | easy | viridale | 2026-08-03 08:44:27.825252+00:00 | succeeded | 5449bba31f012b075dd009b6b4ada565f5189313f2d748397738a123cf7c330c | 126,493 | null | """Signed sparse quadratics plus four-locus source-edit transducers.
hypothesis: v130 closes the requested I/tau/F/O edit set but still omits
source-natural signed maps requiring simultaneous square, linear, and
constant coefficients. Append four sparse signed quadratic transition
bases without constructin... | {
"id": "23e18605-63e2-40fa-b7fe-8c8a8b6c0071",
"created_at": "2026-08-03 08:44:27.825252+00:00",
"db_md5": "7f81c5b62a719ec6df76d60df36c24bd",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "55dc7299-6b60-4ea5-b985-6056d1018df7",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 0.003014546214626436,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.00301454615599774,
"example_count": 800,
"... |
23e2c962-b56c-410f-b411-5c154d9e0b16 | easy | himalalps | 2026-08-07 02:29:47.504970+00:00 | succeeded | 8a79d6546c0e38b0793d33750d7727243f903b15fed6f973a4b5fca3e56afc3a | 65,432 | null | """Algorithmically constrained latent recurrence for One Layer Deeper.
One masked cross-attention pass creates two independent representations:
``state_0`` from X only and an immutable context from N only. T is excluded
from both representations and is used solely to choose how many times the
learned recurrent MLP tr... | {
"id": "23e2c962-b56c-410f-b411-5c154d9e0b16",
"created_at": "2026-08-07 02:29:47.504970+00:00",
"db_md5": "cea68cc104c51bd8258f5243e52053ea",
"submitter": "Haoyu Tang",
"github_login": "himalalps",
"run_id": "2a0a69a5-586a-4c5f-b90c-7c63b8c77ec7",
"tier": "easy",
"dataset_id": "e5",
"status": "succe... | {
"score": {
"mean_loss": 5.976281541894685,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007500000052774946
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.962568283081055,
"example_count": 600,... |
23eaff89-acb1-4dfa-bd8e-2a5b5685ad58 | easy | sankalp1999 | 2026-08-04 11:35:14.520463+00:00 | succeeded | 4c96250b8293772a30be2bd04dc9e02c61d5edcc03cf8b6fdaf14cb5818d8398 | 4,869 | null | """Parametrized flattened-grid residual MLP for E1 sweeps."""
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,
assert_model_state,
)
... | {
"id": "23eaff89-acb1-4dfa-bd8e-2a5b5685ad58",
"created_at": "2026-08-04 11:35:14.520463+00:00",
"db_md5": "2db60cf366eebbface8061700d1fb8d0",
"submitter": "Sankalp ",
"github_login": "sankalp1999",
"run_id": "50341cd3-8229-4f6b-aa35-5b673aafc7d9",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 8.376094341278076,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.06000000052154064
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.841968536376953,
"example_count": 100,
... |
23fcf423-b55f-4843-9faa-1eea61af0b99 | easy | lzy54 | 2026-08-26 07:59:11.717472+00:00 | succeeded | 7bfa0118f2e8ae689773bfeb74b26e056e30e813728872817e6697881a4e286a | 3,786 | 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 = 128
NUM_HEADS = 4
... | {
"id": "23fcf423-b55f-4843-9faa-1eea61af0b99",
"created_at": "2026-08-26 07:59:11.717472+00:00",
"db_md5": "36525f55514567a93d01e3e0b380b39d",
"submitter": "lzy54",
"github_login": "lzy54",
"run_id": "a8b19192-ef03-4bcd-ba92-d2e70fddc91a",
"tier": "easy",
"dataset_id": "e6",
"status": "succeeded",
... | {
"score": {
"mean_loss": 5.240458011627197,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03851351607590914
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.746231555938721,
"example_count": 60,
... |
23fe554b-c89b-4737-b23f-7a306ab4c6cc | easy | c-nnr | 2026-08-04 12:26:17.155685+00:00 | succeeded | a49fcf76c4a3f31720ced8aa734f57abeac6d95304c0d00eb04ba62e2fe4c28d | 5,645 | null | """One Layer Deeper E2 round-two experiment: Official AdamW baseline control on E2."""
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 = 128
NUM_HEADS ... | {
"id": "23fe554b-c89b-4737-b23f-7a306ab4c6cc",
"created_at": "2026-08-04 12:26:17.155685+00:00",
"db_md5": "ca2842ad8e2c63460a61b9a3da0d8ab0",
"submitter": "c-nnr",
"github_login": "c-nnr",
"run_id": "a6b06d84-347d-4c8c-8295-44d6a86b4320",
"tier": "easy",
"dataset_id": "e2",
"status": "succeeded",
... | {
"score": {
"mean_loss": 2.5131802558898926,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009583333972841501
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.767179489135742,
"example_count": 300... |
24014953-0870-468d-a489-09e409508934 | easy | adwhit | 2026-08-13 13:32:37.787695+00:00 | succeeded | 5d6caf6e6dd5ab8df3e6e0827432c2eb37b7efef7d4ff4bcae8a3d596d7f414a | 17,860 | null | """DeepThinker v4a: depth-recurrent transformer with manual CUDA-graph capture.
The evaluation harness is kernel-launch-bound on unrolled-loop models
(measured on H100/Easy: eager 206 steps/60s with the GPU >95% idle;
torch.compile default -> 120 steps net; reduce-overhead -> crash). v4a
captures the ENTIRE training u... | {
"id": "24014953-0870-468d-a489-09e409508934",
"created_at": "2026-08-13 13:32:37.787695+00:00",
"db_md5": "bb2d4aa7a6a51e373d31d9a0b38b549b",
"submitter": "adwhit",
"github_login": "adwhit",
"run_id": "110ce1a5-9e7e-4774-8215-515b385e0b57",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
... | {
"score": {
"mean_loss": 2.4033267498016357,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03166666720062494
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.757272243499756,
"example_count": 100,... |
24019573-26df-4168-978f-1db118276cc8 | easy | viridale | 2026-08-26 01:45:14.983891+00:00 | succeeded | b89fe250b06c10da80899a8340245e8e2f85b2cdcde23d610fce55fcc5dda991 | 23,517 | null | """Predictive tied-depth latent-program Transformer.
hypothesis: v926 solves the frozen million-scale direct-T1 law through a forward
predictive program marginal, but selecting T1 rows alone cannot learn when training
supplies only deeper endpoints. Push each learned program draw through the requested
tied... | {
"id": "24019573-26df-4168-978f-1db118276cc8",
"created_at": "2026-08-26 01:45:14.983891+00:00",
"db_md5": "69156315fe57ea236b08c3ba7d8cc718",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "563fa543-d6bf-46d1-b89d-69384068e105",
"tier": "easy",
"dataset_id": "e6",
"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": 60,
"exact_accuracy": 1.0,
"c... |
240bddd8-f757-4020-9366-b60b75943b2b | easy | khushidahi | 2026-08-21 03:12:49.988951+00:00 | succeeded | f8dd5ede6de581025c60ab187982048c0c592c03404a712ffc28a22275d22814 | 15,504 | null | """R325: R323 control with smaller batches for more optimizer decisions.
The research claim is deliberately narrow: on fixed-modulus Easy E6, can the
published Abacus representation plus a maximally recurrent Transformer learn
the one-step modular-squaring transition on held-out operands?
McLeish et al.'s arithmetic ... | {
"id": "240bddd8-f757-4020-9366-b60b75943b2b",
"created_at": "2026-08-21 03:12:49.988951+00:00",
"db_md5": "af7e92ac53a08bf7b385d10ff6636f1d",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "5b71ff39-fe67-4d25-b99e-f7665ea7d492",
"tier": "easy",
"dataset_id": "e6",
"status": "succ... | {
"score": {
"mean_loss": 3.490508388166558,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.06351351803420363
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.460662841796875,
"example_count": 60,
... |
2417a293-e7d9-4920-9355-3c58c9581837 | easy | himalalps | 2026-08-07 01:36:45.089157+00:00 | succeeded | 580900e67e764e17ad9aacae1640b69398d834ceef6cc27356acddcb3190d67f | 65,546 | null | """Algorithmically constrained latent recurrence for One Layer Deeper.
One masked cross-attention pass creates two independent representations:
``state_0`` from X only and an immutable context from N only. T is excluded
from both representations and is used solely to choose how many times the
learned recurrent MLP tr... | {
"id": "2417a293-e7d9-4920-9355-3c58c9581837",
"created_at": "2026-08-07 01:36:45.089157+00:00",
"db_md5": "d955687cc4cd8cdd6da9f596ff8037a5",
"submitter": "Haoyu Tang",
"github_login": "himalalps",
"run_id": "f827ee29-5c99-4f4f-914e-2936f802c51e",
"tier": "easy",
"dataset_id": "e5",
"status": "succe... | {
"score": {
"mean_loss": 2.2159992233925574,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009583333221574624
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2313084602355957,
"example_count": 60... |
2425c1a6-30ea-49f6-b5a0-d8aacacc8f80 | easy | anirudh-chakravarthy | 2026-08-12 03:24:22.901205+00:00 | succeeded | ba94bf16495996ba4a73dbb98afd8b888fe2a0004f786b31daaf36eee703a8d2 | 14,572 | null | """A learned state transition applied once per requested time step."""
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,
asser... | {
"id": "2425c1a6-30ea-49f6-b5a0-d8aacacc8f80",
"created_at": "2026-08-12 03:24:22.901205+00:00",
"db_md5": "a06bc017fc24953f9f636c3f3cfd620e",
"submitter": "Anirudh S Chakravarthy",
"github_login": "anirudh-chakravarthy",
"run_id": "21a42eb8-c329-4813-8566-5d8837d5dc35",
"tier": "easy",
"dataset_id": "... | {
"score": {
"mean_loss": 4.318755785161855,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.010000000037252904
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.344288077589642,
"example_count": 600,... |
242f27ba-5e3d-4eab-b4ab-43d14e64e62f | easy | khushidahi | 2026-08-21 03:27:53.915800+00:00 | succeeded | dd7cea57211fa121567db11fc323223058555abfaad4b681b6a9b3eb04a57b97 | 15,424 | null | """R330: R326 with evaluation-matched deterministic recurrent state.
The research claim is deliberately narrow: on fixed-modulus Easy E6, can the
published Abacus representation plus a maximally recurrent Transformer learn
the one-step modular-squaring transition on held-out operands?
McLeish et al.'s arithmetic mode... | {
"id": "242f27ba-5e3d-4eab-b4ab-43d14e64e62f",
"created_at": "2026-08-21 03:27:53.915800+00:00",
"db_md5": "43a85ad3381ae2c3d4be2772a7ca81ef",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "836fadef-ba8d-48e0-ab7a-752515fd791b",
"tier": "easy",
"dataset_id": "e6",
"status": "succ... | {
"score": {
"mean_loss": 7.573250313327737,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.060135139605483494
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.620667934417725,
"example_count": 60,
... |
242f9320-edf4-4d07-8a1e-6a70764abe44 | easy | sapient-sapiens | 2026-08-30 22:00:42.923912+00:00 | succeeded | 5bb2b36a7ea15c10a362d73b84b58663d59e0f9ada6a9b389c1084f168b81002 | 16,264 | null | """Eligible E5 repeated digit-content Horner operator with strict S-to-A routing.
The tied S operator receives full immutable N/X, current A/S, and the current
digit content in one fixed role lane. It receives no step index or step tag;
the loop iterates over digit contents directly. The A operator receives only
A/S... | {
"id": "242f9320-edf4-4d07-8a1e-6a70764abe44",
"created_at": "2026-08-30 22:00:42.923912+00:00",
"db_md5": "0b1fdb75e9f8ddfbc1c7f09ca20ac2e0",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "c0ba3a20-65d9-4536-a093-2beb1ee9456c",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 2.1514369747906397,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005833333345750968
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.149328834807392,
"example_count": 600... |
2437d5e3-37e6-4e77-8c1d-c1e4caaf7fcd | easy | chad-atexpedient | 2026-08-04 22:14:06.284569+00:00 | succeeded | c9031e85e2c9ee52518b9dab7ec3601709bf11369bc4ea24961f603a83d58c81 | 6,826 | null | """R019: one-pass RoPE Transformer with eight position-agnostic latent slots."""
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,
)
WIDT... | {
"id": "2437d5e3-37e6-4e77-8c1d-c1e4caaf7fcd",
"created_at": "2026-08-04 22:14:06.284569+00:00",
"db_md5": "1f2dbb2651b10abc8562dc16b87cafad",
"submitter": "chad-atexpedient",
"github_login": "chad-atexpedient",
"run_id": "b1237d3f-020e-444f-819e-fd07d5939740",
"tier": "easy",
"dataset_id": "e5",
"st... | {
"score": {
"mean_loss": 4.20269264336029,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006250000024835269
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.673482798674716,
"example_count": 600,
... |
243bcd82-8ba0-494c-8051-58ed508d8518 | easy | KaustubhKumar05 | 2026-08-25 16:08:52.309443+00:00 | succeeded | 1c97e6ade7b7c4704016b0ee590b894d107a7dd895518e2d223d26813d2c9766 | 13,655 | 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": "243bcd82-8ba0-494c-8051-58ed508d8518",
"created_at": "2026-08-25 16:08:52.309443+00:00",
"db_md5": "f9a601f96038214772004fc5576b07aa",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "6a34f2d4-568d-4b9c-9217-60d19006e94b",
"tier": "easy",
"dataset_id": "e8",
"status": "succee... | {
"score": {
"mean_loss": 5.552034616470337,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.22281640768051147
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.965521812438965,
"example_count": 85,
... |
244285c2-7b89-4cd0-8816-017e7e914853 | easy | KaustubhKumar05 | 2026-08-27 03:32:00.078771+00:00 | succeeded | b3b819e524ac25c61449344d380f2aeb54a7d3c5a37c44bff7d3d80354abf1a1 | 13,095 | null | """abacus_wdc — candidate B4: B at the reported critical weight decay.\n\nOnly weight_decay changes: 0.05 (Muon group) and 0.1 (AdamW group) both -> 0.0158.\nVerma (arXiv 2605.20441) reports weight decay acting as a scalar control\nparameter with three regimes -- memorization, generalization, and COLLAPSE --\nand local... | {
"id": "244285c2-7b89-4cd0-8816-017e7e914853",
"created_at": "2026-08-27 03:32:00.078771+00:00",
"db_md5": "a18cccef56929d5c6cd9f9edfa23816a",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "b2081a93-f9e1-4a16-b455-ff141469ae0f",
"tier": "easy",
"dataset_id": "e3",
"status": "succee... | {
"score": {
"mean_loss": 10.03135422059879,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.011875
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 10.0382527795518,
"example_count": 800,
"e... |
24435fc4-cd1a-41ea-81b7-0d3edba678d0 | easy | chuk-uzowihe | 2026-08-12 19:31:11.381016+00:00 | succeeded | a8d63a3f01cdab77e23a22c8f610db1477e7678843856425a7dde293ca9afd3b | 26,030 | null | """v10: dense fixed-loop recurrence with deep supervision.
Lineage: FPRM structure (alpha/beta-tied residual scaling, pre-rms_norm,
input injection every iteration) + the v8 2D digit lattice / RoPE task
encoding. Three deliberate departures from the v9 MoEUT line:
1. **Full BPTT.** The loop runs a fixed N_ITER iterat... | {
"id": "24435fc4-cd1a-41ea-81b7-0d3edba678d0",
"created_at": "2026-08-12 19:31:11.381016+00:00",
"db_md5": "773488d9294224f12b95e81a4dfb1f4d",
"submitter": "Chuk Uzowihe",
"github_login": "chuk-uzowihe",
"run_id": "44c76c2f-b874-4e50-8324-1c21dd47c5aa",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 33.90426540374756,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0833333320915699
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 37.74502182006836,
"example_count": 100,
... |
24445e4d-3ec9-4dc5-9510-835396ee0fce | easy | DDanlov | 2026-08-09 00:49:46.520548+00:00 | succeeded | 2b39027ad89a3ca20516f816b05c71f03e4804da979089620da593dfe585eed7 | 15,981 | 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
class RohanShampoo(Optimizer):
"""Self-contained RohanShampoo optimizer with eigenvalue ma... | {
"id": "24445e4d-3ec9-4dc5-9510-835396ee0fce",
"created_at": "2026-08-09 00:49:46.520548+00:00",
"db_md5": "5f4617b3d98c01ac509187eab29e4c79",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "39592253-becf-4348-ad13-c998bbf28658",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 133.53384437778908,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0004166666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 133.96552122654936,
"example_count": 6... |
24451bb4-c1e8-4f83-93bb-78fbb911d3aa | easy | erdavis0 | 2026-08-26 03:51:31.309665+00:00 | succeeded | f323ea27a95b25c17ef240c64fd37d3915d7f3e1d86c503b6dc262e4235c4217 | 30,352 | null | """Latent-role continuous REV C1: selector-oracle distillation.
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 first e... | {
"id": "24451bb4-c1e8-4f83-93bb-78fbb911d3aa",
"created_at": "2026-08-26 03:51:31.309665+00:00",
"db_md5": "1658fcec88828e061d8640031c475500",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "c74b47c1-9fd4-48cd-988f-51043e2481b6",
"tier": "easy",
"dataset_id": "e4",
"status": "succeeded",... | {
"score": {
"mean_loss": 2.2581273235264825,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005625
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2633019094747517,
"example_count": 1200,
... |
24463201-eb83-4096-94bc-feeeed5133d1 | easy | DDanlov | 2026-08-11 19:32:53.638126+00:00 | failed | bbf4b74e2860dd6b5455e7048884898afde86c279c7c65c1b9dc6536f971f366 | 19,071 | 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
class RohanShampoo(Optimizer):
"""Self-contained RohanShampoo optimizer with bias-correcte... | {
"id": "24463201-eb83-4096-94bc-feeeed5133d1",
"created_at": "2026-08-11 19:32:53.638126+00:00",
"db_md5": "dbb47a483640cada0c86336a1ee83963",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "af6d5ae5-9a9e-4a0b-be55-e2eefdf697ea",
"tier": "easy",
"dataset_id": "e1",
"status": "failed",
... | null
|
2447c7da-5d22-4253-88d1-c21e78896762 | easy | alirezashirvani-jr | 2026-08-31 00:48:10.388261+00:00 | succeeded | d30c15c4b819074a695cfe7f9ab43fd9530cdf8d2f612f55533983ffcf4e7dfb | 22,445 | null | from __future__ import annotations
import math
import time
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_model_state,
)
PERIOD_MINIMUM = 2
PERIOD_MAXIMUM = 96
EXP... | {
"id": "2447c7da-5d22-4253-88d1-c21e78896762",
"created_at": "2026-08-31 00:48:10.388261+00:00",
"db_md5": "a726edbe0af9de151bd033a13eebcee4",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "c085f048-600e-42f5-8626-f20201c24c5c",
"tier": "easy",
"dataset_id": "e4",
... | {
"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": 1200,
"exact_accuracy": 1.0,
... |
2448e8aa-b11c-4559-9860-981191a350c0 | easy | ehonig | 2026-08-20 17:59:39.630685+00:00 | succeeded | cafed339bec368ebff3c5a39cf089f29824bf2f77fbf22ab70455b17a2c65b48 | 25,078 | 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": "2448e8aa-b11c-4559-9860-981191a350c0",
"created_at": "2026-08-20 17:59:39.630685+00:00",
"db_md5": "1a3143490b22a2defefb48e3b67014e2",
"submitter": "Edouardo Honig",
"github_login": "ehonig",
"run_id": "254329c0-688f-4396-8277-43da08b80e75",
"tier": "easy",
"dataset_id": "e1",
"status": "succ... | {
"score": {
"mean_loss": 2.191460132598877,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.014999999664723873
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.134831190109253,
"example_count": 100,... |
244d2f5e-2dfe-4d1d-9538-94373f569459 | easy | himalalps | 2026-08-06 12:21:29.681056+00:00 | succeeded | caf12a74991f36892686ee67d5e1f244116c8f09df880eed04c910a65b51c908 | 53,864 | null | """Algorithmically constrained latent recurrence for One Layer Deeper.
One masked cross-attention pass creates two independent representations:
``state_0`` from X only and an immutable context from N only. T is excluded
from both representations and is used solely to choose how many times the
learned recurrent MLP tr... | {
"id": "244d2f5e-2dfe-4d1d-9538-94373f569459",
"created_at": "2026-08-06 12:21:29.681056+00:00",
"db_md5": "c05d5c6a80e6dcf561f29d2b28719dfd",
"submitter": "Haoyu Tang",
"github_login": "himalalps",
"run_id": "aa394a5c-3233-4433-a75b-a779d076adc6",
"tier": "easy",
"dataset_id": "e5",
"status": "succe... | {
"score": {
"mean_loss": 2.1677449982579766,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.010000000229726235
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.177643299102783,
"example_count": 600... |
24583751-2179-41e8-8d6c-153d45ceb937 | easy | DDanlov | 2026-08-27 22:41:02.898855+00:00 | succeeded | bdd87667fe62aab72df659417d17171a169068995eece75c1ff40dc68fdb2f2f | 40,876 | null | import math
import time
from dataclasses import dataclass
from typing import Optional, Tuple, Dict, Any, List, Union
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Optimizer
try:
from onelayerdeeper.api import Submission, OptimizerBundle, ModelSpec, OptimizerSpec
except ... | {
"id": "24583751-2179-41e8-8d6c-153d45ceb937",
"created_at": "2026-08-27 22:41:02.898855+00:00",
"db_md5": "8d20de342c25334afd1eada318f7ad28",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "9b588cb4-afd6-4cae-9650-0eddc710292e",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.55464506149292,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0025000000605359674
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.5650806427001953,
"example_count": 600... |
245d62a5-8e31-4f13-8651-2cd371a6c335 | easy | DDanlov | 2026-08-26 20:42:40.786391+00:00 | succeeded | 8d0972715660cc522ea742c80cb60e71c9026ea9b8d11cab6c7e04a95da97925 | 34,343 | null | import math
import time
from dataclasses import dataclass
from typing import Optional, Tuple, Dict, Any, List, Union
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Optimizer
try:
from onelayerdeeper.api import Submission, OptimizerBundle, ModelSpec, OptimizerSpec
except ... | {
"id": "245d62a5-8e31-4f13-8651-2cd371a6c335",
"created_at": "2026-08-26 20:42:40.786391+00:00",
"db_md5": "a4d6119f2ae1ad3813ba8ff4eb1af660",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "aff98873-45be-4dbf-82ff-83853c95bcc7",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 7.905915622563416,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004583333333333333
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.934739713711589,
"example_count": 600,... |
24612d46-a904-44ca-a946-8cebfa43b7bf | easy | Ranoobaba | 2026-08-17 22:47:18.836792+00:00 | succeeded | 5ede32a90522f01aac66b095b32bc29555d982e721140caeab3422e12c558388 | 5,856 | 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": "24612d46-a904-44ca-a946-8cebfa43b7bf",
"created_at": "2026-08-17 22:47:18.836792+00:00",
"db_md5": "ba2327560f8e45986ad7d9c3435af30d",
"submitter": "Syed Rayyan Ali",
"github_login": "Ranoobaba",
"run_id": "25d13b10-66a3-4239-b6b1-1b96c312315e",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 6.345899397001005,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.053333332762122154
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.5018792152404785,
"example_count": 100... |
2462fd69-585a-4d8f-b8c9-278828406479 | easy | velocizapkar | 2026-08-30 16:38:39.202936+00:00 | succeeded | 5c8ee47bbc801493bd7ba5eb52143381406519f108fd1d2986fa986067fa1ac7 | 7,486 | null | """HeightGrid: a (sequence x depth x height) transformer for One Layer Deeper.
The computation is a 3D grid. Each cell (t, d, h) holds a width-W vector and
receives three interactions:
* sequence: softmax attention over cells (., d, h) in the same plane;
* height: the running state ascending from (t, d, h-1);
... | {
"id": "2462fd69-585a-4d8f-b8c9-278828406479",
"created_at": "2026-08-30 16:38:39.202936+00:00",
"db_md5": "60f12ad7eb595dacf72033c30dfd11dd",
"submitter": "Aakanksh Zarapkar",
"github_login": "velocizapkar",
"run_id": "c312f313-3447-492c-a35e-85a581a66288",
"tier": "easy",
"dataset_id": "e6",
"statu... | {
"score": {
"mean_loss": 2.0619055032730103,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.02342342445626855
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1770942211151123,
"example_count": 60,... |
24675af2-d9cd-4be8-a0ba-25b122d51bba | easy | jayyala | 2026-08-18 13:11:17.196978+00:00 | succeeded | e6c747c86aa58869f964c195b1efe51c52ccb8bbc525e3613aff1138a5113d25 | 24,443 | null | """One N-equivariant step: wrap a learned bit-pair product, then invert.
exp(2*pi*i*k/N) depends only on k mod N, so the integer product wraps into
the reduced residue. That wrapping is the transition, reused T times. A
tied coarse-to-fine decoder inverts the phase; it is not a table of (N, x).
Prompt reconstruction... | {
"id": "24675af2-d9cd-4be8-a0ba-25b122d51bba",
"created_at": "2026-08-18 13:11:17.196978+00:00",
"db_md5": "4da96419acfd905f77442ec1acb5a1d0",
"submitter": "jayyala",
"github_login": "jayyala",
"run_id": "27da9f07-c2f2-498f-a3f8-c4962dc5f96b",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded"... | {
"score": {
"mean_loss": 4.244229332859761,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.8612500000000001
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.940260866283117,
"example_count": 800,
... |
2467fb93-93ab-4453-978b-da8fcb44b196 | easy | jordanrubin | 2026-08-05 18:26:13.527628+00:00 | succeeded | ef83786e8b5d6b9bde5ee341b69d241cfe4f15d944d01fd186a692875dfc2a67 | 14,276 | 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": "2467fb93-93ab-4453-978b-da8fcb44b196",
"created_at": "2026-08-05 18:26:13.527628+00:00",
"db_md5": "5d8a00fb87b44fa3031bccc04f6a5780",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "d3e6b781-41b9-49b0-acc9-ff06f27ecf6b",
"tier": "easy",
"dataset_id": "e2",
"status": "s... | {
"score": {
"mean_loss": 2.3854598411272576,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.015624999650754035
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2323610131887333,
"example_count": 30... |
247ca834-0be6-40d5-a220-f5b37b0f8e59 | easy | sapient-sapiens | 2026-08-21 21:27:11.910765+00:00 | succeeded | a35dff741d0ede7a43da0c4067565fd8791af541056b144a96ba3b8b3d235395 | 33,481 | 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": "247ca834-0be6-40d5-a220-f5b37b0f8e59",
"created_at": "2026-08-21 21:27:11.910765+00:00",
"db_md5": "6cb41323ec4eea3acf3b5d4a4e3c2ab7",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "5beabc4e-0cd0-4c3c-8601-0677b3e374fb",
"tier": "easy",
"dataset_id": "e9",
"status": "su... | {
"score": {
"mean_loss": 5.463748181196545,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.02666666763524214
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.750673294067383,
"example_count": 90,
... |
247ef5f6-3239-4b03-9cae-b6e14258644d | easy | yashkant | 2026-08-14 17:35:12.307457+00:00 | succeeded | 6c53cfae9b9dc392b14016b07cbf7adc9d7833723c46ae7691fa404596548965 | 42,050 | null | """Round1816 parent-preserving cyclic residue tensor.
The complete Round1792 AdamW5/EMA.995 carrier remains byte-identical at a
matched seed. New-module initialization is isolated from the inherited CPU
RNG stream, and an exact-zero scalar gate keeps the scored path equal to the
parent until terminal CE opens the sea... | {
"id": "247ef5f6-3239-4b03-9cae-b6e14258644d",
"created_at": "2026-08-14 17:35:12.307457+00:00",
"db_md5": "ea5d985f1a28649deea946f3e8c11b5b",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "f7fff863-b621-4a3f-9ab5-5ac7e4e0bf01",
"tier": "easy",
"dataset_id": "e3",
"status": "succeed... | {
"score": {
"mean_loss": 2.1234401454770877,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.017500000158324838
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.122389782864254,
"example_count": 800... |
24838fed-1d11-40fe-ad4c-53a34ecec49f | easy | erdavis0 | 2026-08-27 00:54:54.980099+00:00 | succeeded | 795df3f9652030e991e92998273fd7a4939c775c6b6c6c9395c3028f971d4e8b | 31,304 | null | """Typed-prefix reversible transition B4: oracle-weighted step training.
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": "24838fed-1d11-40fe-ad4c-53a34ecec49f",
"created_at": "2026-08-27 00:54:54.980099+00:00",
"db_md5": "86ddb36d905f79e81813767c4d60fc8f",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "d0b338cd-c39f-4fc9-aa09-a168d456d926",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded",... | {
"score": {
"mean_loss": 2.683643082941251,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009583333358168602
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.721094788991817,
"example_count": 600,... |
248c3b3b-d647-4b5e-94d3-fbd5b1ae6b8d | easy | christopherhwood | 2026-08-06 07:37:37.897411+00:00 | succeeded | 1ec0746efdf4d97028eb91d0cccf2782139c79be7e03d533c83f7a8cceaa8718 | 22,423 | null | """Grounded global-program register machine for repeated modular squaring.
Every example executes the same short learned program. A program phase makes
only observable generic choices: which canonical tape to read twice, which
canonical tape to write, and which direction to scan. One finite-state local
cell is tied ... | {
"id": "248c3b3b-d647-4b5e-94d3-fbd5b1ae6b8d",
"created_at": "2026-08-06 07:37:37.897411+00:00",
"db_md5": "7681afdb1f4832ff65dcf0099949ae9e",
"submitter": "Chris Wood",
"github_login": "christopherhwood",
"run_id": "dfbb9c29-07a7-4c8f-9d43-1fc40f39d83c",
"tier": "easy",
"dataset_id": "e5",
"status":... | {
"score": {
"mean_loss": 2.2956860427425743,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007500000037252903
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.3043038021823214,
"example_count": 60... |
248d52b9-6be8-434c-9266-6d0e90a9573b | easy | sapient-sapiens | 2026-08-16 20:20:37.893726+00:00 | succeeded | 0c81e73a06a5d396dc8038768738c16fe7af9c95105a9f4ccbd3bea0b77a3dbb | 21,218 | null | """E6 SAM-rho probe: frozen v55 with SAM_RHO=0.05.
The only change relative to act64_e6_w1024_dh4_L6_v55 is a larger SAM
perturbation radius (0.02 -> 0.05). T is diagnostic-only.
"""
from __future__ import annotations
import math
import time
import torch
import torch.nn.functional as F
from torch import Tensor, nn
... | {
"id": "248d52b9-6be8-434c-9266-6d0e90a9573b",
"created_at": "2026-08-16 20:20:37.893726+00:00",
"db_md5": "6285acf7bce3b74ceae23213f3639666",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "ac6855e8-123e-4b78-99b7-bbbd815964b0",
"tier": "easy",
"dataset_id": "e6",
"status": "su... | {
"score": {
"mean_loss": 2.9774999128629083,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.960388422012329,
"example_count": 60,
"exact... |
24920c22-a003-41a4-8bb0-4df02617a488 | easy | liam-gb | 2026-08-13 13:12:34.581153+00:00 | succeeded | f95e6bbf65c5dc56d5416ecc07dde324f0f2e00e4471a97c51ee6e8969d47ec1 | 20,183 | null | """rns_pm_gather_ct_lift: a doubling loop trained by its own semigroup law.
Changes vs hardA. The counted loop is replaced by BINARY LIFTING: seven level
operators B_0..B_6, with B_k standing for the 2^k-fold iterate, applied once for
each set bit of T. And the loop states are additionally trained by a label-free
cons... | {
"id": "24920c22-a003-41a4-8bb0-4df02617a488",
"created_at": "2026-08-13 13:12:34.581153+00:00",
"db_md5": "17189dd642d43a6655afbe44d27e3dc1",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "e09c6e1b-2fa0-4be9-89e4-5db938c15f28",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 6.170367198283071,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05708333317190409
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.575867652893066,
"example_count": 600,
... |
2496b04f-9a15-4027-96d3-5ce027093b0d | easy | ArkAung | 2026-08-04 13:20:56.625638+00:00 | failed | b47fa5a6bbedaa7e3f41976b89eaf4dd79f1d329d92e417e35d9f5443d70bab7 | 9,118 | null | import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_model_state,
)
# ---------------------------------------------------------------------------
# Config
# -------... | {
"id": "2496b04f-9a15-4027-96d3-5ce027093b0d",
"created_at": "2026-08-04 13:20:56.625638+00:00",
"db_md5": "14ca4bb4fdf14913cbba587ab04b53d1",
"submitter": "Arkar Min Aung",
"github_login": "ArkAung",
"run_id": "aea1e824-a62d-4272-b3c6-409081b2a977",
"tier": "easy",
"dataset_id": "e1",
"status": "fai... | null
|
249d0236-298f-4403-bef6-750373bfee10 | easy | DDanlov | 2026-08-10 07:53:39.481002+00:00 | succeeded | 742aa178c8f34e0df18080215389c3a77fc7507be7138cf63c986ebc254c2449 | 16,672 | 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 initialization to ensure initial Jaco... | {
"id": "249d0236-298f-4403-bef6-750373bfee10",
"created_at": "2026-08-10 07:53:39.481002+00:00",
"db_md5": "5c9396a9e78cad54a98e0b4a56c30ec9",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "b34a8f9e-c2c6-4bf7-bffc-02d7b7ef25ca",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 4.373632324604024,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.053333330353101095
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.495085716247559,
"example_count": 100,... |
249f2737-b2ee-44b4-8848-7bbd077a0572 | easy | khushidahi | 2026-08-22 18:30:39.192175+00:00 | succeeded | 2ac724ec0859f589ade0a598d902cdb7296fd695e6767fb63db4de38fc1fe28c | 31,426 | null | """R399: post-normalized Universal Transformer with internal linear recall.
This submission tests the strongest rules-safe generic baseline suggested by
the public competition discussion. It does not implement squaring, modular
reduction, or any other task solver. Numeric prompt fields are only rearranged
onto equal... | {
"id": "249f2737-b2ee-44b4-8848-7bbd077a0572",
"created_at": "2026-08-22 18:30:39.192175+00:00",
"db_md5": "b0e56c45312b5b1c16e80dc48e19a226",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "7e98ea5e-415e-4600-a94c-bbed66e1af0a",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.6455170412477567,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004583333358168602
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.648225756504076,
"example_count": 600... |
24a530fb-54e8-4aa8-87aa-a031be46b669 | easy | karanganesan | 2026-08-07 08:14:52.996924+00:00 | succeeded | b35e0163af386136b0b282f874ae269528427964389cc99eca2d29126c20bdb6 | 25,600 | 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": "24a530fb-54e8-4aa8-87aa-a031be46b669",
"created_at": "2026-08-07 08:14:52.996924+00:00",
"db_md5": "08d1e335b2fc6d9fbe7cc2804f460670",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "ab2152e6-2be8-47e4-b15a-70442ff053b8",
"tier": "easy",
"dataset_id": "e5",
"status": ... | {
"score": {
"mean_loss": 3.365779399871826,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0033333332976326346
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.546499729156494,
"example_count": 600... |
24a9027d-195c-435d-a623-4a0bbed5a02f | easy | DDanlov | 2026-08-30 00:17:54.289461+00:00 | failed | a447c87bf78aded843388b4f56d834ab7afe0ae1f34193416bf7cbf40053d792 | 28,426 | null | """
Unified Continuous Hopfield Equilibrium Propagation Deliberation Engine
=======================================================================
Architecture:
- Deliberation: Continuous Hopfield Multi-Layer Energy System
- Integration: In-Place Strang-Split Conformal Symplectic Velocity Verlet (with LaSalle Resets)
... | {
"id": "24a9027d-195c-435d-a623-4a0bbed5a02f",
"created_at": "2026-08-30 00:17:54.289461+00:00",
"db_md5": "69ab29d95093168673628409b454acb7",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "91ebdcf6-014a-4f7a-99a2-e0865cca6617",
"tier": "easy",
"dataset_id": "e5",
"status": "failed",
... | null
|
24ab039f-8fbf-4067-9ba3-1befc8d443d3 | easy | AFeinfield | 2026-08-19 23:51:34.889080+00:00 | succeeded | 5bc0ecc31e8b6b50986d6f46bbd85c3654b5a82dd35ef6b6f0262326a8c2a89b | 19,376 | null | """092: INSTRUMENTED BASE — two measurement fixes, no architecture change.
(1) Step-indexed LR schedule with a fixed screening step budget. The old
schedule decayed on WALL CLOCK, so throughput jitter changed the
LR-per-step trajectory: identical files on e2 gave 1470 vs 1025 steps
and train acc 0.725 vs 0... | {
"id": "24ab039f-8fbf-4067-9ba3-1befc8d443d3",
"created_at": "2026-08-19 23:51:34.889080+00:00",
"db_md5": "401a4eb2e300d91e9f14cc8a470105dc",
"submitter": "AFeinfield",
"github_login": "AFeinfield",
"run_id": "f71ab5c4-eb81-4ad1-a645-2266a378841b",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.1532120380100785,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00583333338300387
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1614393150859885,
"example_count": 600... |
24cad0b8-95c0-486b-8d97-478a99cd143c | easy | viridale | 2026-08-28 12:43:31.584867+00:00 | succeeded | 25b1883751cd4406d6b3ab706b6f5b66931f5cdfd1f3b0707619174f58c4f0ee | 33,469 | null | """Globally learned quadratic-program evidence slab.
hypothesis: E5's few 10-bit modulus identities repeatedly trigger the same-N branch,
allowing the N-conditioned controller to learn modular aliases: latent evidence falls
below .001 while full-batch hard accuracy stays exactly one bit-setting out of two.
... | {
"id": "24cad0b8-95c0-486b-8d97-478a99cd143c",
"created_at": "2026-08-28 12:43:31.584867+00:00",
"db_md5": "a81ce9296f55a044e1ab64b02b2dc26e",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "31e01436-a9a2-4086-940b-9505452b5bc0",
"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,
"... |
24e77426-b66b-4620-a27e-63858884758c | easy | jordanrubin | 2026-08-03 20:41:21.233693+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": "24e77426-b66b-4620-a27e-63858884758c",
"created_at": "2026-08-03 20:41:21.233693+00:00",
"db_md5": "ef7d67d872a65aa96e81ca53991ab27d",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "4309dc01-6e54-47db-8d28-0f861bbf684f",
"tier": "easy",
"dataset_id": "e2",
"status": "s... | {
"score": {
"mean_loss": 3.6584284979518635,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.20062499454555413
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.692923481040715,
"example_count": 300,... |
24e85235-bfcc-42b5-b5e1-19671a84e3a5 | easy | lzy54 | 2026-08-27 05:28:08.840005+00:00 | succeeded | ab2dc1ee24b4fecae66530c2b8ab19a4db6a965bfc7d6c03e18e0a7d67ed845f | 6,884 | null | """B2 plus weak supervision at the sampled endpoint's penultimate 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,
TokenLossBatch,
assert_model_... | {
"id": "24e85235-bfcc-42b5-b5e1-19671a84e3a5",
"created_at": "2026-08-27 05:28:08.840005+00:00",
"db_md5": "b9b46e0dc16403cca0598be80994351e",
"submitter": "lzy54",
"github_login": "lzy54",
"run_id": "4e6faa30-8079-4faf-b860-3fe402cba3e1",
"tier": "easy",
"dataset_id": "e6",
"status": "succeeded",
... | {
"score": {
"mean_loss": 2.5376970767974854,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04842342482879758
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.4251060485839844,
"example_count": 60,... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.