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 |
|---|---|---|---|---|---|---|---|---|---|---|
39681aea-aea1-477a-be81-b9e1bb7c3045 | easy | blake-camp-surge | 2026-08-05 17:57:02.907579+00:00 | succeeded | 776923fcf83ca43d58d03f274b4abfdbaff70175150f886e505fda309122c54f | 18,087 | null | """Width-16 linear-memory DAN with differentiable scratch nodes."""
from __future__ import annotations
import math
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_m... | {
"id": "39681aea-aea1-477a-be81-b9e1bb7c3045",
"created_at": "2026-08-05 17:57:02.907579+00:00",
"db_md5": "a850412b7f139e1cc1dcd39687b08f3b",
"submitter": "Blake Camp",
"github_login": "blake-camp-surge",
"run_id": "2e1270ec-6324-4ffe-8a53-9c49eeef609d",
"tier": "easy",
"dataset_id": "e3",
"status":... | {
"score": {
"mean_loss": 2.133608462282596,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.013125000009313225
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.134752860428357,
"example_count": 800,... |
3979b5ef-cdd3-4483-81a8-fc2912dbbce2 | easy | shirvani-jr | 2026-08-31 02:13:38.797539+00:00 | succeeded | 8287e623e81e690e9789e418e091edbc30c2c910f1d7989c44c06500efcf995f | 21,497 | 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 = 3
PERIOD_MAXIMUM = 100
EX... | {
"id": "3979b5ef-cdd3-4483-81a8-fc2912dbbce2",
"created_at": "2026-08-31 02:13:38.797539+00:00",
"db_md5": "b30b82037f1c986f946187e16d28e90b",
"submitter": "Ali",
"github_login": "shirvani-jr",
"run_id": "adfa165e-14dd-4cc4-9267-e41381965863",
"tier": "easy",
"dataset_id": "e9",
"status": "succeeded"... | {
"score": {
"mean_loss": 5.894297122955322,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.5155555605888367
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 10.097929000854492,
"example_count": 90,
... |
3980fe81-e213-4d0f-8488-4d052cd5e7d9 | easy | ehonig | 2026-08-21 00:43:55.189046+00:00 | succeeded | db1f4951420af8c26e8ee03186c289cea79061dd3023124ac720c9ff546bd30d | 30,222 | 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": "3980fe81-e213-4d0f-8488-4d052cd5e7d9",
"created_at": "2026-08-21 00:43:55.189046+00:00",
"db_md5": "6aaf761bbb73c332b808e69bd55ac7f3",
"submitter": "Edouardo Honig",
"github_login": "ehonig",
"run_id": "20a38b59-8de3-4220-bf2a-776720eee4ec",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.1620881735674433,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0020833333333333333
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.166863203048706,
"example_count": 60... |
39836592-e8fc-48bb-b2e7-34d34b5f7ee4 | easy | sapient-sapiens | 2026-08-20 15:17:16.235183+00:00 | succeeded | f5622aa759caadb03ae8d5d7d5813b1b14c9a7e3e7c2f7f332184755deb708ec | 34,705 | null | """Categorical-modulus-conditioned recurrent operator.
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 separate deco... | {
"id": "39836592-e8fc-48bb-b2e7-34d34b5f7ee4",
"created_at": "2026-08-20 15:17:16.235183+00:00",
"db_md5": "6d11b56e8ce1d6845c5f087b29267a72",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "de60ef95-7fb0-40ee-86c2-ecc18550dac5",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 2.6065437237564346,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.002916666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.5759166356159433,
"example_count": 60... |
39844448-4224-41af-b05f-276d47ebe9eb | easy | velocizapkar | 2026-08-16 07:11:47.288763+00:00 | succeeded | d50fb9cafd08dac007facd7dc21890865e69281b96784628f08876b136f40ac3 | 11,484 | null | """CE-to-exact-expected-whole-answer objective control.
The model, logits, initialization, optimizer, batch size, and time-based
learning-rate schedule are identical to the matched standard Transformer.
After a fixed CE warmup, the loss exactly marginalizes the reward of sampling
an entirely correct answer under the m... | {
"id": "39844448-4224-41af-b05f-276d47ebe9eb",
"created_at": "2026-08-16 07:11:47.288763+00:00",
"db_md5": "bacc3c38c410c317876579b584f9f8e4",
"submitter": "Aakanksh Zarapkar",
"github_login": "velocizapkar",
"run_id": "36f5171a-5427-4377-8eb3-ef2a71623f22",
"tier": "easy",
"dataset_id": "e6",
"statu... | {
"score": {
"mean_loss": 10.471948623657227,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.07522523030638695
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.706766128540039,
"example_count": 60,
... |
39859c4d-76d5-427d-aac6-df59ee0e25c2 | easy | Yalyenea | 2026-08-12 08:35:14.480800+00:00 | succeeded | d720fe5edd7c8b1f3eb471dce1190b7d8d38354fbda24621c88fdee8bbf6805f | 10,749 | null | """Compact decimal recurrence with stronger one-step supervision."""
from __future__ import annotations
import math
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_... | {
"id": "39859c4d-76d5-427d-aac6-df59ee0e25c2",
"created_at": "2026-08-12 08:35:14.480800+00:00",
"db_md5": "85c4084c0651afd8c199f2ca3018b3ac",
"submitter": "yfff",
"github_login": "Yalyenea",
"run_id": "f1a329a2-92e2-4429-9245-66d2bb98e6e0",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded",
... | {
"score": {
"mean_loss": 3.176294514336213,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0066666666915019355
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.7776544329296846,
"example_count": 60... |
398d7447-944f-4f7a-9a16-f0050ed03878 | easy | alirezashirvani-jr | 2026-08-28 23:43:44.288017+00:00 | succeeded | 73dad23ac67ce34a8f5ba4257342fbd955f792c46dd910f2668b969b51df72ec | 53,560 | null | """Adaptive learned discrete recurrent fabric for One Layer Deeper.
A gradient-trained, operation-free recurrent computational machine. The
model learns, from the competition's endpoint supervision: how to parse the
prompt into a bank of categorical digit registers, a reusable transition
circuit applied serially, a l... | {
"id": "398d7447-944f-4f7a-9a16-f0050ed03878",
"created_at": "2026-08-28 23:43:44.288017+00:00",
"db_md5": "c68d20cdc8ff54eb88edee065db13428",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "9f964f12-c131-4c5b-917b-9a5e59037dbb",
"tier": "easy",
"dataset_id": "e10",
... | {
"score": {
"mean_loss": 1.9343796372413635,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004000000189989805
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.9377200603485107,
"example_count": 12... |
398d9d71-5791-4898-a363-82a323af9c6a | easy | gauravmishra | 2026-08-08 06:16:37.856633+00:00 | succeeded | 75b766be3bdb202352ea93a6e16ba69e78f53670d40d6dfd8e80bb467030a160 | 3,787 | null | """Four-block Transformer stack reused for two complete 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,
)
D_MODEL = 128
NUM_... | {
"id": "398d9d71-5791-4898-a363-82a323af9c6a",
"created_at": "2026-08-08 06:16:37.856633+00:00",
"db_md5": "e5cb4f65a2ccf87b2b6b891df6235afe",
"submitter": "Gaurav Mishra",
"github_login": "gauravmishra",
"run_id": "68e006d7-3a8b-49e5-8c0f-a44f63b49ddb",
"tier": "easy",
"dataset_id": "e5",
"status": ... | {
"score": {
"mean_loss": 9.660651373991685,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006250000024835269
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 10.017485430422385,
"example_count": 600... |
3991d5ae-9afb-4800-8387-331721aae921 | easy | erdavis0 | 2026-08-24 16:23:37.877957+00:00 | succeeded | 51c1ac6f1c1b63d62814de1d34409c49ecd688b4e0533ed3562b45bbe4363796 | 9,114 | null | """Iterative latent workspace for soft prompt-role discovery and slot readout."""
from __future__ import annotations
import math
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_s... | {
"id": "3991d5ae-9afb-4800-8387-331721aae921",
"created_at": "2026-08-24 16:23:37.877957+00:00",
"db_md5": "8c1df270c6f26cdb1236360125a41b1a",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "7eb7b544-d4f8-4bbc-95f6-ba585c4a8178",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded",... | {
"score": {
"mean_loss": 6.115540266036987,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0054166666232049465
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.131560802459717,
"example_count": 600... |
399493d9-e10b-4845-854f-f9724abcfb6d | easy | ehonig | 2026-08-19 16:52:21.241367+00:00 | succeeded | 84807a9b3fbdd1623e837d3c104c3be3068d7103ca97f7e3e356c67dbc4d2b64 | 16,887 | 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": "399493d9-e10b-4845-854f-f9724abcfb6d",
"created_at": "2026-08-19 16:52:21.241367+00:00",
"db_md5": "d64131975e4b45cab4c8754d2100fe0b",
"submitter": "Edouardo Honig",
"github_login": "ehonig",
"run_id": "f6cf4cae-aa0d-428c-8b74-05dca627d68c",
"tier": "easy",
"dataset_id": "e4",
"status": "succ... | {
"score": {
"mean_loss": 2.0916379788317974,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0046875
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.105814338383772,
"example_count": 1200,
... |
39962e78-a957-47e5-91be-d7395156e9e0 | easy | ehonig | 2026-08-13 22:32:51.300572+00:00 | succeeded | 160f717c1731ca308a8cbdc57042bb04dab99268336c0a978ed68dfd22b40c80 | 42,906 | null | """DigitLoop — a place-value recurrent model for repeated modular squaring.
The task is ``y = x^(2^T) mod N`` with the answer read back tail aligned, one
decimal place value per token position from the right. A stack of attention and
MLP layers has no primitive for the two operations this needs — a partial-product
co... | {
"id": "39962e78-a957-47e5-91be-d7395156e9e0",
"created_at": "2026-08-13 22:32:51.300572+00:00",
"db_md5": "9ae78a6dfeca2c4ffdc4279d8205b868",
"submitter": "Edouardo Honig",
"github_login": "ehonig",
"run_id": "b5db3ca6-9fbc-4447-b8ea-72ab2454701f",
"tier": "easy",
"dataset_id": "e1",
"status": "succ... | {
"score": {
"mean_loss": 4.196907877922058,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.07999999821186066
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.492194652557373,
"example_count": 100,
... |
399660dc-a6d7-466c-9b41-8dfea1f732ec | easy | erdavis0 | 2026-08-26 03:45:18.501741+00:00 | succeeded | c67927e1f1a3eb13d9778a95b41512101bbd0baa70d6fd25c9f9b89607e3826f | 30,360 | null | """Latent-role continuous REV C2: exact length-normalized sequence mixture.
Public delimiter syntax only routes opaque token categories into distinct,
right-aligned N, X, and T tapes. Learned local cells compile separate
interaction, carry, and modulus-conditioned workspaces. N is isolated from
interaction and carry... | {
"id": "399660dc-a6d7-466c-9b41-8dfea1f732ec",
"created_at": "2026-08-26 03:45:18.501741+00:00",
"db_md5": "b3917270504f5239445a48889c48e9ae",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "8d84f8de-e1cf-4c4e-bd6e-a3034b3850a7",
"tier": "easy",
"dataset_id": "e9",
"status": "succeeded",... | {
"score": {
"mean_loss": 4.94569206237793,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.052222223952412605
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.626527786254883,
"example_count": 90,
... |
399b77b5-6a9f-4bc4-9211-239251909195 | easy | jordanrubin | 2026-08-01 20:19:27.613390+00:00 | succeeded | 2c6adbc91db77091eaa1fdb53b2b15fdc6e05306608a2440be5e6a2aa2591b8a | 18,183 | null | """Looped bidirectional selective-SSM (Mamba-style), pure PyTorch.
The statefulness bet: input-dependent SSM dynamics along the token axis,
with the block weight-tied and applied NUM_LOOPS times in depth with input
injection — recurrence in depth is where the task's T-iteration lives.
The selective scan is computed i... | {
"id": "399b77b5-6a9f-4bc4-9211-239251909195",
"created_at": "2026-08-01 20:19:27.613390+00:00",
"db_md5": "35fec302dbad6ab1de27b5092ede0b6a",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "88f9d11d-2965-47a4-bfc3-678af9901239",
"tier": "easy",
"dataset_id": "e1",
"status": "s... | {
"score": {
"mean_loss": 5.067467241810444,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04833333070079486
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.312114715576172,
"example_count": 100,
... |
399c3e2c-f969-4f2a-8367-d9bdc09fda0a | easy | liam-gb | 2026-08-13 00:51:34.822336+00:00 | succeeded | 9db49abf7e909af0722b300a0081859b559063890d7ec3fa0650817e9e1d46d6 | 14,529 | null | """rns_pm_gather_ct_widet_nc: widet with N-conditioning re-injected each loop.
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-fr... | {
"id": "399c3e2c-f969-4f2a-8367-d9bdc09fda0a",
"created_at": "2026-08-13 00:51:34.822336+00:00",
"db_md5": "e8ff57bb0d37941db3f6534309a0d17b",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "9e7423da-507d-46b8-a474-14e53a473f79",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded"... | {
"score": {
"mean_loss": 9.085074186325073,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.09999999578576535
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 10.349267959594727,
"example_count": 800,... |
39a425de-9304-4eac-b7a5-ab80f6aecf18 | easy | christopherhwood | 2026-08-08 20:04:20.812433+00:00 | succeeded | ec39724bffc0be2b426c5b2a9aa9982dc6b76ce361d91a254f132a8547015e5d | 18,306 | null | """Length-scaled gated convolutional reasoner for repeated composition.
The learned transition is a small Neural-GPU-style cellular state machine. A
digit register and immutable context are projected into one cell per digit;
the same local gated update is then applied a number of times proportional to
the register le... | {
"id": "39a425de-9304-4eac-b7a5-ab80f6aecf18",
"created_at": "2026-08-08 20:04:20.812433+00:00",
"db_md5": "7c418b05e8b1125a0fa5e8d0f0821db2",
"submitter": "Chris Wood",
"github_login": "christopherhwood",
"run_id": "5bce141d-de34-46af-b785-c6d499359e18",
"tier": "easy",
"dataset_id": "e1",
"status":... | {
"score": {
"mean_loss": 3.4251967072486877,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.7350000143051147
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.6053892374038696,
"example_count": 100,... |
39a482ba-aae4-4b7a-9ceb-38b429ce0bc4 | easy | anirudh-chakravarthy | 2026-08-12 03:39:25.012204+00:00 | succeeded | caeb9bcb33b341c64e71dc56c0dd58200dc6c6b2386204096034183d14a0b650 | 15,396 | 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": "39a482ba-aae4-4b7a-9ceb-38b429ce0bc4",
"created_at": "2026-08-12 03:39:25.012204+00:00",
"db_md5": "1f76b3b20b9ba66d1177a1a6e2a99032",
"submitter": "Anirudh S Chakravarthy",
"github_login": "anirudh-chakravarthy",
"run_id": "4168a7f4-8d47-4f40-8a09-8caafde22ffb",
"tier": "easy",
"dataset_id": "... | {
"score": {
"mean_loss": 6.500515215195506,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.01312500001396984
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.451594427146478,
"example_count": 800,
... |
39a4975c-e9e8-40a0-ab3d-b8d02a436df5 | easy | AFeinfield | 2026-08-18 23:36:57.389410+00:00 | succeeded | 17ee37d7d9236ff3f94a266b3b5a93306c0ae3e73fa9e075b814251c3b829d8b | 13,845 | null | """082: low-rank bilinear — mul_a/mul_b factored through r=128.
Bilinear costs ~3x per step on e2; if the product only needs a
low-dimensional interaction space, most of that is recoverable.
Original 073 notes follow.
073: BILINEAR multiply + K=5 agreement heads.
The multiply sublayer becomes W((Ah)*(Bh)) — a native ... | {
"id": "39a4975c-e9e8-40a0-ab3d-b8d02a436df5",
"created_at": "2026-08-18 23:36:57.389410+00:00",
"db_md5": "243823cea1f7b180e4c36aac8d654d67",
"submitter": "AFeinfield",
"github_login": "AFeinfield",
"run_id": "f095a1ae-5cdc-4769-af89-f26c8a878da4",
"tier": "easy",
"dataset_id": "e2",
"status": "succ... | {
"score": {
"mean_loss": 3.2741048336029053,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008541666902601719
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8144114017486572,
"example_count": 30... |
39a5184a-4209-4287-a07e-b46dd9dee4cf | easy | yashkant | 2026-08-12 03:23:01.999547+00:00 | succeeded | a3f46dfb02451852233cb93cbd5e79d729f2c77a3cd5494c861547487f3b677d | 32,532 | null | """Round1781 common-mode-free Dykstra correction residuals.
The proven Round1766 W8/D2 carrier is retained. Before the parent's bounded
tanh stores each initial or recurrent correction, the candidate removes that
W8 residual's coordinate mean. This reserves correction memory for relative
latent-coordinate disagreeme... | {
"id": "39a5184a-4209-4287-a07e-b46dd9dee4cf",
"created_at": "2026-08-12 03:23:01.999547+00:00",
"db_md5": "0bb48cbf4769aae62ef4e2491fcd1345",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "9b74a43d-8374-40ae-b89a-59a49e14f43f",
"tier": "easy",
"dataset_id": "e2",
"status": "succeed... | {
"score": {
"mean_loss": 2.8735029697418213,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0475000012665987
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.5106327533721924,
"example_count": 300,... |
39a52ce1-bdbb-4cc1-823d-c1f3750840bc | easy | karanganesan | 2026-08-04 01:01:47.748669+00:00 | succeeded | f920e3874c2c981beb0a84aac70162c3c5594d7deaf0f9ad656cb63e87a989f9 | 11,992 | 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": "39a52ce1-bdbb-4cc1-823d-c1f3750840bc",
"created_at": "2026-08-04 01:01:47.748669+00:00",
"db_md5": "82c14aa3f2a3e4f8be1bc6f6172a2b10",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "55f526d7-95d5-4454-ac7b-4602962e0bbe",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 1.9753868579864502,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03333333507180214
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.9759811162948608,
"example_count": 100... |
39a6ae8e-fb8d-4324-9ac6-0ba98d4c1ddb | easy | kushasareen | 2026-08-12 02:14:02.517459+00:00 | succeeded | 4422681916dc28a5e632125ded3b55482da3c86e141ed64dcaf0d9a052e4639d | 6,912 | null | """v2 capacity descent: d_model 128, lr 4e-3 (1/width scaling).
T-conditioned recurrent Transformer with the measured stability fixes:
- 2 untied pre-blocks (pre-norm), 1 untied post-block, tied loop block
- prompt re-injection every loop iteration (gradient + information path)
- UT-style double post-norm inside the l... | {
"id": "39a6ae8e-fb8d-4324-9ac6-0ba98d4c1ddb",
"created_at": "2026-08-12 02:14:02.517459+00:00",
"db_md5": "3a6ce0d2de590731e6a6be87901557e3",
"submitter": "Kusha Sareen",
"github_login": "kushasareen",
"run_id": "4b23127a-3a23-4f37-b8fc-5a8abf777089",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 8.566758734219938,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0033333333333333335
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.5569625777514,
"example_count": 600,
... |
39a77924-953d-48da-8e94-786999245a43 | easy | DDanlov | 2026-08-09 00:51:31.684745+00:00 | succeeded | e58f0a07f3be732b92b95df0057bb94f22708c24be0c4d33ca4afba9d31c8f6f | 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": "39a77924-953d-48da-8e94-786999245a43",
"created_at": "2026-08-09 00:51:31.684745+00:00",
"db_md5": "6003f5b2d3b79f35aff4b4ab453eeb4b",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "703ac3ee-eace-470d-b5a0-8ee95a4d62c3",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 22.6761402001702,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.013333333333333334
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 27.330598831176758,
"example_count": 100,... |
39b7366b-a256-4426-a180-ff7abcae93ce | easy | satyaki-up | 2026-08-07 17:08:31.736555+00:00 | failed | 4115f517dfd5031d18193a586945ab7ae2ae93eb0fbf0c0baf6e515555708302 | 11,908 | null | """Leaky LSTM with RNN optimization tricks from arXiv:1212.0901."""
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 = 2151
LAYE... | {
"id": "39b7366b-a256-4426-a180-ff7abcae93ce",
"created_at": "2026-08-07 17:08:31.736555+00:00",
"db_md5": "5f403029e3813add14fbefed070f6636",
"submitter": "Satyaki",
"github_login": "satyaki-up",
"run_id": "c2d52a4a-682b-4a1f-b7fc-42f0467c5e40",
"tier": "easy",
"dataset_id": "e5",
"status": "failed"... | null
|
39bde393-9534-4fc8-8c16-12f876812b68 | easy | repst | 2026-08-26 21:16:31.792600+00:00 | succeeded | 1e8a4ffc87e6f8c9c9714243243299621b0abfd79de902095d12dd29280e9149 | 168,027 | null | # R752: dynamic-loader-compatible packaging of frozen R751.
#
# This revision changes only the standalone namespace alias. The benchmark's
# uploaded-source loader need not register its temporary module in sys.modules;
# R752 resolves the inherited-module-style `R.name` references through a proxy
# over this source's ... | {
"id": "39bde393-9534-4fc8-8c16-12f876812b68",
"created_at": "2026-08-26 21:16:31.792600+00:00",
"db_md5": "62c30bf0797612d917bbbc321a744579",
"submitter": "Ryan Epstein",
"github_login": "repst",
"run_id": "975e057c-fa20-4e64-8e81-06cbb7c57ec3",
"tier": "easy",
"dataset_id": "e1",
"status": "succeed... | {
"score": {
"mean_loss": 0.0210082208751666,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.021008392795920372,
"example_count": 100,
"e... |
39c48635-e0c8-47d4-84fa-23b0e257a118 | easy | viridale | 2026-08-02 15:51:04.140389+00:00 | succeeded | 5b3397af07e2feedaf903f7a2d9542e4948491ceb6cbafb2cd28945b30a90525 | 56,953 | null | """Contradiction-reopen split conditional five-monomial expansion.
hypothesis: v109 decides catalogue closure from only the first training batch,
so a lower-family alias on that batch permanently hides family 7 even when
later provided final strings falsify the executed learned winner. Keep the
exact-zero... | {
"id": "39c48635-e0c8-47d4-84fa-23b0e257a118",
"created_at": "2026-08-02 15:51:04.140389+00:00",
"db_md5": "4e19f6f9ad49995d2472f858327384e3",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "acbebfa6-80c0-4a46-a08c-c6b629f9eb6d",
"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,
"... |
39c82cac-7fe8-4334-bddf-8075d72a996b | easy | sapient-sapiens | 2026-08-23 17:45:08.329506+00:00 | succeeded | 0adf6a3c2ea9f19f3cb37ad2f596902f0d350f07e392bd4425978deb54ec7f98 | 16,079 | null | """Scaled six-digit two-phase scan with adaptive recurrent reduction.
The model uses categorical learned embeddings and marker-derived alignment only.
No N, X, or T field is decoded into a number. All recurrent sweep readouts are
supervised exclusively against the same final answer.
"""
from __future__ import annota... | {
"id": "39c82cac-7fe8-4334-bddf-8075d72a996b",
"created_at": "2026-08-23 17:45:08.329506+00:00",
"db_md5": "305ad8b5056c0f90badf187e86cff720",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "205de4d4-1bab-498d-93c7-25085db63523",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 2.1519421381597525,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006666666679084301
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.153577539017382,
"example_count": 600... |
39c8c92b-e0f2-421a-b539-e4aacabea969 | easy | khushidahi | 2026-08-23 19:33:10.789566+00:00 | failed | dc30f96367eeccb5fe562f1ff3bdff32abfe024a824930c0ad6f787d71aa41f0 | 11,734 | null | """ART-DT recall candidate.
A learned recurrent sequence model with right-to-left Abacus positions and
global-sequence rotary attention. This variant changes only optimizer LR and
warmup schedule from its architectural parent. This branch ties the two
within-visit layer calls to one TransformerLayer object. One imm... | {
"id": "39c8c92b-e0f2-421a-b539-e4aacabea969",
"created_at": "2026-08-23 19:33:10.789566+00:00",
"db_md5": "25f8e724d7ec7f02d7c56a4ef3401964",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "c1a0ec0f-f711-4c32-86d4-2717ac7f8b41",
"tier": "easy",
"dataset_id": "e9",
"status": "fail... | null
|
39cc79fc-5270-4d50-bbf4-a016992302d5 | easy | DDanlov | 2026-08-24 22:22:18.301580+00:00 | succeeded | 65d720bd68fff379e91fe9dc2db6662f6c9acda4aa9020c4953e19f8e604926d | 51,673 | 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": "39cc79fc-5270-4d50-bbf4-a016992302d5",
"created_at": "2026-08-24 22:22:18.301580+00:00",
"db_md5": "a31a6572715239dbf5f66f7fe212a110",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "a1eaff19-6a8f-4beb-9fb3-884ff2bd0f2d",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.9746164943477638,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007916666691501935
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.950336300204153,
"example_count": 600... |
39cc7df3-bee0-4578-b55a-fd1ca4805ecc | easy | TZZheng | 2026-08-07 05:13:31.108208+00:00 | succeeded | ca08a1187f3cb4b88257443d9a72bfdd3b9e5df5e5682b627d5f01c49d031a2e | 12,525 | null | """Learned recurrent modular-transition model.
The mutable digit state is initialized from x. A learned square-oriented
self-mixing network and an N-conditioned reduction network form one transition,
and exactly the same transition parameters are reused T times. No arithmetic
operation is hard-coded; only the public p... | {
"id": "39cc7df3-bee0-4578-b55a-fd1ca4805ecc",
"created_at": "2026-08-07 05:13:31.108208+00:00",
"db_md5": "5921d6cee1ed200d873e9c1a5d9c2d22",
"submitter": "TZZheng",
"github_login": "TZZheng",
"run_id": "8b8fc564-3add-4811-937d-09d4ad978d7f",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 6.353623867034912,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.044999998062849045
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.125143051147461,
"example_count": 100,... |
39cf30b8-268b-4b66-89d3-4ba865ae27a4 | easy | lpbb | 2026-08-18 10:16:18.723688+00:00 | succeeded | d0302a6f360f9cd450bde9d5a9fe5faf9def60c8b1d2057059e71ed6e25bad82 | 6,792 | null | """S0 + abacus embeddings: same 12 untied layers as s0, plus a per-digit
significance tag (units=1, tens=2, ...) embedded additively, with random
per-row offsets at train time so tag rows beyond the trained digit counts
still receive gradient (McLeish et al. 2024)."""
from __future__ import annotations
import torch
i... | {
"id": "39cf30b8-268b-4b66-89d3-4ba865ae27a4",
"created_at": "2026-08-18 10:16:18.723688+00:00",
"db_md5": "c1e6e39cf76ca90dfa41517cdfede47a",
"submitter": "Lpbb",
"github_login": "lpbb",
"run_id": "17753c77-f5e2-4aa3-874b-209ebd860742",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
"s... | {
"score": {
"mean_loss": 2.3591853380203247,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.06333333067595959
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.5566511154174805,
"example_count": 100... |
39d2f662-de9b-4b35-90c9-f870769db7c1 | easy | MarcoF1 | 2026-08-13 16:42:52.519463+00:00 | succeeded | f0915f788e1e674926b16b4dc5854d94aee9f8f5afc6319f025517e676d18a55 | 15,350 | null | """Residue-clock model for fixed-N generalization.
Representation: parsed integer values of x and N feed (a) exact residue-phase
features over a bank of small integer periods, (b) learned double-hashed value
embeddings (per-residue identity), and T feeds cyclic phase features
(T mod 2..48) plus saturating channels — p... | {
"id": "39d2f662-de9b-4b35-90c9-f870769db7c1",
"created_at": "2026-08-13 16:42:52.519463+00:00",
"db_md5": "96ea10c35cb00331a503ef880c7a84e7",
"submitter": "Marco Fleming",
"github_login": "MarcoF1",
"run_id": "883f0148-d834-4fb0-87b4-136783b1cc51",
"tier": "easy",
"dataset_id": "e1",
"status": "succ... | {
"score": {
"mean_loss": 2.7961326350000433,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.044999998609224955
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.1804513931274414,
"example_count": 10... |
39d4a83a-6794-413f-a46e-94ff0edf4240 | easy | chad-atexpedient | 2026-08-09 05:25:43.084965+00:00 | succeeded | 1ca1a06d5c6d39c7d7d7422f8baac1a4cf076b82551f8ae01a1766693e2d259a | 18,589 | null | """X06: pair register with functional LSD-relative place features.
The mutable register is initialized from x once. Every subsequent application
of the tied cell receives only the current LSD-first digit state and immutable
N digits. Requested T controls only the number of applications. The state
logits are also t... | {
"id": "39d4a83a-6794-413f-a46e-94ff0edf4240",
"created_at": "2026-08-09 05:25:43.084965+00:00",
"db_md5": "a2059a2de93730fd195bcfe19388c5a3",
"submitter": "chad-atexpedient",
"github_login": "chad-atexpedient",
"run_id": "d32f5572-ad79-404a-98ea-7f37b94241d0",
"tier": "easy",
"dataset_id": "e5",
"st... | {
"score": {
"mean_loss": 2.9832798683562793,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006666666759798924
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.4769837856292725,
"example_count": 60... |
39d6aca0-10b8-478a-8fef-f1eac2f7caea | easy | christopherhwood | 2026-08-08 04:42:46.995602+00:00 | succeeded | 4c05f9224562d1a5e3ed959faad257804d347b71edfd50ffa95f7815f0112c54 | 26,658 | null | """Hard-tier global posterior over recurrent binary programs.
The architecture supplies randomly initialized trainable particles over a
symmetric 5 x 7 x 2 class of small affine local relations. A second randomly
initialized learned posterior chooses one complete program from final evaluator
answers. In addition to ... | {
"id": "39d6aca0-10b8-478a-8fef-f1eac2f7caea",
"created_at": "2026-08-08 04:42:46.995602+00:00",
"db_md5": "718d05c043b22f0ab39ab7e76919e781",
"submitter": "Chris Wood",
"github_login": "christopherhwood",
"run_id": "4c86ee47-80ea-452d-ba19-cb7732ef5d7d",
"tier": "easy",
"dataset_id": "e5",
"status":... | {
"score": {
"mean_loss": 1.8505724706819945e-06,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.8502813487642183e-06,
"example_count": 600,
... |
39e043c5-45af-45ec-a403-1bf8eed9af80 | easy | rougee | 2026-08-28 03:58:13.223453+00:00 | succeeded | 605770b8b621ad1f686f32cf18130084d57eb59e033fa889fbd79f6e6ab6b728 | 7,818 | null | """Official Submittable Champion for One Layer Deeper (Differentiable Algebraic Recurrence).
End-to-End Differentiable Recurrent Architecture:
- Input Representation: Dynamic Token Embedding + Positional Embedding
- Differentiable Residue Extraction: Algebraic position-weighted digit accumulation for N, X, T
- Unrolle... | {
"id": "39e043c5-45af-45ec-a403-1bf8eed9af80",
"created_at": "2026-08-28 03:58:13.223453+00:00",
"db_md5": "f873d737a8cf254a6129903d6dc51a9a",
"submitter": "Rougee",
"github_login": "rougee",
"run_id": "bf210d41-741a-4f11-ba91-bbfbe2f01a03",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
... | {
"score": {
"mean_loss": 4.931945795193823e-06,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.940945927955909e-06,
"example_count": 100,
... |
39e3fa87-a32c-4234-98ff-1245a487c204 | easy | MarcoF1 | 2026-08-13 06:27:50.000447+00:00 | succeeded | fe91768d49bb4e34841d504a4c071f4c850772ae6ae818579e1aef2284a4740f | 3,816 | null | """Untied 4-block transformer control (fixed feedforward depth, no recurrence)."""
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_... | {
"id": "39e3fa87-a32c-4234-98ff-1245a487c204",
"created_at": "2026-08-13 06:27:50.000447+00:00",
"db_md5": "c8caae964ed19b3274d1b0a37496e85b",
"submitter": "Marco Fleming",
"github_login": "MarcoF1",
"run_id": "f519b517-b089-4bd1-9945-f89bdc1e1820",
"tier": "easy",
"dataset_id": "e1",
"status": "succ... | {
"score": {
"mean_loss": 11.416754107820125,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.036666665996114414
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 11.052905082702637,
"example_count": 10... |
39ec9841-4e89-4e04-bd9f-af17629232bb | easy | erdavis0 | 2026-08-26 03:54:53.195502+00:00 | succeeded | c67927e1f1a3eb13d9778a95b41512101bbd0baa70d6fd25c9f9b89607e3826f | 30,360 | null | """Latent-role continuous REV C2: exact length-normalized sequence mixture.
Public delimiter syntax only routes opaque token categories into distinct,
right-aligned N, X, and T tapes. Learned local cells compile separate
interaction, carry, and modulus-conditioned workspaces. N is isolated from
interaction and carry... | {
"id": "39ec9841-4e89-4e04-bd9f-af17629232bb",
"created_at": "2026-08-26 03:54:53.195502+00:00",
"db_md5": "b3917270504f5239445a48889c48e9ae",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "d0b7382b-b7cc-4599-8b07-2c1df9934e0d",
"tier": "easy",
"dataset_id": "e4",
"status": "succeeded",... | {
"score": {
"mean_loss": 2.23816806235525,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005625
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2389061591728563,
"example_count": 1200,
... |
39ed77eb-9314-486d-83e8-c95000d40be6 | easy | khushidahi | 2026-08-03 05:34:37.372677+00:00 | succeeded | aef42e525f4edb1146f5f3304628cfd6d7aee967fe2f80d1989a5424909a85b9 | 14,365 | 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": "39ed77eb-9314-486d-83e8-c95000d40be6",
"created_at": "2026-08-03 05:34:37.372677+00:00",
"db_md5": "590a5d7a1bdd7afece48d22717e51325",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "c0c427d8-2559-4ffe-bd6d-66595f2a1bba",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.2050609588623047,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007500000298023224
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2011234760284424,
"example_count": 60... |
39efa9a2-30d8-443a-8bcd-16ce57ee6b72 | easy | msnow0 | 2026-08-08 11:16:19.884934+00:00 | succeeded | 368cbb92c3ec4e910ef024d57c72fbed2622c93b3df4173e3c7bce75c3a5dc70 | 16,877 | null | """One Layer Deeper submission — mechanism 1 (see BRIEF.md §6, research/RULES.md).
Weight-tied recurrent operator over a discrete token tape. One outer iteration per
squaring step (loop count conditioned on the prompt's T, which is explicitly allowed);
between outer steps the latent is snapped back onto token embeddin... | {
"id": "39efa9a2-30d8-443a-8bcd-16ce57ee6b72",
"created_at": "2026-08-08 11:16:19.884934+00:00",
"db_md5": "b9182f4cd6fb33d3bf6b28cdaca6d3e9",
"submitter": "msnow0",
"github_login": "msnow0",
"run_id": "9a323570-e75f-4dcb-adc1-db6b9f43136c",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
... | {
"score": {
"mean_loss": 4.627742052078247,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.06000000052154064
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.7111897468566895,
"example_count": 100,... |
39fbfb09-6ef7-4ba1-9258-b8a1c259725b | easy | sapient-sapiens | 2026-08-22 18:43:04.690333+00:00 | succeeded | 3234378ea16030f37d9a6f7535c56d40e2d21c7742cd1a423eca4501ba7608b6 | 19,834 | null | """Batch-64 LR-1.5e-4 eight-layer T-private Q+V TRM for E5.
Research hypothesis: halving peak LR to 1.5e-4 improves batch-64 stability and
turns its greater example throughput into stronger fixed-clock learning.
Primary experimental variable: peak learning rate, 3e-4 versus 1.5e-4, at fixed
batch 64 with model, optim... | {
"id": "39fbfb09-6ef7-4ba1-9258-b8a1c259725b",
"created_at": "2026-08-22 18:43:04.690333+00:00",
"db_md5": "631e5d871abae1f7ed939e61669302bc",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "02b8722c-c61a-499e-91a7-47e9a8310509",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 2.1839355126057005,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00666666670391957
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.194129805260175,
"example_count": 600,... |
3a00aac7-ecf7-4599-b22c-0f66706c315d | easy | yashkant | 2026-08-12 04:12:12.104702+00:00 | succeeded | 858802f662e9673a92687ee901546bc8d4d6f929f0fd0c36ea0a082047d71baa | 31,750 | null | """Round1701 Dykstra width16/hidden32 architecture pair on the D416 MemGLU carrier.
The fixed Round1141 carrier first produces its ordinary pooled representation.
A compact primal state and two Dykstra correction memories are initialized
from that immutable representation. Every parsed public-T cycle applies two
cond... | {
"id": "3a00aac7-ecf7-4599-b22c-0f66706c315d",
"created_at": "2026-08-12 04:12:12.104702+00:00",
"db_md5": "b602aa508c65835af490b097651aa139",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "77f57497-1bd5-4e10-b0f6-46462845d66b",
"tier": "easy",
"dataset_id": "e3",
"status": "succeed... | {
"score": {
"mean_loss": 2.125429086937961,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.018125000158324838
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1257397042235557,
"example_count": 800... |
3a0421d0-4f7a-4e4c-b450-14fcb3978542 | easy | ehonig | 2026-08-18 12:58:58.788084+00:00 | succeeded | 09d959ee8fc4e115982b637b98bae4307d0a5885a8a8c2653926d4bf972fde75 | 12,391 | 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": "3a0421d0-4f7a-4e4c-b450-14fcb3978542",
"created_at": "2026-08-18 12:58:58.788084+00:00",
"db_md5": "fe07d9c04433d83ed581d6f74f0fcdad",
"submitter": "Edouardo Honig",
"github_login": "ehonig",
"run_id": "9d3d21b9-44f4-413a-88d5-70e01127deb4",
"tier": "easy",
"dataset_id": "e2",
"status": "succ... | {
"score": {
"mean_loss": 4.219460725784302,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.01458333432674408
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.220640182495117,
"example_count": 300,
... |
3a10857e-a959-48e0-a7f1-9e770eb4f77a | easy | AdityaVG13 | 2026-08-15 11:07:20.917575+00:00 | succeeded | 669fa1075f017ca370fdfeef3a8e7d471c8952d468e0052ef14183b43ad10ab1 | 109,743 | null | """AEGIS V12 v12_fc_gated: frontier-carry with soft-carry only on continuing hops.
One Layer Deeper submission candidate. All task transitions remain learned
end-to-end. The deterministic shell only identifies marker-delimited roles,
constructs a finite semantic tape, schedules learned interval programs, and
applies a... | {
"id": "3a10857e-a959-48e0-a7f1-9e770eb4f77a",
"created_at": "2026-08-15 11:07:20.917575+00:00",
"db_md5": "66940f452858262486a75afefa5d8e5e",
"submitter": "AdityaG",
"github_login": "AdityaVG13",
"run_id": "b4a1c71a-3f2b-487f-9fab-2c01189cf588",
"tier": "easy",
"dataset_id": "e1",
"status": "succeed... | {
"score": {
"mean_loss": 1.3944930732250214,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.6649999916553497
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.6032988429069519,
"example_count": 100,... |
3a12ccf1-9a8f-4af8-ac84-b7319e3a7c38 | easy | Avyukth | 2026-08-12 20:02:26.232795+00:00 | succeeded | 3195d76dc631d014ad4f9b13630af8c7588b55e17e43e63b6e54813b0dcf9821 | 21,767 | null | """T-501 — compose T-101b training augmentation with T-401 averaging.
The wide-f0 training path uses the fixture-verified random-half, equal-width
sign transform from T-101b. In eval mode, the same trained weights are run on
x and (-x mod N), and every returned logit position is averaged for eligible
rows; the separa... | {
"id": "3a12ccf1-9a8f-4af8-ac84-b7319e3a7c38",
"created_at": "2026-08-12 20:02:26.232795+00:00",
"db_md5": "6b9ca9e23cd9a914f71176179b866b11",
"submitter": "Subhrajit Makur",
"github_login": "Avyukth",
"run_id": "2aa74ee2-7ebe-45d3-88a2-4519bddded01",
"tier": "easy",
"dataset_id": "e1",
"status": "su... | {
"score": {
"mean_loss": 7.1331703662872314,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.1966666765511036
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.789929389953613,
"example_count": 100,
... |
3a1644ef-9110-4752-8a3c-96012ace5828 | easy | DDanlov | 2026-08-09 00:33:25.121793+00:00 | succeeded | 5a8316c1fda32cdcc7113500881f90a36609af4c3d9816f6b24e016fda3b13a5 | 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": "3a1644ef-9110-4752-8a3c-96012ace5828",
"created_at": "2026-08-09 00:33:25.121793+00:00",
"db_md5": "9ccc676495d3f97ec43e708e8259e252",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "54af4f0b-d84e-4c61-ab6b-9718af96b3c9",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 95.97483528567696,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0004166666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 102.34238793924786,
"example_count": 60... |
3a17722a-7421-4711-8fc6-2170b5e05ffb | easy | khushidahi | 2026-08-03 19:36:11.668367+00:00 | succeeded | a9f656fce019a7cae44db3db1dd60742378aa46a76a4e948fc9c9d05b4edbd75 | 15,243 | 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": "3a17722a-7421-4711-8fc6-2170b5e05ffb",
"created_at": "2026-08-03 19:36:11.668367+00:00",
"db_md5": "607f0e9ecba8e97b03f8b945ccfdbff4",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "5df136bc-fd9e-4ef9-94f0-60d1849ba0cf",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.205930471420288,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008333333767950535
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2003540992736816,
"example_count": 600... |
3a19a6e2-c9a1-43ed-b18c-1e745fc4f822 | easy | mnida | 2026-08-23 21:34:32.087696+00:00 | succeeded | 42bd71c66e74ac257e43530d5b9cee77fffbebab291628aa322f2387723341a4 | 6,666 | null | """Basic single-pass Transformer with PyTorch AdamW."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_model_state,
)
D_MODEL =... | {
"id": "3a19a6e2-c9a1-43ed-b18c-1e745fc4f822",
"created_at": "2026-08-23 21:34:32.087696+00:00",
"db_md5": "c75308790c86c472e87b9f14cdceabc3",
"submitter": "mnida",
"github_login": "mnida",
"run_id": "675b7c5c-d03d-4d8e-b9aa-989edb437350",
"tier": "easy",
"dataset_id": "e6",
"status": "succeeded",
... | {
"score": {
"mean_loss": 4.3748050928115845,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.07027027569711208
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.9193379878997803,
"example_count": 60,... |
3a24cdea-3672-4d15-8c0a-f3a51ac8d6a2 | easy | christopherhwood | 2026-08-06 01:29:02.584501+00:00 | succeeded | 7d74e31ae8abf31dade5dc40f25a33e18035e6d4d8ed404feb12405269e3039f | 14,642 | null | """GPU-scaled general Transformer transition for repeated composition.
The learned transition sees a current digit register and immutable modulus
context, but no arithmetic-specific features or process labels. Four distinct
Transformer blocks refine answer and scratch slots, allowing depth to learn a
sequence of oper... | {
"id": "3a24cdea-3672-4d15-8c0a-f3a51ac8d6a2",
"created_at": "2026-08-06 01:29:02.584501+00:00",
"db_md5": "011c4ccaa9b02ef179bcbb0376ea1b45",
"submitter": "Chris Wood",
"github_login": "christopherhwood",
"run_id": "cdd649c4-e573-44c5-ac70-b600d37cb986",
"tier": "easy",
"dataset_id": "e5",
"status":... | {
"score": {
"mean_loss": 4.145235421676387,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005000000012417635
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.3046754807100167,
"example_count": 600... |
3a448f67-d817-406d-b6a5-86b7bc855e5c | easy | khushidahi | 2026-08-02 22:41:30.935319+00:00 | succeeded | 5fb47536ee1d63cfcffff014726c0e9a66a788a4004e7e630c23d5ea740729ca | 8,517 | null | """Prelude -> recalled shared core -> coda model for One Layer Deeper.
R2 architecture reset:
- use a real one-block prelude and one-block coda;
- apply one shared recurrent Transformer core four times;
- inject immutable prelude context at every recurrent update;
- use sandwich RMS normalization for stable repeated a... | {
"id": "3a448f67-d817-406d-b6a5-86b7bc855e5c",
"created_at": "2026-08-02 22:41:30.935319+00:00",
"db_md5": "b6b0a6718b9a681df70af98775d49266",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "051300a0-d00e-43e7-9afc-096420bae039",
"tier": "easy",
"dataset_id": "e2",
"status": "succ... | {
"score": {
"mean_loss": 2.2449424266815186,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008750000270083547
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2491157054901123,
"example_count": 30... |
3a512535-1abc-459c-92cd-affc7129429a | easy | liam-gb | 2026-08-14 02:12:50.809320+00:00 | succeeded | 3c39c52b433ddd2675e5ce63fea469f9c9bd5e09a62e7e6fda5ccefcd8242585 | 14,760 | null | """rns_pm_gather_ct_widet_nore: widet with batch reuse off.
The wide readout is what makes long answers reachable; the trim is what pays
for it, by not running loop iterations whose update is multiplied by zero.
Base docstring:
Single change vs rns_pm_gather_ct: the readout carries one
distance-from-units slot per s... | {
"id": "3a512535-1abc-459c-92cd-affc7129429a",
"created_at": "2026-08-14 02:12:50.809320+00:00",
"db_md5": "940211fbc28f5e5b59bd7c1149be5b02",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "13dc3869-cfb5-4a19-a196-777562086685",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 3.558953722385136,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.018750000045013926
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.146072864532471,
"example_count": 600,... |
3a5174a4-a69d-4f9c-924d-b8091854a06a | easy | kushasareen | 2026-08-27 03:54:16.457842+00:00 | succeeded | 2e3982e5fbcef4ca89caa9995b2aee9d72b87925d237b7502308beeadc574466 | 10,901 | null | """loop2_noT: digitcomp (4 blocks) where blocks 2..3 repeat T times; T token dropped.
At T = 1 this is exactly the digitcomp model. Muon recipe, 3 updates per batch.
Each number's token is a learned function of its digits: every place has
its own embedding table, the place embeddings are summed and passed
through a s... | {
"id": "3a5174a4-a69d-4f9c-924d-b8091854a06a",
"created_at": "2026-08-27 03:54:16.457842+00:00",
"db_md5": "971517701ab57cdaec681846e1761047",
"submitter": "Kusha Sareen",
"github_login": "kushasareen",
"run_id": "0fe040fe-880c-4f81-8163-1e4b30a64306",
"tier": "easy",
"dataset_id": "e2",
"status": "s... | {
"score": {
"mean_loss": 2.5838997215032578,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.42104168608784676
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.755796909332275,
"example_count": 300,... |
3a528880-4cae-4ca2-8efa-349dbb4dc2ba | easy | jordanrubin | 2026-08-06 08:06:19.030752+00:00 | succeeded | eadfbdc25f04113dc0d33a24f6554f923ee822fe348e3d7774e3d7641f884522 | 17,144 | null | """T1-emphasized diagnostic for the categorical looped Transformer.
This is a learned transition model, not an arithmetic program. The fixed
parts expose only the public prompt grammar and six representation priors:
right-origin decimal slots, a shared ordered digit geometry, unordered pair
tokens, immutable N recall... | {
"id": "3a528880-4cae-4ca2-8efa-349dbb4dc2ba",
"created_at": "2026-08-06 08:06:19.030752+00:00",
"db_md5": "b327020424fe1e416e817fbd0773de27",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "0c86732b-fe0c-4d7a-a6be-cd23ef539d49",
"tier": "easy",
"dataset_id": "e1",
"status": "s... | {
"score": {
"mean_loss": 2.052024245262146,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0033333334140479565
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.8202879428863525,
"example_count": 10... |
3a565f9a-4871-4b04-aa1f-0512e05dddaf | easy | chuk-uzowihe | 2026-08-12 22:01:20.887744+00:00 | succeeded | f760e96cf3d9b03fed1f947fee70053205a72d1eeeec048950326b4431b3fce5 | 27,100 | 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": "3a565f9a-4871-4b04-aa1f-0512e05dddaf",
"created_at": "2026-08-12 22:01:20.887744+00:00",
"db_md5": "038c47e50dd24d5cc969050963fbf2a4",
"submitter": "Chuk Uzowihe",
"github_login": "chuk-uzowihe",
"run_id": "a66f358e-f52b-42f5-9151-ba3e150a0dc5",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 20.170616149902344,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.07999999821186066
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 16.95235252380371,
"example_count": 100,... |
3a5897d4-e0ee-4f1e-8b9f-e6bfef8cceb9 | easy | newjordan | 2026-08-09 16:30:13.106663+00:00 | succeeded | aa76ac0f314e1dec4fb3a43f0778c784c05575a2a058960f631c730b7a4764bb | 93,178 | null | """Helix-Crawler — a recursive-weight-shared "crawler" ALU for modular squaring.
The task is r_{i+1} = r_i^2 mod n, chained T times: a *recurrence* over a single
operator. This build embodies the crawler / Frugendorff lineage (spark:~/sota_crawler)
mapped onto that structure. n never enters as a table row (unseen modu... | {
"id": "3a5897d4-e0ee-4f1e-8b9f-e6bfef8cceb9",
"created_at": "2026-08-09 16:30:13.106663+00:00",
"db_md5": "6b92e8e5ad290ffbd821c78f8ccb7e02",
"submitter": "Frosty40",
"github_login": "newjordan",
"run_id": "99694fc2-ad2c-49b4-afd3-9dbca9d282ee",
"tier": "easy",
"dataset_id": "e3",
"status": "succeed... | {
"score": {
"mean_loss": 0.00020847561558823502,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.00020836068478154922,
"example_count": 800,
... |
3a5cd871-4ec0-4651-b686-f88c78cb2056 | easy | erdavis0 | 2026-08-25 02:13:50.808219+00:00 | succeeded | 35ff275a7c933b6d0b5e1dc217d55253aa63433308fe14fd797471c38b44cb91 | 24,148 | null | """Matched categorical-boundary panel: C3 ST oracle selector.
Public delimiter syntax is used only to route opaque token identities onto
separate right-aligned tapes. One T-blind learned macro maps a categorical X
tape and a digitwise N tape to a new categorical tape through freshly reset
interaction, carry, and modu... | {
"id": "3a5cd871-4ec0-4651-b686-f88c78cb2056",
"created_at": "2026-08-25 02:13:50.808219+00:00",
"db_md5": "5fb89a7c7ac87a7ea6e553c7a9be46f2",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "bb5ccb32-ef47-40b6-9b8c-e06c91ffd8ce",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",... | {
"score": {
"mean_loss": 1.8353776335716248,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.07999999821186066
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.7811980247497559,
"example_count": 100... |
3a5e71aa-7a7b-4dbd-b44e-6ef9d134338d | easy | rougee | 2026-08-22 07:16:40.523229+00:00 | succeeded | 7db28211300245e4dd82204e5819a74cef3beb8842e9f9cbd4dc259e98915a38 | 7,047 | null | """Official Submittable Champion for One Layer Deeper (Deep Looped Recurrent Model).
End-to-End Differentiable Grokking Architecture:
- Recurrent Depth: Multi-loop shared transformer block with SiLU gated updates
- Topology: d_model=128, num_heads=4, num_loops=4
- Optimizer: AdamW (lr=2.5e-3, weight_decay=0.04)
- Sche... | {
"id": "3a5e71aa-7a7b-4dbd-b44e-6ef9d134338d",
"created_at": "2026-08-22 07:16:40.523229+00:00",
"db_md5": "9d4b689957054b8f8b57eab99e2b96fd",
"submitter": "Rougee",
"github_login": "rougee",
"run_id": "affcaf02-141f-4c69-b25b-1fd1d9a44fa8",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
... | {
"score": {
"mean_loss": 16.5187931060791,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 20.468690872192383,
"example_count": 100,
"exact... |
3a603956-dde2-4928-9238-7e32459fe0b5 | easy | yashkant | 2026-08-12 03:42:00.893360+00:00 | succeeded | 20b6688f7b4e4b727395cfd99a07b398e63a90f4995736f5c766ef7a4fb8fd79 | 31,750 | null | """Round1701 Dykstra width16/hidden32 architecture pair on the D416 MemGLU carrier.
The fixed Round1141 carrier first produces its ordinary pooled representation.
A compact primal state and two Dykstra correction memories are initialized
from that immutable representation. Every parsed public-T cycle applies two
cond... | {
"id": "3a603956-dde2-4928-9238-7e32459fe0b5",
"created_at": "2026-08-12 03:42:00.893360+00:00",
"db_md5": "78be01b213383e39de0c9a468834e2be",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "2db41c7b-01c7-4fcc-8858-83c38f2b8492",
"tier": "easy",
"dataset_id": "e2",
"status": "succeed... | {
"score": {
"mean_loss": 2.4870200157165527,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0422916691750288
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8511292934417725,
"example_count": 300,... |
3a743bf3-a15d-47c0-81c5-11ea30efeb1e | easy | karanganesan | 2026-08-04 01:03:00.948236+00:00 | succeeded | 0288d32eea17dadc8b101f8da499c63f774566aba83e053db242a9dc63090e5b | 12,014 | 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": "3a743bf3-a15d-47c0-81c5-11ea30efeb1e",
"created_at": "2026-08-04 01:03:00.948236+00:00",
"db_md5": "80ad5819a19dbb79ecde3652dcfafde5",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "816fc4f8-3312-4f90-9633-58f1cfafae98",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 2.364505887031555,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03333333320915699
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.4134185314178467,
"example_count": 100,... |
3a7d9ec6-0c14-47c1-b5b0-1532477a49d6 | easy | sapient-sapiens | 2026-09-01 00:24:36.923534+00:00 | succeeded | 2a2f63253487329b000800f27bc9c2d539c124393ab0ba4653b738e9301a0da3 | 20,399 | null | """One-block nested-Horner E5 model with an associative character carrier."""
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": "3a7d9ec6-0c14-47c1-b5b0-1532477a49d6",
"created_at": "2026-09-01 00:24:36.923534+00:00",
"db_md5": "cb804c580586644d1db1bcf64d5e5f35",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "4ea98017-aa37-4c8b-9d4c-c4095995b764",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 2.13989470452305,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007916666666666666
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.143192904145194,
"example_count": 600,
... |
3a8426c1-7e71-4bcf-84b9-421489866348 | easy | DDanlov | 2026-08-27 22:55:00.996476+00:00 | succeeded | 36d05da0ff24d04d107502844bee7612bd99befa8a8f927877617d6a5465dfb4 | 41,055 | 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": "3a8426c1-7e71-4bcf-84b9-421489866348",
"created_at": "2026-08-27 22:55:00.996476+00:00",
"db_md5": "a76fcfc2389f3706b2024e9c0222a392",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "44d1c7e7-b950-4d6f-bb3e-a8cc77cc2e29",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.573031187057495,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.002083333383779973
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.5866267681121826,
"example_count": 600... |
3a902822-ea34-4e4f-bc7a-a3ba6767927d | easy | khushidahi | 2026-08-16 12:31:31.316492+00:00 | succeeded | fc384cea1783dc9437f0f52c892b9ef0b88cd6102dc8b87503f731fadbde017b | 37,726 | null | """R101: final-label-grounded binary latent recurrence.
This candidate is deliberately diagnostic. It trains on evaluator-owned final
labels at several depths so that one shared transition must compose. The model
uses tensor-only base conversion to expose input bits, but never uses native
arithmetic to compute a tar... | {
"id": "3a902822-ea34-4e4f-bc7a-a3ba6767927d",
"created_at": "2026-08-16 12:31:31.316492+00:00",
"db_md5": "bbe4539e6c9e5d60b6229497ff923b06",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "2c86c012-5997-4a0d-9fe7-26d104c5feef",
"tier": "easy",
"dataset_id": "e4",
"status": "succ... | {
"score": {
"mean_loss": 2.2234530902522742,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.002291666679084301
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.214552278951085,
"example_count": 120... |
3a954627-5007-4a5f-ac55-cfb9ef9e0b2e | easy | oupadhyay | 2026-08-15 22:29:20.721023+00:00 | succeeded | 2c22960c47c04744fcec5acae44b161d3cf914642ac0d54c71de767c75befc57 | 3,994 | null | """Conditional optimizer phase 3: warm8_wd10."""
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... | {
"id": "3a954627-5007-4a5f-ac55-cfb9ef9e0b2e",
"created_at": "2026-08-15 22:29:20.721023+00:00",
"db_md5": "4aa3204f4eaa733a8e4235b72dd88b75",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "199ef6ce-7cc5-405f-87d5-0d8f13774937",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.179788927605162,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0033333333333333335
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.155991435585535,
"example_count": 600... |
3a9a341c-fe7a-43a9-890e-13415e6187f2 | easy | pepinu | 2026-08-26 17:55:41.186457+00:00 | succeeded | 74307ba76d035d71e8330e51db839434ca24ae8d3efbab3f4b12dd2ec11dd273 | 46,270 | null | """Squaring mod N as a learned permutation on CRT components.
WHY THIS SHAPE. (Z_N, x) is a commutative monoid, so every finite-dimensional
linear representation splits into multiplicative characters and squaring is
diagonal there -- the component basis is the only one in which this operation is
shallow. Two consequen... | {
"id": "3a9a341c-fe7a-43a9-890e-13415e6187f2",
"created_at": "2026-08-26 17:55:41.186457+00:00",
"db_md5": "d7f34595d7f996ce3ab1db374501668a",
"submitter": "pepinu",
"github_login": "pepinu",
"run_id": "1a9d4a26-e3b0-49c1-a32b-91dd33ab0d5d",
"tier": "easy",
"dataset_id": "e9",
"status": "succeeded",
... | {
"score": {
"mean_loss": 0.013724563643336296,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.013644742779433727,
"example_count": 90,
"... |
3aa86849-2c19-41b6-a4ce-b52a161f6373 | easy | chad-atexpedient | 2026-08-05 16:23:09.484671+00:00 | succeeded | bd5a2854c83027c05d7faa97f85ba50ef20fc26b9616c77d380c68664325976e | 12,045 | null | """T-S: learned T-conditioned selector over recurrent trajectory states."""
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,
)
LANES = 6... | {
"id": "3aa86849-2c19-41b6-a4ce-b52a161f6373",
"created_at": "2026-08-05 16:23:09.484671+00:00",
"db_md5": "0cd6bdf5228941c1c58344fe234c9d6b",
"submitter": "chad-atexpedient",
"github_login": "chad-atexpedient",
"run_id": "69092866-fab1-4fea-aa72-20916a3b4f65",
"tier": "easy",
"dataset_id": "e5",
"st... | {
"score": {
"mean_loss": 2.1787503783522277,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006250000024835268
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1808096934861667,
"example_count": 60... |
3aad40e8-d6c4-4ac6-a474-dd3ab500e90a | easy | viridale | 2026-08-13 22:23:12.624131+00:00 | succeeded | 5a9271f1223653704060a677d99104d70c338b79a1c863f8d1dbbaefb2048153 | 26,232 | null | """Joint-likelihood neural finite-state bit transducer.
hypothesis: exact final-string likelihood marginalized through one coherent anonymous
carry path will identify reusable bitwise operator state better than independent
terminal-bit CE, even when both use the same explicit learned product streams.
axis: arc... | {
"id": "3aad40e8-d6c4-4ac6-a474-dd3ab500e90a",
"created_at": "2026-08-13 22:23:12.624131+00:00",
"db_md5": "82196903d91d8a18eafcf98c3fd28728",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "a6a05d14-beab-444a-aeff-57a9f64f0485",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 16.76236629486084,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.024999999441206455
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 16.05078887939453,
"example_count": 100,... |
3ab3d13a-513a-40f1-a3c6-3789f63aa972 | easy | jordanrubin | 2026-08-11 01:30:30.396722+00:00 | succeeded | aa0c446a01b1c50a3739ac906b125fbe05d6b1cd9cf0a23d1a564e062285e8d6 | 21,670 | null | """Weight-tied MLP-cell loop over soft digit states (mlploop).
Cell = MLP over [expected digits, pairwise digit products, RNS residue
simplices (fixed differentiable mixing over Z_p), N digits]. The cell that
learns one-step modular squaring from direct pairs (52% unseen at 14k rows,
day-20 screen) inside the exact-T ... | {
"id": "3ab3d13a-513a-40f1-a3c6-3789f63aa972",
"created_at": "2026-08-11 01:30:30.396722+00:00",
"db_md5": "344e901468d7fc4aff5006fca563aa6e",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "40825773-a5de-4b3c-a863-3ac3e29adab6",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.5269918895453314,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006666666517655054
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1729412108258814,
"example_count": 60... |
3ab5d641-d355-4c9f-9b51-d7a608b16425 | easy | 0Chris5R | 2026-08-07 12:19:36.232099+00:00 | succeeded | 2294a88fac21e02693957acb20a1ca3d23e9c3c4b5b0f83c38efea9034baad9b | 20,840 | null | """Shared-digit Neural GPU with a compute-matched operator curriculum.
The model aligns decimal digits by significance, forms a learned low-rank
second-order feature map of the current state, and evolves a two-dimensional
workspace with the original convolutional-GRU equations. Each arithmetic
transition applies the ... | {
"id": "3ab5d641-d355-4c9f-9b51-d7a608b16425",
"created_at": "2026-08-07 12:19:36.232099+00:00",
"db_md5": "8c66085ff4ca0e48a710c4d40e3bb1a8",
"submitter": "Chris ",
"github_login": "0Chris5R",
"run_id": "114ffdf2-d73b-4e94-87ff-407aa553511d",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.7303053962160604,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.040416666703919565
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.663192602551037,
"example_count": 600... |
3ab5e421-ee6b-4a9c-9cea-ae9e77dbeb4e | easy | sapient-sapiens | 2026-08-23 08:08:19.602254+00:00 | succeeded | b7e98e91392cc43a7b63dc3129ad69ed65af7e91e1320da5f9a42375a1a82076 | 19,718 | null | """Approximately 10x-scale adaptive child of the D6 pair/quotient architecture.
The architecture uses the required learned additive token/group/left-relative
representation. Marker-derived field lengths additionally route categorical
digit embeddings into right-aligned lanes; this alternative place alignment is
an ex... | {
"id": "3ab5e421-ee6b-4a9c-9cea-ae9e77dbeb4e",
"created_at": "2026-08-23 08:08:19.602254+00:00",
"db_md5": "52539bd707d43027cb388da4c9030153",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "a24477c1-a3f2-4d53-8fe6-3d23aa943a9e",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 2.2196170690375423,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007083333333333334
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.22114139728482,
"example_count": 600,... |
3aba1954-a249-4b03-8a6e-0acdf4626e5d | easy | kushasareen | 2026-08-12 02:25:48.954333+00:00 | succeeded | afba4d4403d8474a5e012d368265a44162bf21d4d818def99b236bf8075f946b | 6,884 | null | """v2 weight decay bracket: wd=1.0.
T-conditioned recurrent Transformer with the measured stability fixes:
- 2 untied pre-blocks (pre-norm), 1 untied post-block, tied loop block
- prompt re-injection every loop iteration (gradient + information path)
- UT-style double post-norm inside the loop block (state bounded at ... | {
"id": "3aba1954-a249-4b03-8a6e-0acdf4626e5d",
"created_at": "2026-08-12 02:25:48.954333+00:00",
"db_md5": "4ccaf9e5fc18bd923a75e183ee7b1b86",
"submitter": "Kusha Sareen",
"github_login": "kushasareen",
"run_id": "0204a2f7-0938-41ee-89ba-5f86b4972a09",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 4.995699530757326,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005833333358168602
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.08588410706798,
"example_count": 600,
... |
3abca907-aca6-4273-98f8-dfb0315fe2b1 | easy | CodeReclaimers | 2026-08-20 15:18:28.489416+00:00 | succeeded | ecdc5bc3549f21ec4b5703f65802288e92bc05cd2c1587f9346ab4e680aba37d | 42,274 | null | """v39_keyed: keyed-modulus CRT learner — Run A of T1_CERT_PLAN_20260819.
HYPOTHESIS A (alias modulus): h1's displayed N is a stable per-identity
alias phi(M) of a hidden modulus M <= 2^20 from enumerated factor pairs
(<=10-bit primes), while x and y are displayed plainly and the step closes
per prime channel (squarin... | {
"id": "3abca907-aca6-4273-98f8-dfb0315fe2b1",
"created_at": "2026-08-20 15:18:28.489416+00:00",
"db_md5": "4734e439309d100106007d6b654ccab0",
"submitter": "CodeReclaimers",
"github_login": "CodeReclaimers",
"run_id": "79b0e2b4-a49f-4e54-b44a-7d2c20c1a2f7",
"tier": "easy",
"dataset_id": "e3",
"status... | {
"score": {
"mean_loss": 0.1649315705202102,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.99375
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.3065545397944244,
"example_count": 800,
... |
3ac34124-3110-458b-afd2-2ba4a36432e0 | easy | nikolageorgiev2000 | 2026-08-04 08:55:01.124590+00:00 | succeeded | 68fb245cb2c9a086b685cf6b34f7066b141e162ca6ba8f8a86fec5fd4d7e87b7 | 9,266 | null | """T-composed categorical-digit reasoner with fully learned pair/reduction maps."""
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,
TokenLossBat... | {
"id": "3ac34124-3110-458b-afd2-2ba4a36432e0",
"created_at": "2026-08-04 08:55:01.124590+00:00",
"db_md5": "e1255ad9173a18876df62b4caa434f25",
"submitter": "Nikola Georgiev",
"github_login": "nikolageorgiev2000",
"run_id": "68ebef64-73d6-4078-84ab-f13586ad325b",
"tier": "easy",
"dataset_id": "e3",
"s... | {
"score": {
"mean_loss": 3.8118512497386376,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006250000009313226
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.1862875791372347,
"example_count": 80... |
3ac46733-4fc0-4182-b6ff-20fd02a84c0b | easy | 0Chris5R | 2026-08-07 09:17:59.414649+00:00 | succeeded | 2b524c2d222e1807f541f698d1c9df17cd47f5077db840e1a982890033fcb02b | 30,585 | null | """Shared-digit Neural GPU with late full-objective EMA evaluation.
The model aligns decimal digits by significance, forms a learned low-rank
second-order feature map of the current state, and evolves a two-dimensional
workspace with the original convolutional-GRU equations. Each arithmetic
transition applies the sam... | {
"id": "3ac46733-4fc0-4182-b6ff-20fd02a84c0b",
"created_at": "2026-08-07 09:17:59.414649+00:00",
"db_md5": "b036ec836fac35a5f3465096501360a5",
"submitter": "Chris ",
"github_login": "0Chris5R",
"run_id": "a6c0fe87-775d-4aa5-a924-396c351600fd",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.3180081318521055,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008749999999999999
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.161729108057749,
"example_count": 600... |
3add00f7-c194-4949-b208-20a31f04934f | easy | yashkant | 2026-08-14 15:30:27.508215+00:00 | succeeded | fe7aeaca1989c7966b94c730f30e332c50be747e3d658e37b0e68f929b7f15f0 | 38,960 | null | """Round1818 parent-preserving temporal Anderson-Dykstra mixing.
The complete Round1792 AdamW5/EMA.995 W8/D2 carrier remains byte-identical at
a matched seed. Initialization of the added temporal mixer is isolated from
the inherited CPU RNG stream, and one exact-zero scalar gate keeps the scored
path equal to the par... | {
"id": "3add00f7-c194-4949-b208-20a31f04934f",
"created_at": "2026-08-14 15:30:27.508215+00:00",
"db_md5": "8eb54a571912ddbb1e6aa8dc7b3c2208",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "755e7f93-a2b4-454a-ac7f-295eebe43cb7",
"tier": "easy",
"dataset_id": "e3",
"status": "succeed... | {
"score": {
"mean_loss": 2.1205559220834838,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.014375000027939677
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1208835875832537,
"example_count": 80... |
3ae04eb3-a24c-4301-ad9b-51c93e9cda50 | easy | TZZheng | 2026-08-06 21:07:17.333496+00:00 | succeeded | a7f51e1b4f35491d7245225398862605760974d0531f191864e1aa06cbe98239 | 7,301 | null | """Self-contained standard Transformer scaling-sweep submission."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
)
VARIANT = {'block... | {
"id": "3ae04eb3-a24c-4301-ad9b-51c93e9cda50",
"created_at": "2026-08-06 21:07:17.333496+00:00",
"db_md5": "0640470de367f227b4d66e618db11928",
"submitter": "TZZheng",
"github_login": "TZZheng",
"run_id": "1322ddfc-ef4b-4849-8118-14e95b159c28",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 3.117733061313629,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.056666663847863674
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.310667037963867,
"example_count": 100,... |
3ae21bac-bf10-4757-b4c5-e47415f5f502 | easy | AFeinfield | 2026-08-27 22:19:30.350554+00:00 | succeeded | 82ee9a809fa3260504112ca47ed47d24df8557f06ade4766f27ad749a6c78d2e | 12,086 | null | """155: 152 + digit-decode init scaled to the working range
(randn*3.0 — still random; the decode no longer has to grow 100x).
151 notes: throughput/learnability pass.
Dynamic alignments (m <= batch N-width), carry passes 6 post-square / 3
in reduction, single correction try, explicit y/10 feature (carry is
linear in... | {
"id": "3ae21bac-bf10-4757-b4c5-e47415f5f502",
"created_at": "2026-08-27 22:19:30.350554+00:00",
"db_md5": "a3e2330961bb7e57846c9220189e9a1c",
"submitter": "AFeinfield",
"github_login": "AFeinfield",
"run_id": "dbff30e3-4d86-4865-a025-f05e24e9c17a",
"tier": "easy",
"dataset_id": "e2",
"status": "succ... | {
"score": {
"mean_loss": 2.230340838432312,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2263360023498535,
"example_count": 300,
"exac... |
3ae7c384-0294-4c98-8aa3-0ae932b424f6 | easy | yashkant | 2026-08-19 02:19:18.713620+00:00 | succeeded | 4bf48738e795a7ee1545a7695c8f22652e00337f245c69b97ce43d83ecaf31c1 | 46,509 | null | """Round2202 factorizes one tiny recurrent correction into residue and carry.
The measured correction-reset official carrier stays intact. A pooled
four-value/four-carry state runs one shared transition twice per parsed-T
cycle and reinjects its immutable input halves on every call. Carry is
learned from signed resi... | {
"id": "3ae7c384-0294-4c98-8aa3-0ae932b424f6",
"created_at": "2026-08-19 02:19:18.713620+00:00",
"db_md5": "a1a52e211c1ef193c4ff2934dfdc88e1",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "c8cb10ff-d78d-4036-b22c-abc8deae0f71",
"tier": "easy",
"dataset_id": "e5",
"status": "succeed... | {
"score": {
"mean_loss": 2.1483980671201315,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008333333345750968
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1527680206726485,
"example_count": 60... |
3ae7f901-e1c1-4a16-baab-094c9b0af1b2 | easy | sapient-sapiens | 2026-08-09 21:36:39.700486+00:00 | succeeded | 5bb4e61f0c3687019fbff8d90d6b56ec8467719a08ba00c1fe0e69800328728e | 15,567 | null | """T-independent categorical recurrence with learned latent-depth selection."""
from __future__ import annotations
import math
import time
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
Toke... | {
"id": "3ae7f901-e1c1-4a16-baab-094c9b0af1b2",
"created_at": "2026-08-09 21:36:39.700486+00:00",
"db_md5": "d5518602a2c831b1a0df7d2053fd26b5",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "52cb58fe-8622-4021-ba4f-cd40b6942a57",
"tier": "easy",
"dataset_id": "e2",
"status": "su... | {
"score": {
"mean_loss": 2.305056333802526,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004166666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2967537432459033,
"example_count": 300... |
3aecdf84-8c92-4547-84eb-aabda17b6077 | easy | oupadhyay | 2026-08-07 12:01:40.481001+00:00 | succeeded | 97cb568c45894404023fc51613bed22228a97a9383614c865f98daac3fb4d8e2 | 9,350 | null | """Compressed axial baseline with zero-initialized global pair summaries."""
import torch
import torch.nn.functional as F
from torch import nn
from benchmark import OptimizerBundle, Submission, assert_model_state
PAD, N, X, T, ANS, DIGIT = 0, 2, 3, 4, 5, 7
MAX_STEPS, CHANNELS, STATE_ELEMENTS = 64, 64, 54954
class ... | {
"id": "3aecdf84-8c92-4547-84eb-aabda17b6077",
"created_at": "2026-08-07 12:01:40.481001+00:00",
"db_md5": "6b8c96938ca5fa8f2a3daf59e9a6b589",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "fcc777c6-c723-4e0e-b691-969fb11c037a",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.569478794974806,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.002916666679084301
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.582455228796989,
"example_count": 600,... |
3afc05f5-7396-4c45-967e-5a52c7056a34 | easy | isaac0804 | 2026-08-06 15:24:49.786004+00:00 | succeeded | 07d60888eaa823230461af42aa47317e8e30791ad92cf305a2937ca5d55d48ae | 6,604 | null | """E3 candidate #6: narrow continuous information bottleneck (128 -> 8 -> 128
squeeze), mid-stack, on every token position -- fixes a capacity-escape flaw
found in e3_vq_bottleneck.
e3_vq_bottleneck quantized every one of ~13 sequence positions independently
against a 64-entry codebook. That is not a real bottleneck: ... | {
"id": "3afc05f5-7396-4c45-967e-5a52c7056a34",
"created_at": "2026-08-06 15:24:49.786004+00:00",
"db_md5": "93e773d92a653185b9c14ec75c5913fb",
"submitter": "Isaac Yong",
"github_login": "isaac0804",
"run_id": "ec313cc9-df0f-4cb2-ba10-cb723639cfaa",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 8.050480024361121,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007500000004656613
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.856465736055354,
"example_count": 800,... |
3afd2f02-4d70-4277-bbf5-2a3f95e61c90 | easy | shirvani-jr | 2026-08-30 23:56:56.059399+00:00 | succeeded | 6b6108fc2b74c3ea42bc1eb17b33be22ea62a61a3784738e9e3841cb938c8c93 | 19,005 | 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 = 11
PERIOD_MAXIMUM = 100
E... | {
"id": "3afd2f02-4d70-4277-bbf5-2a3f95e61c90",
"created_at": "2026-08-30 23:56:56.059399+00:00",
"db_md5": "98a1b794d9d15babf69521185b874db6",
"submitter": "Ali",
"github_login": "shirvani-jr",
"run_id": "03104e9f-8480-4d09-b6a2-a60d1e262056",
"tier": "easy",
"dataset_id": "e4",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.2612608426471295,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.027604166716337204
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.481648467224988,
"example_count": 120... |
3b0de07b-e939-4a97-b457-cc3805a9cf41 | easy | jordanrubin | 2026-08-03 21:37:46.240875+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": "3b0de07b-e939-4a97-b457-cc3805a9cf41",
"created_at": "2026-08-03 21:37:46.240875+00:00",
"db_md5": "ef7d67d872a65aa96e81ca53991ab27d",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "4bc010b9-6790-466b-92b8-815fd1169675",
"tier": "easy",
"dataset_id": "e2",
"status": "s... | {
"score": {
"mean_loss": 3.426506173458692,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.2208333301047484
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.504126330562862,
"example_count": 300,
... |
3b0fc1ce-7c0e-41c0-8718-81699476a3c2 | easy | DDanlov | 2026-08-28 21:50:32.080978+00:00 | succeeded | 1ee00c3e3f7802efd9f0f6659f676708744f14febe5962649530211e21667b6b | 14,938 | null | """
Trial 19: Model B: D=384, H=6, M=16, u=2 (T=32)
"""
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": "3b0fc1ce-7c0e-41c0-8718-81699476a3c2",
"created_at": "2026-08-28 21:50:32.080978+00:00",
"db_md5": "abede554170b67712c2a4e0377d76528",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "43b9f2a2-2581-4348-8994-57fb387a194b",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.83321353302519,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0025
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8332135109356167,
"example_count": 600,
"ex... |
3b117588-3b23-4cf2-8882-8733b11367fa | easy | viridale | 2026-08-03 09:35:08.918368+00:00 | succeeded | 2ab7cba2961297bfd1d3fd25e786d3b53a173f72d2647e0b47600e40a5f5622b | 134,838 | null | """Prompt-depth I/O wrappers over the exact global source posterior.
hypothesis: the hidden source edit may use prompt T in initialization or the
returned observation, loci still absent after the T-conditioned clock and
transition closure despite being explicitly elevated by the API hint.
axis: arch
target: re... | {
"id": "3b117588-3b23-4cf2-8882-8733b11367fa",
"created_at": "2026-08-03 09:35:08.918368+00:00",
"db_md5": "317a01daef78471923437720d63f2031",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "7ecebef3-4c88-4c39-90f4-0d3409e818ec",
"tier": "easy",
"dataset_id": "e5",
"status": "succe... | {
"score": {
"mean_loss": 0.003014760072946975,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.0030147603728898254,
"example_count": 600,
... |
3b13af7a-7808-456c-9d33-129cedeee818 | easy | erdavis0 | 2026-08-24 09:14:36.383647+00:00 | succeeded | 489dda64c8ff330175bd25f897aab56f96331ab4e8010edd0ffa7d00d2da65bb | 9,227 | null | """Bidirectional learned Horner registers with coupled first- and second-order state."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
... | {
"id": "3b13af7a-7808-456c-9d33-129cedeee818",
"created_at": "2026-08-24 09:14:36.383647+00:00",
"db_md5": "80e0b97bc03413b4cce19f20f78b5a4b",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "8bddadc0-b46c-442d-9da9-8c72fb9bf912",
"tier": "easy",
"dataset_id": "e9",
"status": "succeeded",... | {
"score": {
"mean_loss": 7.461805582046509,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05111111141741276
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.354795455932617,
"example_count": 90,
... |
3b1d46c8-66c7-4557-8852-711cf124ded0 | easy | Maxwells-Demons | 2026-08-11 10:28:49.485607+00:00 | succeeded | 7290a066f1b354b1aaf687c8d52e2665bbb7498162209fdf5f78aede7d7114d4 | 24,771 | null | """The transition as one wide MLP, queried on a free orbit of the input, with every
query supervised on its own.
Three measurements drive this shape.
1. The tied attention block (v43-v46) tops out around mean 0.75-0.79 on e1 while
the older flat-MLP transition reached 0.87 on the same data. Attention mixes
slo... | {
"id": "3b1d46c8-66c7-4557-8852-711cf124ded0",
"created_at": "2026-08-11 10:28:49.485607+00:00",
"db_md5": "71676094ff318417c65789994ea8ad72",
"submitter": "Maxwells_Ayakashi",
"github_login": "Maxwells-Demons",
"run_id": "182b3a0c-e0a6-4989-b936-9374e1ba0c4d",
"tier": "easy",
"dataset_id": "e3",
"st... | {
"score": {
"mean_loss": 2.4002820084050045,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.3900000002980232
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.017852335453635,
"example_count": 800,
... |
3b1d8a24-eafc-4028-9cbf-34c673ea460f | easy | jordanrubin | 2026-08-05 20:53:38.700338+00:00 | succeeded | cca11f9ae8ee62059d3fbba8e4bdb9001b823bd08f9011a4d99bd689f6dd303f | 24,768 | 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": "3b1d8a24-eafc-4028-9cbf-34c673ea460f",
"created_at": "2026-08-05 20:53:38.700338+00:00",
"db_md5": "d03721b4820a08aef275f13d4ee162a4",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "48bc4c40-c187-4fea-b6dc-585a4dc9cb89",
"tier": "easy",
"dataset_id": "e2",
"status": "s... | {
"score": {
"mean_loss": 4.1150767979866325,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.1687499947147444
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.098171230807952,
"example_count": 300,
... |
3b2c8033-76d0-447b-ab80-49b3bc94e550 | easy | liam-gb | 2026-08-16 19:59:56.915304+00:00 | succeeded | 73377d899037f347ce9872a62e038450eb59f5c03a550d9d438f7ba6dc0e4190 | 18,288 | null | """rns_pm_gather_ct_widet_l32nore_muon: single change vs rns_pm_gather_ct_widet_l32nore —
the optimizer. Dense 2-D matrices step with Muon (momentum + Newton-Schulz
orthogonalisation, the standard practice split); lookup tables and all 1-D /
scalar parameters keep AdamW exactly as the base ships them (tables lr 3e-2,
n... | {
"id": "3b2c8033-76d0-447b-ab80-49b3bc94e550",
"created_at": "2026-08-16 19:59:56.915304+00:00",
"db_md5": "8e64703b1ca5e11db9484c1cff485ce2",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "3ce713ee-db8e-4192-b808-46923ce44fb7",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 6.81921058329376,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05333333331470688
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.6944990158081055,
"example_count": 600,
... |
3b2d1f03-fe95-499e-bcd7-2a23482265ec | easy | khushidahi | 2026-08-03 20:47:03.555077+00:00 | succeeded | 625e8f88a95810d12a7116d99f48b29c55685a85c01ccb3498c2586fb4d7925a | 15,250 | 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": "3b2d1f03-fe95-499e-bcd7-2a23482265ec",
"created_at": "2026-08-03 20:47:03.555077+00:00",
"db_md5": "aadf9eb278ba6397a5d544ff0c6a90e5",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "730b330b-9f0b-494d-adc1-c87ea9ba5519",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.2063772678375244,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008333333767950535
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2008235454559326,
"example_count": 60... |
3b3d0cc6-67ef-467b-9161-ea35cb251330 | easy | khushidahi | 2026-08-21 17:23:51.906869+00:00 | succeeded | 65a94e9ab5682c3a07dce8fdf21f2e1a91a1ecb9829898bd797986bf1735410d | 29,570 | null | """R344: full-clock two-visit inner-recurrence arm of R340.
This submission tests the strongest rules-safe generic baseline suggested by
the public competition discussion. It does not implement squaring, modular
reduction, or any other task solver. Numeric prompt fields are only rearranged
onto equal-length, right-a... | {
"id": "3b3d0cc6-67ef-467b-9161-ea35cb251330",
"created_at": "2026-08-21 17:23:51.906869+00:00",
"db_md5": "226fa30987f323507eff2970a54d9890",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "fc27cad3-e4d1-4037-95ec-ac30b28c990a",
"tier": "easy",
"dataset_id": "e6",
"status": "succ... | {
"score": {
"mean_loss": 1.0138767659664154,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.8015766143798828
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.8087849020957947,
"example_count": 60,
... |
3b43cd21-705f-4596-8e10-ae9d7ffd29f2 | easy | viridale | 2026-08-13 06:21:06.994027+00:00 | succeeded | e40e77242845f015063826581b9ddf99f29621c8953d71aff31f0bd8c76eac20 | 22,429 | null | """Symmetry-broken tensor-renormalized stochastic one-step operator.
hypothesis: a depth-64 gradient audit of v528 found healthy base-operator
gradients but only 1e-8 gradients in its power compressors: near-uniform softmax
fusion makes every virtual state exchangeable. Use a randomly initialized hard
stra... | {
"id": "3b43cd21-705f-4596-8e10-ae9d7ffd29f2",
"created_at": "2026-08-13 06:21:06.994027+00:00",
"db_md5": "d066dc02722390615030e2e75219a00d",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "c3cf8ad5-eeaf-4eb1-b707-020a96d63b12",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 15.760794386304822,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.018333332998057207
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 14.984068870544434,
"example_count": 10... |
3b458aec-1417-45f4-8de2-745692ba8765 | easy | k-penchev | 2026-08-05 02:43:34.876700+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": "3b458aec-1417-45f4-8de2-745692ba8765",
"created_at": "2026-08-05 02:43:34.876700+00:00",
"db_md5": "eaac69fe345f82973fcf392c504a40ff",
"submitter": "Kaloyan Penchev",
"github_login": "k-penchev",
"run_id": "bd4dee51-a643-4362-af02-a137b8611084",
"tier": "easy",
"dataset_id": "e5",
"status": "... | {
"score": {
"mean_loss": 2.7235996911734666,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0033333333457509677
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8735547685836997,
"example_count": 6... |
3b4eccf4-89bc-4e9e-82a7-23c29ce58cb2 | easy | AdityaVG13 | 2026-08-13 04:21:03.208178+00:00 | succeeded | 8423d031d7e713f171fb050d73ad91127022aa2f22c8cdc50027d3e179140b50 | 16,467 | null | """TAPE: a tied arithmetic program executor for One Layer Deeper.
The deterministic shell only separates the public numeric fields, stores decimal
values on least-significant-digit-first tapes, and applies one learned transition
the requested number of times. The transition itself is a generic neural
program: it cont... | {
"id": "3b4eccf4-89bc-4e9e-82a7-23c29ce58cb2",
"created_at": "2026-08-13 04:21:03.208178+00:00",
"db_md5": "9b1885eb929fe8904c89f537654f600c",
"submitter": "AdityaG",
"github_login": "AdityaVG13",
"run_id": "84365c17-57e5-4a11-91d2-d76bdbb859e0",
"tier": "easy",
"dataset_id": "e1",
"status": "succeed... | {
"score": {
"mean_loss": 2.085818290710449,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0566666666418314
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.050222158432007,
"example_count": 100,
... |
3b50b98c-8634-4c0e-b88a-9324e961de48 | easy | DDanlov | 2026-08-15 20:49:08.199666+00:00 | succeeded | 3eb517cb9667863720ce75d439fd67ebc36b79a26534e0b145dad12284313550 | 17,950 | null | """
Official Submission for One Layer Deeper Challenge
Architecture: Deep Equilibrium Port-Hamiltonian Deliberation Network (Model C v3 / DEPH-Net)
- Input: Multi-Scale Linear Harmonic Torus Frequency Encoder (omega_j = 1..64)
- Recurrent Core: Continuous Port-Hamiltonian Dynamical Flow dz/dt = (J - R) * grad_E
* J =... | {
"id": "3b50b98c-8634-4c0e-b88a-9324e961de48",
"created_at": "2026-08-15 20:49:08.199666+00:00",
"db_md5": "623970f9d216ffb45b2ef4f8507de941",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "fe76881e-a470-463c-a1bd-3a4f5b1c601d",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.3486146243812476,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03666666680326064
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.297792738053336,
"example_count": 100,... |
3b53f2c3-526a-45fd-b03d-dec0f7f2d8c0 | easy | oupadhyay | 2026-08-25 07:29:17.115889+00:00 | succeeded | 2bb0a0ad19ee5b135c236dee81d211d3a8d74edb5f06265a50d9bf8c900ef29c | 4,091 | null | """Dynamic-width shared conditional quadratic recurrence."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import nn
from benchmark import OptimizerBundle,Submission,TokenLossBatch,assert_model_state
E,H,K=32,16,32
class C:
def __init__(s,vocab_size,max_seq_len):s.vocab_siz... | {
"id": "3b53f2c3-526a-45fd-b03d-dec0f7f2d8c0",
"created_at": "2026-08-25 07:29:17.115889+00:00",
"db_md5": "a6fe18d7aa099e57d437966517e7ffa6",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "2058d985-c652-4852-9e60-b143647b01b7",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.2869872596417826,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0054166666790843015
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.134454908926925,
"example_count": 60... |
3b55bd9e-4b37-40ac-b3e3-c6312cbab54f | easy | AFeinfield | 2026-08-28 00:44:25.022092+00:00 | succeeded | ac26defeaf4f0163a71bd057aae9291848ee382f3742358a0df94eab7f213c66 | 15,305 | null | """177: 174 + SCHEDULED INTEGER SNAP (soft -> hard at call 400 + eval).
The frontier's test accuracy equals per-op accuracy ^ ops-per-row: the
missing multiplier is per-op EXACTNESS. 170 proved snapping from step 1
removes early partial credit; scheduling it after the operators reach
~+-0.5 accuracy (and always at eva... | {
"id": "3b55bd9e-4b37-40ac-b3e3-c6312cbab54f",
"created_at": "2026-08-28 00:44:25.022092+00:00",
"db_md5": "9eb28c6f655044b7dc2489eab3a591c9",
"submitter": "AFeinfield",
"github_login": "AFeinfield",
"run_id": "ad011784-bfed-433b-8283-aa01a7fb1415",
"tier": "easy",
"dataset_id": "e1",
"status": "succ... | {
"score": {
"mean_loss": 2.0551676750183105,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.07999999821186066
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.014720916748047,
"example_count": 100,... |
3b59893f-e594-45e4-9f86-6ee8a8df45e6 | easy | liam-gb | 2026-08-13 00:44:44.661352+00:00 | succeeded | eb2504bb7d0157e632cf799a1f18f53674416b6d2c2037173c708bf726abbc5b | 14,358 | null | """rns_pm_gather_ct_widet: dataset-sized digit head, dead loop tail trimmed.
The wide readout is what makes long answers reachable; the trim is what pays
for it, by not running loop iterations whose update is multiplied by zero.
Base docstring:
Single change vs rns_pm_gather_ct: the readout carries one
distance-from... | {
"id": "3b59893f-e594-45e4-9f86-6ee8a8df45e6",
"created_at": "2026-08-13 00:44:44.661352+00:00",
"db_md5": "1a548bea555b987cf0a175795dc3aa83",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "ba381970-9596-44e6-ab53-29d80519d581",
"tier": "easy",
"dataset_id": "e2",
"status": "succeeded"... | {
"score": {
"mean_loss": 6.790793299674988,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.393125019967556
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 11.702922821044922,
"example_count": 300,
... |
3b5d91d1-c0e4-41ce-bb51-717e88f21142 | easy | karanganesan | 2026-08-05 10:35:25.455046+00:00 | succeeded | e0f4d1398e0528baa904d2809b64050c6c5bac0fe85dbc775ef5cb98bd587fb6 | 22,081 | null | """Parametric looped-transformer family (P1).
One weight-tied transformer block applied k times in latent space. Config
flags cover four P1 families with one file:
looped recall=0 gated=0 tfilm=0 plain weight-tied loop
looped-recall recall=1 re-inject the input embedding each
... | {
"id": "3b5d91d1-c0e4-41ce-bb51-717e88f21142",
"created_at": "2026-08-05 10:35:25.455046+00:00",
"db_md5": "1ae9bcd4922ba8ee3a36d0d013bf0eca",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "cec38839-b7b5-4bb4-8634-44af4cd5a740",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 9.673227310180664,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.011666666716337204
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.388511657714844,
"example_count": 100,... |
3b611d12-b9ba-454f-a352-d887fae761f8 | easy | briheet | 2026-08-07 01:44:33.980614+00:00 | succeeded | 96cf2f738d7911de387a6a21d648a7272ef717cabfe022aaa242be09a2ce6761 | 5,030 | 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 = 96
NUM_HEADS = 4
X_... | {
"id": "3b611d12-b9ba-454f-a352-d887fae761f8",
"created_at": "2026-08-07 01:44:33.980614+00:00",
"db_md5": "3c362223c46ba6cd4b80e246e978d77e",
"submitter": "Aizenn",
"github_login": "briheet",
"run_id": "b9d30b18-1c47-470d-aa32-4c8c6be218a6",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",... | {
"score": {
"mean_loss": 2.596865653991699,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03999999910593033
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.598268747329712,
"example_count": 100,
... |
3b67413d-960d-4af4-a27e-f50790908468 | easy | rougee | 2026-08-25 01:18:47.056172+00:00 | succeeded | 4513f81270d2f6904c293deebeb4eabd636767e31d75dc2b14c5ff597f1f3858 | 11,321 | null | """Official Submittable Champion for One Layer Deeper.
Pareto Champion Architecture:
- Core Type: Degree-2 Quadratic Polynomial Recurrent Core (h_{t+1} = (W_a h) * (W_b h) * (1/sqrt(d)) + W_lin h + W_p h_prompt)
- Lie Group Metric: Dynamic Skew-Symmetric Matrix Exponential W_orth = exp(S - S^T) anchored at ANS prompt ... | {
"id": "3b67413d-960d-4af4-a27e-f50790908468",
"created_at": "2026-08-25 01:18:47.056172+00:00",
"db_md5": "3d60031c6be935c4b27ef62c7a9ed8bf",
"submitter": "Rougee",
"github_login": "rougee",
"run_id": "bf037d87-7d90-4e79-baa6-17cbc02887b2",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
... | {
"score": {
"mean_loss": 6.159562528133392,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05833333171904087
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 10.358636856079102,
"example_count": 100,... |
3b6cc33e-6593-4616-8732-25a53e812bf0 | easy | jordanrubin | 2026-08-14 01:12:05.396425+00:00 | succeeded | fc98447fc19a656845f48580dd88f53c0b2d7d00c35e32a8fe2d3dfff79a73f4 | 29,581 | null | """Weight-tied MLP-cell loop over soft digit states (mlploop).
Cell = MLP over [expected digits, pairwise digit products, RNS residue
simplices (fixed differentiable mixing over Z_p), N digits]. The cell that
learns one-step modular squaring from direct pairs (52% unseen at 14k rows,
day-20 screen) inside the exact-T ... | {
"id": "3b6cc33e-6593-4616-8732-25a53e812bf0",
"created_at": "2026-08-14 01:12:05.396425+00:00",
"db_md5": "d08ba312cd6460ffc449c5d2476626a5",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "20c691d8-06af-43a1-8996-fa3e0ad157bc",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.8467480656076676,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.018749999503294628
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.199169605302169,
"example_count": 600... |
3b6ec722-bfa0-4fe7-88ff-17d43a26624e | easy | chad-atexpedient | 2026-08-05 15:01:32.202848+00:00 | succeeded | 19fceed08cfca04c2eb5a85658c2083ca5897b25288c14b31f560345bb6ef62a | 13,067 | null | """R031G: hard-choice processor with two persistent typed global variables."""
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,
)
LANES ... | {
"id": "3b6ec722-bfa0-4fe7-88ff-17d43a26624e",
"created_at": "2026-08-05 15:01:32.202848+00:00",
"db_md5": "bbc73c0dbc8abd8f22225f1d45a1b707",
"submitter": "chad-atexpedient",
"github_login": "chad-atexpedient",
"run_id": "6f2130a8-135a-47df-bb81-1fc88832f090",
"tier": "easy",
"dataset_id": "e3",
"st... | {
"score": {
"mean_loss": 2.1823152243067856,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00125
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1787602544382536,
"example_count": 800,
... |
3b6ed1d0-bc35-47ba-bd8b-db1ff52a7d04 | easy | benjaminW2025 | 2026-08-17 18:31:42.764634+00:00 | succeeded | ed79a247b14f2557860207b7e3dc9b2820d2e4ac6f3951a47fb1723c7159e92d | 4,654 | null | """Basic single-pass Transformer with PyTorch AdamW."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
)
D_MODEL = 256
NUM_HEADS = 4
... | {
"id": "3b6ed1d0-bc35-47ba-bd8b-db1ff52a7d04",
"created_at": "2026-08-17 18:31:42.764634+00:00",
"db_md5": "44c1bd17a8b263801aa442008de7a8eb",
"submitter": "benjawesome",
"github_login": "benjaminW2025",
"run_id": "2f89cdba-1c1c-424e-9e74-820ac89e43d7",
"tier": "easy",
"dataset_id": "e5",
"status": "... | {
"score": {
"mean_loss": 2.6771715065782002,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00541666670391957
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8443192601738487,
"example_count": 600... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.