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 |
|---|---|---|---|---|---|---|---|---|---|---|
3fa5e222-528f-41ce-9144-961161de584f | easy | pculaf | 2026-08-15 17:51:17.428668+00:00 | succeeded | 6b5f7bec1e47d978ebf3ac7b266f71930957747a138500857e0f061fd5d7101f | 8,354 | null | """Muon-optimized recurrent Transformer for the E6 activation/depth sweep."""
from __future__ import annotations
from collections.abc import Iterable
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submissio... | {
"id": "3fa5e222-528f-41ce-9144-961161de584f",
"created_at": "2026-08-15 17:51:17.428668+00:00",
"db_md5": "39089c1185e17913d9460f7e644cd6e4",
"submitter": "Pavle Culafic",
"github_login": "pculaf",
"run_id": "24f1e0ca-2eff-48cb-8b25-502c541a0c66",
"tier": "easy",
"dataset_id": "e6",
"status": "succe... | {
"score": {
"mean_loss": 1.9792985320091248,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.021846847608685493
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.999637484550476,
"example_count": 60,... |
3fa847e9-142a-494b-8277-c851281cc55a | easy | sapient-sapiens | 2026-08-17 19:21:15.116242+00:00 | succeeded | 0913b8c59f49e06a226560c1427ac2717d25466d0252742367888c3a58ce2da9 | 28,656 | null | """v40 lineage: T-free operator, N memory, PonderNet mix, no T in the loss.
Field-local encoder, T-free core, countable halt, answer-slot readout.
N is seeded and restored every step. T is never parsed in forward or loss.
"""
from __future__ import annotations
import math
import time
import torch
import torch.nn.fu... | {
"id": "3fa847e9-142a-494b-8277-c851281cc55a",
"created_at": "2026-08-17 19:21:15.116242+00:00",
"db_md5": "9ed62cf0be6dd3c368fb2ffd302295e6",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "1c50289e-35d7-4f22-8d7f-70e586d0ab73",
"tier": "easy",
"dataset_id": "e9",
"status": "su... | {
"score": {
"mean_loss": 1.9197689335352228,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03777777808407942
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.9537475109100342,
"example_count": 90,... |
3fa982f7-1113-463a-bb0e-89280f8151a6 | easy | newjordan | 2026-08-24 14:51:41.691708+00:00 | succeeded | 40fabd3a2ad74a9d0374e500eea4e580f11565a981bb9814af775398ce9dc00f | 29,672 | null | """Joint-pair transducer with a tied bidirectional dyadic prefix mixer.
Decimal N and X fields are packed into fixed-width little-endian categorical
registers. Each learned transition retains all 144 ordered pairs of X places
through a random learned categorical relation table. A learned nonseparable
position router... | {
"id": "3fa982f7-1113-463a-bb0e-89280f8151a6",
"created_at": "2026-08-24 14:51:41.691708+00:00",
"db_md5": "76b00730937bcc61c536f178214c23b7",
"submitter": "Frosty40",
"github_login": "newjordan",
"run_id": "23c38c37-6b6f-4c86-b581-3a7b7b91e481",
"tier": "easy",
"dataset_id": "e5",
"status": "succeed... | {
"score": {
"mean_loss": 4.741273910860093,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.08
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.603316972073952,
"example_count": 600,
"exac... |
3fb03f2f-cf5c-40c9-bb6d-7025186da747 | easy | DDanlov | 2026-08-20 11:26:42.793596+00:00 | succeeded | ff4722ee88c04ad0eff67549689dbb8caec4ad49b2592252f52b1af8a701266f | 17,808 | null | from __future__ import annotations
import math
import time
from typing import Any, Optional, Tuple
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Optimizer
try:
from benchmark.api import ModelSpec, OptimizerBundle, OptimizerSpec, Submission, assert_model_state
except Imp... | {
"id": "3fb03f2f-cf5c-40c9-bb6d-7025186da747",
"created_at": "2026-08-20 11:26:42.793596+00:00",
"db_md5": "28b0bb41c0e437d2aa84c9e99862f580",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "e8860594-0aa4-4f88-9a95-9bb124ec56ce",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.287452014397237,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.012500000012417634
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.282691561053152,
"example_count": 600,... |
3fb4bec5-e4f1-4a07-95aa-eda1b939353e | easy | jordanrubin | 2026-08-14 17:03:36.685520+00:00 | succeeded | ce48f872679d82707db7fc4f24ddbcf58d1afa8a5604030210935de426033a3a | 23,106 | null | """Clean categorical pair tree with immutable-modulus recall.
The model parses public decimal fields into right-origin categorical slots.
One learned, swap-symmetric pair encoder turns unordered state-slot pairs into
evidence. A single position-shared reducer composes all leaves through a
balanced tree; the unchanged... | {
"id": "3fb4bec5-e4f1-4a07-95aa-eda1b939353e",
"created_at": "2026-08-14 17:03:36.685520+00:00",
"db_md5": "8fa78a6356fa27f85ee1aefe5c55ffaf",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "25d26020-7869-49bf-9732-22e70ad20824",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.149505256320797,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.010000000012417635
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1494620637508786,
"example_count": 600... |
3fb786ca-56ba-4c5e-8983-58b866368112 | easy | Yalyenea | 2026-08-11 21:32:54.681526+00:00 | succeeded | 6d78943e5f57f580ac7c177a353da4fb6a5993d04baab0720705b6f27c87e5b7 | 12,919 | null | """R4 full-bandwidth candidate with RoPE and a 2000-step cosine horizon."""
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": "3fb786ca-56ba-4c5e-8983-58b866368112",
"created_at": "2026-08-11 21:32:54.681526+00:00",
"db_md5": "e1ab4883f30066856bcfaaad405fd185",
"submitter": "yfff",
"github_login": "Yalyenea",
"run_id": "76fe7573-9253-431c-b00f-3176ed392216",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded",
... | {
"score": {
"mean_loss": 8.8658447265625,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.014999999664723873
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.701698303222656,
"example_count": 800,
... |
3fb93b5b-84df-4e74-b5b6-3e56c2ae0e53 | easy | lpbb | 2026-08-20 15:34:23.794308+00:00 | succeeded | cb09ee1057787da3aac251442bfd03969a5fac2cfaefaeff9ae490f69b1c179e | 15,023 | null | """abacus_rope_loop_t + a continuous flow-matching sub-objective on the loop state.
Main objective unchanged: the shared block runs T times (T parsed from each
row's own tokens), cross-entropy on the answer digits.
Flow-matching sub-objective (training only). At every loop step each active
row's state is, with probab... | {
"id": "3fb93b5b-84df-4e74-b5b6-3e56c2ae0e53",
"created_at": "2026-08-20 15:34:23.794308+00:00",
"db_md5": "92efa14ebbb6432e4c86f39cce17cb55",
"submitter": "Lpbb",
"github_login": "lpbb",
"run_id": "6b932b99-0c3f-486d-9a55-ce4e78be20f8",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
"s... | {
"score": {
"mean_loss": 2.4069327061015766,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03166666643073161
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8980495929718018,
"example_count": 100... |
3fc50e49-7810-4e9b-86ad-7ae093874e99 | easy | TZZheng | 2026-08-07 05:34:33.778462+00:00 | succeeded | f08fd05b7074612ce3672b1a01797678731347344d792100ce91c2247aa009d3 | 12,532 | 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": "3fc50e49-7810-4e9b-86ad-7ae093874e99",
"created_at": "2026-08-07 05:34:33.778462+00:00",
"db_md5": "12ae608771b739cfdc6ff492964f4a39",
"submitter": "TZZheng",
"github_login": "TZZheng",
"run_id": "86815447-8ecc-42aa-9b70-b0f506c6d5f9",
"tier": "easy",
"dataset_id": "e4",
"status": "succeeded"... | {
"score": {
"mean_loss": 8.092047863473487,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004166666679084301
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.19142678773516,
"example_count": 1200,... |
3fc8404d-08f1-4266-9769-a1cb46a23b6a | easy | sapient-sapiens | 2026-08-22 19:10:14.549972+00:00 | succeeded | 7177f8b122c2a5d1ec12ea87f6a89475bb120b3dcbe7691a48b03886a89ac7e4 | 19,828 | null | """Batch-64 LR-2e-3 eight-layer T-private Q+V TRM for E5.
Research hypothesis: approximately linear LR scaling to 2e-3 best exploits
batch 64's high example throughput without destabilizing HybridMuon+SAM.
Primary experimental variable: peak learning rate, 1.2e-3 versus 2e-3, at fixed
batch 64 with model, optimizer, ... | {
"id": "3fc8404d-08f1-4266-9769-a1cb46a23b6a",
"created_at": "2026-08-22 19:10:14.549972+00:00",
"db_md5": "517fddc2cacc32d32c8e0b36fce04fc0",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "fb3790c5-9848-4489-a45a-a2f0e85935e1",
"tier": "easy",
"dataset_id": "e5",
"status": "su... | {
"score": {
"mean_loss": 2.288795522711119,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0141666666790843
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2913822516999436,
"example_count": 600,
... |
3fceee74-de4c-409b-8c2c-f756dfc3b5c6 | easy | erdavis0 | 2026-08-29 01:36:51.499210+00:00 | succeeded | eadd19360be6ecdb82c0d25f4d3447d162d2edd2acd789ba5e873fafaac8fc3b | 19,602 | null | """Four-code N-conditioned circulant operator over an X-initialized tape.
Public delimiters only route opaque categories into independent right-aligned
N, X, and T tapes. N makes a straight-through hard choice among four globally
learned cyclic operators; X alone initializes recurrent state; T is consulted
only after... | {
"id": "3fceee74-de4c-409b-8c2c-f756dfc3b5c6",
"created_at": "2026-08-29 01:36:51.499210+00:00",
"db_md5": "be2bf3989baac5e04978d28d9e6dbbb4",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "e97bdd97-9d7b-4ec6-ad08-83b9390a5906",
"tier": "easy",
"dataset_id": "e9",
"status": "succeeded",... | {
"score": {
"mean_loss": 4.855110168457031,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.035555556416511536
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.566369533538818,
"example_count": 90,
... |
3feaee77-aea3-43bf-82f8-f5c65b2b3772 | easy | jamaliki | 2026-08-04 13:55:55.260472+00:00 | succeeded | d8db75444a5785044750fa48536e2455f9729ff8ed791927ed4a98e44c198c1d | 17,717 | null | """One Layer Deeper submission: clocked latent loop with FSQ restoration.
Design: an encoder reads the input once into a memory. A weight-tied core is
applied m times per macro-step over [register | scratch] slots, cross-attending
to the memory (input re-injection). At each clock edge, registers are snapped
through fi... | {
"id": "3feaee77-aea3-43bf-82f8-f5c65b2b3772",
"created_at": "2026-08-04 13:55:55.260472+00:00",
"db_md5": "5a92f8a99746cd0f623a0887056a0727",
"submitter": "Kiarash Jamali",
"github_login": "jamaliki",
"run_id": "f2c386e4-81d1-4de8-b75e-d000c80f3521",
"tier": "easy",
"dataset_id": "e2",
"status": "su... | {
"score": {
"mean_loss": 7.962217807769775,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.016041667200624943
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.192340850830078,
"example_count": 300,... |
3fefd8a1-6aa2-40d6-a8ee-1fbeaa126c43 | easy | erdavis0 | 2026-08-22 02:02:07.572322+00:00 | succeeded | dd717b6474fda04ebdb5466cb2037b63cbf06831fdcb8eb294b90d300288916b | 5,797 | null | """Soft field factorization with multiplicative field interactions."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
)
WIDTH = 64
FIE... | {
"id": "3fefd8a1-6aa2-40d6-a8ee-1fbeaa126c43",
"created_at": "2026-08-22 02:02:07.572322+00:00",
"db_md5": "464075811335119d26ad5014b80665e5",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "79387ecb-bbfe-4a8c-a0e5-8c594e999fe1",
"tier": "easy",
"dataset_id": "e6",
"status": "succeeded",... | {
"score": {
"mean_loss": 8.275322198867798,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.09684684965759516
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.334005832672119,
"example_count": 60,
... |
3ff04db3-f32b-48f7-94e9-6a05251b0aad | easy | jordanrubin | 2026-08-05 20:56:07.687302+00:00 | succeeded | 4485eb9408916322a5c8c79fd4df1a42ab134ee701659a7f41a838860d08d480 | 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": "3ff04db3-f32b-48f7-94e9-6a05251b0aad",
"created_at": "2026-08-05 20:56:07.687302+00:00",
"db_md5": "238875dca4e317ad4cf2f299c07a8042",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "e80d6271-6189-498c-87f6-29bdcae9d2ba",
"tier": "easy",
"dataset_id": "e2",
"status": "s... | {
"score": {
"mean_loss": 3.8831090748798136,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.1504166660985599
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.619231597887099,
"example_count": 300,
... |
3ff6b296-b47e-44e6-978c-02c927c311c3 | easy | khushidahi | 2026-08-11 13:37:21.703975+00:00 | succeeded | 584807bb38cd916504beee6524bec3ea52dcfa61cfda5a5298a274b55f23ed51 | 24,959 | null | """R51 learned bilinear diagonal-routing model.
The architecture uses a learned low-rank pair feature for decimal-place slots
i and j, then a parameter-free scatter-add to slot i+j. The pair features,
carry propagation, modulus conditioning, recurrent update, and output decoder
are all learned end-to-end.
No digit mu... | {
"id": "3ff6b296-b47e-44e6-978c-02c927c311c3",
"created_at": "2026-08-11 13:37:21.703975+00:00",
"db_md5": "14a040042cf1457cae512f517978fc77",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "88a86e25-e95a-4b91-ad10-abca5b23cadc",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.1610899839604585,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007916666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1621624311524124,
"example_count": 60... |
3ffa0497-40da-4941-b94d-6a7263d45424 | easy | pculaf | 2026-08-15 18:46:06.155634+00:00 | succeeded | 25be811902e2fab36a384c75f72023ff55e7816a34bb750cb486eea1e0a58df4 | 10,276 | null | """Muon-optimized recurrent Transformer for the E6 activation/depth sweep."""
from __future__ import annotations
from collections.abc import Iterable
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submissio... | {
"id": "3ffa0497-40da-4941-b94d-6a7263d45424",
"created_at": "2026-08-15 18:46:06.155634+00:00",
"db_md5": "22e669afae2bd367b9f4a946624983a6",
"submitter": "Pavle Culafic",
"github_login": "pculaf",
"run_id": "c36ba6de-7ad8-4126-875e-db5244e72300",
"tier": "easy",
"dataset_id": "e6",
"status": "succe... | {
"score": {
"mean_loss": 2.0470550060272217,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.025225226767361164
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.076007127761841,
"example_count": 60,... |
3ffbab26-d9bd-4e55-a5ac-c00d8f17cb9c | easy | int3 | 2026-08-18 10:24:45.495680+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": "3ffbab26-d9bd-4e55-a5ac-c00d8f17cb9c",
"created_at": "2026-08-18 10:24:45.495680+00:00",
"db_md5": "eaac69fe345f82973fcf392c504a40ff",
"submitter": "Jez Ng",
"github_login": "int3",
"run_id": "b11e9f27-050d-4b11-99b0-fada59e0f1b7",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
... | {
"score": {
"mean_loss": 3.0456100702285767,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0283333333209157
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.061513900756836,
"example_count": 100,
... |
3ffcf495-7d63-4fa3-8746-8638b2740984 | easy | oupadhyay | 2026-08-23 23:18:58.121936+00:00 | succeeded | 83a0ff4fca1353949b9d0f2dc10cf21b919540c4be3eb25e7fad0c9c257f27a0 | 7,414 | null | """BiScan entropy preference with hard-forward straight-through feedback."""
import torch
import torch.nn.functional as F
from torch import nn
from benchmark import OptimizerBundle, Submission, assert_model_state
PAD, N, X, T, ANS, DIGIT = 0, 2, 3, 4, 5, 7
MAX_STEPS = 64
class Config:
def __init__(self, spec):... | {
"id": "3ffcf495-7d63-4fa3-8746-8638b2740984",
"created_at": "2026-08-23 23:18:58.121936+00:00",
"db_md5": "24e793cc87a026b65b2405bcf1a6c014",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "7321730d-a1a1-4622-84e3-5afff559de71",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.168167421514423,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004583333345750968
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.159226609452423,
"example_count": 600,... |
4001a23b-a167-452e-9eeb-2a7c2a4fae29 | easy | jordanrubin | 2026-08-02 20:20:12.754753+00:00 | succeeded | 582ab6e152cfd42d52aab09c3b9ffeb9337357a604516045d534185dcedcf29f | 13,994 | 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": "4001a23b-a167-452e-9eeb-2a7c2a4fae29",
"created_at": "2026-08-02 20:20:12.754753+00:00",
"db_md5": "2793c6be04b6965a9e22d80c93c62b12",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "80d33d99-e927-41f6-bef5-00adf56005a4",
"tier": "easy",
"dataset_id": "e1",
"status": "s... | {
"score": {
"mean_loss": 5.376524159837924,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.2149999945734938
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.363896764926057,
"example_count": 100,
... |
400b462f-1b50-49a9-bfae-b8bab57e76fd | easy | DDanlov | 2026-08-14 07:43:19.496498+00:00 | succeeded | 52c8f780bd80643100bd4df9fb6975672fa53422c49f0c797c4d45f15cba1465 | 21,975 | null | import math
from typing import Optional
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Optimizer
try:
from benchmark import Submission, assert_model_state, OptimizerBundle
except ImportError:
try:
from client import Submission, assert_model_state, OptimizerBu... | {
"id": "400b462f-1b50-49a9-bfae-b8bab57e76fd",
"created_at": "2026-08-14 07:43:19.496498+00:00",
"db_md5": "02f5b2add60f865a1a867fb13b8d4762",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "ffeffc3d-fa3c-4fcd-a4d5-c69397214f68",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.8328714291173647,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.018125000000000002
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8328732059039035,
"example_count": 80... |
400f2e8c-7b92-45de-ba9d-4b8f9e8f5d0d | easy | AFeinfield | 2026-08-13 00:10:34.191256+00:00 | succeeded | 8ff9e474523f546822130a2fad538e6d806c77e28eb231d473e5d6aa274e899f | 7,214 | null | """Looped (weight-tied) bidirectional Transformer, d=256, 8 iterations.
014: 013 at d=128 h=4 — capacity probe: smaller nets grok sooner relative
to memorization (813K -> ~210K params for 600 train rows)
to amortize the evaluator's ~200ms/iteration overhead into 8x more steps.
prior. Same parameter count order as a 1-... | {
"id": "400f2e8c-7b92-45de-ba9d-4b8f9e8f5d0d",
"created_at": "2026-08-13 00:10:34.191256+00:00",
"db_md5": "4c6c2b1f642451dceb6097bf0b2f8806",
"submitter": "AFeinfield",
"github_login": "AFeinfield",
"run_id": "e8100e64-6479-4dd1-a1a4-c92e7f781667",
"tier": "easy",
"dataset_id": "e1",
"status": "succ... | {
"score": {
"mean_loss": 6.6893861293792725,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03999999910593033
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.763472557067871,
"example_count": 100,... |
4012fcd2-bca9-423e-b6af-c67cc2461b08 | easy | newjordan | 2026-08-15 22:25:59.405891+00:00 | succeeded | 6da6e937508b55c0269fa90d648775dd5beecc4c8d7f4934fa089f933655e8a8 | 178,654 | 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": "4012fcd2-bca9-423e-b6af-c67cc2461b08",
"created_at": "2026-08-15 22:25:59.405891+00:00",
"db_md5": "e6c32df21410d4316bc488cf172a8410",
"submitter": "Frosty40",
"github_login": "newjordan",
"run_id": "b1601dde-5261-4676-9278-b3e7d1c61477",
"tier": "easy",
"dataset_id": "e1",
"status": "succeed... | {
"score": {
"mean_loss": 0.0027389402966946363,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.0027631628327071667,
"example_count": 100,
... |
401eed6e-50e3-4116-ad35-0bf093b3bebe | easy | DDanlov | 2026-08-17 08:08:12.010529+00:00 | succeeded | 8f005bc98f513478159344be6c150ffa9051acb60ea2208e391c9ce5aa807742 | 30,801 | null | """
Scalable Recurrent Deliberation Transformer (ScalableDEPHNet)
============================================================
Architecture:
- Pure No-Norm: Single unit RMS normalization strictly isolated to the input token embedding layer.
Zero LayerNorm / RMSNorm inside recurrent deliberation blocks.
- Pure No-Skip... | {
"id": "401eed6e-50e3-4116-ad35-0bf093b3bebe",
"created_at": "2026-08-17 08:08:12.010529+00:00",
"db_md5": "72c8d20e10400f0cbdfc033476b63f79",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "78adc31a-45bf-4ae1-9816-5afa3d7fe5c0",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 1.9629075322067946,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04999999701976776
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.013435125350952,
"example_count": 100,... |
40216198-2d4d-4d12-a5d3-04e8b5b9e58b | easy | alirezashirvani-jr | 2026-08-31 00:24:33.508489+00:00 | succeeded | d9be1270c66f4cec678728ecb8d3888d74765a4fd36126c152a8eb335ac95f65 | 19,728 | 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": "40216198-2d4d-4d12-a5d3-04e8b5b9e58b",
"created_at": "2026-08-31 00:24:33.508489+00:00",
"db_md5": "d28e9bd3af3854ff3baff14eaf466501",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "c0b74e80-5b1c-4478-8587-bd79cc24bdc4",
"tier": "easy",
"dataset_id": "e1",
... | {
"score": {
"mean_loss": 0.4183793905249331,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.9800000190734863
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.00046818313421681523,
"example_count": ... |
4022962b-9560-4eaf-94e5-1cc135274a69 | easy | benjaminW2025 | 2026-08-18 00:00:35.923171+00:00 | succeeded | 9d45820eefecaabe3bc89931e8ea74bf642ad125a1257433f06b526a64ecce89 | 8,017 | 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": "4022962b-9560-4eaf-94e5-1cc135274a69",
"created_at": "2026-08-18 00:00:35.923171+00:00",
"db_md5": "61e690b585987ff8ac6d7afd35a5236c",
"submitter": "benjawesome",
"github_login": "benjaminW2025",
"run_id": "20661591-e611-442e-bd47-9d082320c7e8",
"tier": "easy",
"dataset_id": "e5",
"status": "... | {
"score": {
"mean_loss": 5.116793468954544,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004166666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.260302503012755,
"example_count": 600,... |
4022f52e-040a-4ec4-a4ee-e56c655c3b4e | easy | AdityaVG13 | 2026-08-15 09:06:34.921451+00:00 | succeeded | 7104ddedddaf467cd5539e516758b7c2ae17d62874b65bf59467f41199b7339f | 112,108 | null | """AEGIS V10 v10_donor_bank: first-hop JOIN with a T=1 label memory.
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 trivalent Saf... | {
"id": "4022f52e-040a-4ec4-a4ee-e56c655c3b4e",
"created_at": "2026-08-15 09:06:34.921451+00:00",
"db_md5": "767835d8e5f928dda039d7bdf7dde2ef",
"submitter": "AdityaG",
"github_login": "AdityaVG13",
"run_id": "e3f52f8c-63cd-4306-af2d-fb228df69552",
"tier": "easy",
"dataset_id": "e1",
"status": "succeed... | {
"score": {
"mean_loss": 2.485571324825287,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.6499999761581421
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.897415041923523,
"example_count": 100,
... |
4026181d-3496-4891-982e-4f234ecef0fc | easy | erdavis0 | 2026-08-19 09:06:40.320414+00:00 | succeeded | a7d564f83acafc66aa99aba4fb2d17f46de1cc46969aed4ae9a60533a4da1c66 | 5,994 | null | """Small generic reverse-position Transformer tuned for optimizer throughput."""
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,
ass... | {
"id": "4026181d-3496-4891-982e-4f234ecef0fc",
"created_at": "2026-08-19 09:06:40.320414+00:00",
"db_md5": "4355ae47e0b9320823b2636a4866cece",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "591e02a6-ba7c-451b-8ff8-1b400ea758a6",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded",... | {
"score": {
"mean_loss": 10.799447811421086,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0075000000605359675
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 11.012380049917745,
"example_count": 8... |
40270e04-d2e0-45d3-8cc2-346713f051b4 | easy | sankalp1999 | 2026-08-04 17:10:46.361142+00:00 | succeeded | 3b79fc21de9e0100fcd186c5a038abe5e87d8989a431eb06875bdc25e9f9d319 | 34,659 | null | """p2_program: identify f by hypothesis filtering, with a neural fallback.
The diagnosis this is built on
------------------------------
p1_generic memorises 14.4k examples of x^2 mod N to train loss 0.000 and scores
0% held out, at every width from 12.6M to 402M, at every LR that does not
diverge, on 16k / 50k / 150k... | {
"id": "40270e04-d2e0-45d3-8cc2-346713f051b4",
"created_at": "2026-08-04 17:10:46.361142+00:00",
"db_md5": "1d834f16da97d60fe96aa31d6bab0bff",
"submitter": "Sankalp ",
"github_login": "sankalp1999",
"run_id": "5aeb52d2-cf5b-4cfc-8a41-b25a5316386f",
"tier": "easy",
"dataset_id": "e2",
"status": "succe... | {
"score": {
"mean_loss": 0.0,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.0,
"example_count": 300,
"exact_accuracy": 1.0,
"... |
402a3352-1a90-4978-aacd-ed6cc358c9f7 | easy | c-nnr | 2026-08-04 13:11:53.998581+00:00 | succeeded | a60de7264162881d88f44e5267d657fafcf6a04c82b872bd160c66c3f1bbf8b9 | 6,459 | null | """One Layer Deeper E1 final-15 experiment: Primary candidate: calibrated embeddings plus the E1-winning two-train/eight-eval recurrence."""
from __future__ import annotations
import math
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBun... | {
"id": "402a3352-1a90-4978-aacd-ed6cc358c9f7",
"created_at": "2026-08-04 13:11:53.998581+00:00",
"db_md5": "ef2e279c8d5d287717601c5beb3445f9",
"submitter": "c-nnr",
"github_login": "c-nnr",
"run_id": "9893e0ba-cf1f-4871-aead-2cd4d39e061c",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
... | {
"score": {
"mean_loss": 2.186632990837097,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0833333320915699
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2873404026031494,
"example_count": 100,
... |
402e7ad0-db59-4453-a28c-365531078118 | easy | karanganesan | 2026-08-08 03:38:53.299048+00:00 | succeeded | 8d9eecdc8607c91b190336e24b49597738decc69ce7d4e24ee5930ae97cfcad7 | 27,127 | null | """Parametric looped-transformer family (P1).
One weight-tied transformer block applied k times in latent space. Config
flags cover four P1 families with one file:
looped recall=0 gated=0 tfilm=0 plain weight-tied loop
looped-recall recall=1 re-inject the input embedding each
... | {
"id": "402e7ad0-db59-4453-a28c-365531078118",
"created_at": "2026-08-08 03:38:53.299048+00:00",
"db_md5": "3e8439ba8ab561733e7dc883fab4ecb2",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "f1d36ec7-0815-496d-9b09-42ef743b3bf0",
"tier": "easy",
"dataset_id": "e4",
"status": ... | {
"score": {
"mean_loss": 6.822913408279419,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005833333590999246
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.0636725425720215,
"example_count": 120... |
4030624f-1fc0-41c7-bc00-8b53c36682fe | easy | viridale | 2026-08-07 01:14:23.981159+00:00 | succeeded | 44f022e69387b7dabff39319c86ad6d45835e12e09249a84a587f8d6d417f274 | 8,430 | null | """Supervised causal block exchange on the clean untied serial model.
hypothesis: forcing the same final target through one fixed counterfactual causal
ordering (blocks 3 and 8 exchanged) will discourage brittle depth-slot shortcuts
while preserving the expressive canonical route and its exact inference cost.
... | {
"id": "4030624f-1fc0-41c7-bc00-8b53c36682fe",
"created_at": "2026-08-07 01:14:23.981159+00:00",
"db_md5": "1c62c39b91daff9cd550f8045a89236d",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "467f0e7b-13c7-4303-ae53-379187bd6c85",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 1.850646734237671,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.02333333296701312
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.798162817955017,
"example_count": 100,
... |
4032cdae-1548-4a3e-b92d-e028125ac167 | easy | DDanlov | 2026-08-11 19:52:44.927725+00:00 | succeeded | 2db3eb9bd3ea947ff547d18aedb1670bdcb20fa602bf666cbbd894d9fbbaf36c | 20,019 | null | from __future__ import annotations
import sys
import os
import math
import time
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim.optimizer import Optimizer
VOCAB_SIZE = 17
DIGIT_OFFSET = 7
def init_cutting_the_skip(module):
"""SVD-based orthogonal initialization to ensure initia... | {
"id": "4032cdae-1548-4a3e-b92d-e028125ac167",
"created_at": "2026-08-11 19:52:44.927725+00:00",
"db_md5": "c344f31cff4a37a173c59a3d0429743e",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "4e9c4454-452a-49d1-bc91-ff35430d3666",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 4.446509974853059,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0033333333333333335
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.557558536529541,
"example_count": 100... |
40401f1b-044c-4bdd-918d-de0fc5e04e07 | easy | sapient-sapiens | 2026-08-14 17:46:37.752216+00:00 | succeeded | 28ba469aae3e42ca679270105f219b6bd47433fa80f83dc154b3c3e55e44c7c3 | 14,299 | null | """Digit-space recurrent operator for repeated modular squaring.
Nothing here converts a token into a decimal number. Digits stay as embeddings,
so what a digit is worth, and every carry and reduction, is learned. That also
sidesteps a hard numeric wall: at Hard's modulus scale `x^2` runs past the 53-bit
exact-integer... | {
"id": "40401f1b-044c-4bdd-918d-de0fc5e04e07",
"created_at": "2026-08-14 17:46:37.752216+00:00",
"db_md5": "1efadda9e01ae84ec2857432a666c542",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "47b7edda-4ca9-4ae5-9080-e2a598027649",
"tier": "easy",
"dataset_id": "e3",
"status": "su... | {
"score": {
"mean_loss": 2.1266487759048625,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006875
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.123420696876407,
"example_count": 800,
... |
4043643b-cd48-487f-880d-df6540eacb59 | easy | liam-gb | 2026-08-17 02:09:13.457074+00:00 | succeeded | 24fc878b19987564e88eccf1d98937ae5735fa8ab620fdef470b9eacfec14e2f | 16,740 | null | """rns_pm_gather_ct_widet_l32nore_r8: single change vs
rns_pm_gather_ct_widet_l32nore — the batch-reuse policy returns True, so the
evaluator reuses each fetched batch for the maximum eight optimizer updates.
WHY. On the granular Easy suites one batch is 85-97% of the training split,
so every step ends an epoch and th... | {
"id": "4043643b-cd48-487f-880d-df6540eacb59",
"created_at": "2026-08-17 02:09:13.457074+00:00",
"db_md5": "46c5decfff7d97a850fb297981e9d310",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "196c2509-1c0d-4548-888c-fb7e15258dee",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 4.178189367055893,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.47999999672174454
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.495662689208984,
"example_count": 100,
... |
4043a90e-785b-4bcb-82e7-8340849d5fbd | easy | ArkinDharawat | 2026-08-13 17:23:04.929588+00:00 | succeeded | 109bd036da4ad3adec1bc59b71b70e041dc4508b72b34af2defb0306a5c91285 | 7,632 | null | """Universal-Transformer-style looped block, v2: step embed + pos offset + LR schedule + 6 loops.
Extends ``looped_ut_padded`` with four ideas drawn from
Graves 2016 (ACT), Dehghani et al. 2018 (Universal Transformer), and
Merrill & Sabharwal 2025 (log-depth transformers):
1. **Learned per-step (timestep) embedding**... | {
"id": "4043a90e-785b-4bcb-82e7-8340849d5fbd",
"created_at": "2026-08-13 17:23:04.929588+00:00",
"db_md5": "16175e2a640d8206a119d1aaea9d8df4",
"submitter": "Arkin Dharawat",
"github_login": "ArkinDharawat",
"run_id": "2a1a13d8-7c9c-42fc-807c-28671923a0b5",
"tier": "easy",
"dataset_id": "e3",
"status"... | {
"score": {
"mean_loss": 2.2334494430251626,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005000000009313226
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2966796223436226,
"example_count": 80... |
4047d966-46e3-4d6c-8c07-abc16244ca73 | easy | viridale | 2026-08-05 01:19:32.622702+00:00 | succeeded | 076f1fffeb0a424ad55e25863bb3f0bfa6a65ffbc0b1c2f6570a2d186ef8f12a | 8,777 | null | """Clean fixed unit-horizon neural residual flow.
hypothesis: order-one recurrent jumps produce unstable discrete trajectories; a shared
learned vector field integrated with 64 small fixed steps may select a smooth,
depth-stable procedure and establish the prerequisite for coarse/fine methods.
axis: arch
targe... | {
"id": "4047d966-46e3-4d6c-8c07-abc16244ca73",
"created_at": "2026-08-05 01:19:32.622702+00:00",
"db_md5": "8a623a3ff79f8bf4218ee389a0effad5",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "f7a78f0f-5b6f-4781-9b10-ecfcbb3b12c4",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 1.9005419611930847,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.071666669100523
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.8648622035980225,
"example_count": 100,
... |
404b0306-27ce-4dbf-ab58-347b4f56686b | easy | yashkant | 2026-08-31 17:01:47.690463+00:00 | succeeded | 8aeadecab70131d8bd53b2dcce1a57b210bfc16ce50b36b983c2b4b75e15f801 | 104,475 | null | """One Layer Deeper submission -- a learned register machine.
Architecture (all outputs produced by learned parameters; no task arithmetic on the answering path):
a bank of R=6 digit registers (W=16 positions each) carried as DISCRETE tokens between micro-steps: five
data registers and one COUNTDOWN register holding t... | {
"id": "404b0306-27ce-4dbf-ab58-347b4f56686b",
"created_at": "2026-08-31 17:01:47.690463+00:00",
"db_md5": "4e9c65f4af368bd3681bb8695b99d2fb",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "82c7c6bf-03ef-406b-9e12-c7500fe2f72d",
"tier": "easy",
"dataset_id": "e4",
"status": "succeed... | {
"score": {
"mean_loss": 5.343671333183857,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0012500000046566128
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.232293544744578,
"example_count": 120... |
404c640a-7090-4357-a8eb-9c449beab5da | easy | DDanlov | 2026-08-13 00:00:44.697219+00:00 | failed | 0e073fc3bf751cfba453f859d52a1c3a7ce6b4b8c203dd0e41959920ea689ad2 | 26,540 | null | from __future__ import annotations
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim.optimizer import Optimizer
try:
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
)
except ... | {
"id": "404c640a-7090-4357-a8eb-9c449beab5da",
"created_at": "2026-08-13 00:00:44.697219+00:00",
"db_md5": "af401a0594a1953b6de234df2d54eb30",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "fcf27608-e607-460f-9cd3-785a483357f1",
"tier": "easy",
"dataset_id": "e5",
"status": "failed",
... | null
|
404da815-bc41-4303-a2fb-5fbf87c49ffd | easy | gauravmishra | 2026-08-08 00:06:03.625363+00:00 | succeeded | d74892c24149a43ba0cdffb61184aad8bf53fa05abdc4d223f4d559049fc2bc8 | 3,657 | 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": "404da815-bc41-4303-a2fb-5fbf87c49ffd",
"created_at": "2026-08-08 00:06:03.625363+00:00",
"db_md5": "85d8ec03e0b3a6a9bd08ce2324f23cdd",
"submitter": "Gaurav Mishra",
"github_login": "gauravmishra",
"run_id": "5986421e-06ad-4a32-b38a-d4b42a636548",
"tier": "easy",
"dataset_id": "e5",
"status": ... | {
"score": {
"mean_loss": 2.224404471216678,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007500000012417634
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2590261831411866,
"example_count": 600... |
404f8878-4f3e-4eef-b07e-1c7db68aef3a | easy | Mantissagithub | 2026-08-16 20:20:37.399710+00:00 | succeeded | 9cbd00a0e29750c32e0aa8e88d7a42eaa16a009f340bc1bf8899f3ad2924eb8d | 7,159 | null | """One Layer Deeper v1: cheap gated tied recurrence.
Goal:
- keep recurrence, but make it cheap enough to get ~baseline-like optimizer steps;
- use one shared block repeatedly;
- keep train/eval recurrence identical for this diagnostic run;
- use plain mean token cross-entropy;
- do not hard-code the task or parse N/X... | {
"id": "404f8878-4f3e-4eef-b07e-1c7db68aef3a",
"created_at": "2026-08-16 20:20:37.399710+00:00",
"db_md5": "250fdefd5a920a4194493508996cf8e1",
"submitter": "Pradheep P",
"github_login": "Mantissagithub",
"run_id": "b7249017-4a3a-4eab-a716-b0e2a11fc73a",
"tier": "easy",
"dataset_id": "e6",
"status": "... | {
"score": {
"mean_loss": 3.21683669090271,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.011711712228134274
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.7182655334472656,
"example_count": 60,
... |
4050b43a-3e37-43ec-9d7f-00def7fa0c2b | easy | erdavis0 | 2026-08-26 04:33:39.891246+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": "4050b43a-3e37-43ec-9d7f-00def7fa0c2b",
"created_at": "2026-08-26 04:33:39.891246+00:00",
"db_md5": "b3917270504f5239445a48889c48e9ae",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "643b0db8-24c5-4e66-a222-dc95b626fbd5",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",... | {
"score": {
"mean_loss": 4.725288152694702,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04999999888241291
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.829916000366211,
"example_count": 100,
... |
4052541b-4570-47b5-a376-b766eaf7dd5a | easy | sapient-sapiens | 2026-08-17 19:22:34.506826+00:00 | succeeded | 323bbdb07f2673f09810445a100f6702764663d3c64ee822098f5a942f386314 | 28,658 | null | """v40 lineage: T-free operator, N memory, PonderNet mix, no T in the loss.
Field-local encoder, T-free core, countable halt, answer-slot readout.
N is seeded and restored every step. T is never parsed in forward or loss.
"""
from __future__ import annotations
import math
import time
import torch
import torch.nn.fu... | {
"id": "4052541b-4570-47b5-a376-b766eaf7dd5a",
"created_at": "2026-08-17 19:22:34.506826+00:00",
"db_md5": "7e033ec41d673d046e533619ac937247",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "ca806e5f-9d0a-4a75-bbd1-2493d6cd2868",
"tier": "easy",
"dataset_id": "e9",
"status": "su... | {
"score": {
"mean_loss": 2.1500787617877988,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03333333430190881
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2693467140197754,
"example_count": 90,... |
40547574-7535-48d7-b744-712c340136ba | easy | isaac0804 | 2026-08-04 21:59:27.212216+00:00 | succeeded | e150bfc799bbdeb702d126aed06816d9953c9ce4136dea1a005442383096f757 | 8,352 | null | """Weight-tied looped Transformer that reads T from the prompt to set its own depth.
The task is x_t = x_{t-1}^2 mod N for T steps. A single fixed-depth block has no
mechanism to represent "apply this operation T times" and cannot extrapolate to
depths unseen in training. This submission instead applies one shared blo... | {
"id": "40547574-7535-48d7-b744-712c340136ba",
"created_at": "2026-08-04 21:59:27.212216+00:00",
"db_md5": "fbe215c54c5741274101702dab493b79",
"submitter": "Isaac Yong",
"github_login": "isaac0804",
"run_id": "fc992ced-0903-46e7-a4b2-a05f2e159822",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 2.29105017240063,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.012500000037252903
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.3757468270914464,
"example_count": 800,... |
4054c61f-75cb-4a1d-b3c1-de8faef50cbc | easy | newjordan | 2026-08-09 16:18:09.810130+00:00 | succeeded | 6e528b3da26ea152ee3064782ccd20d9c74915b11e7c5c05c32ffbf220a67510 | 93,405 | 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": "4054c61f-75cb-4a1d-b3c1-de8faef50cbc",
"created_at": "2026-08-09 16:18:09.810130+00:00",
"db_md5": "7ec58ed4a8ede86a7bd2b82a6feec527",
"submitter": "Frosty40",
"github_login": "newjordan",
"run_id": "4853f151-ae49-475b-8f49-f5403264c97e",
"tier": "easy",
"dataset_id": "e3",
"status": "succeed... | {
"score": {
"mean_loss": 0.000855825155080131,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.0008549899256897508,
"example_count": 800,
... |
405de6c6-8b4c-4836-bf78-8f2c54881813 | easy | anirudh-chakravarthy | 2026-08-12 03:52:40.635573+00:00 | succeeded | 32789a0c6a3ca21508f82d947caf34c5a76f121b3cd124a7cbc74fe738b9e12f | 14,824 | 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": "405de6c6-8b4c-4836-bf78-8f2c54881813",
"created_at": "2026-08-12 03:52:40.635573+00:00",
"db_md5": "32558e50be5ace4023c59636bfcd5b13",
"submitter": "Anirudh S Chakravarthy",
"github_login": "anirudh-chakravarthy",
"run_id": "3b9228d6-71b0-41e2-b185-1bf9f155f64f",
"tier": "easy",
"dataset_id": "... | {
"score": {
"mean_loss": 4.482673148681676,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.011250000037252902
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.56286589874815,
"example_count": 600,
... |
40640c5d-dfd1-40e0-9ef0-0e81da712962 | easy | isaac0804 | 2026-08-06 10:28:17.957956+00:00 | succeeded | 394873827c0c1d9e9d31397dd2b11c19c39673036ab4bbbcd3d931e4ba2df75f | 6,622 | null | """Basic single-pass Transformer + Sharpness-Aware Minimization (SAM).
Targets the memorize-vs-generalize gap directly (flat minima generalize
better) rather than just training longer. Uses the evaluator's
backward_passes_per_step=2 + between_backward_passes hook: pass 1 computes a
gradient at the current weights... | {
"id": "40640c5d-dfd1-40e0-9ef0-0e81da712962",
"created_at": "2026-08-06 10:28:17.957956+00:00",
"db_md5": "1dec0b7232a38ef54a45661582ef4c5f",
"submitter": "Isaac Yong",
"github_login": "isaac0804",
"run_id": "3a430126-c03a-4d6a-9a67-fd5dd54a250b",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 3.0153275757832545,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.010000000009313226
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.7477978933845493,
"example_count": 80... |
407c98e2-e0f8-43d4-bd74-b59dbcfa792a | easy | isaac0804 | 2026-08-05 08:55:36.922694+00:00 | succeeded | f836b272089910a4ba0c4460e0669db3ddc6c9b9b00a877b35732b896ca1e908 | 5,186 | null | """Vanilla (non-recurrent, non-weight-tied) deep Transformer -- a control to
isolate "is a single attention+MLP pass too shallow to even use T" from the
looped architecture's own known failure modes.
`baseline_adamw` (this project's original Step-0 baseline) has exactly one
block: one attention+MLP pass has to read N,... | {
"id": "407c98e2-e0f8-43d4-bd74-b59dbcfa792a",
"created_at": "2026-08-05 08:55:36.922694+00:00",
"db_md5": "6a4a79008638fac84c6972fac862e86e",
"submitter": "Isaac Yong",
"github_login": "isaac0804",
"run_id": "99349636-9b8e-40a5-a18d-f003d23b947b",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 11.82491805332291,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009375000013969839
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 12.33074582781569,
"example_count": 800,... |
407dd115-9c09-4260-baa4-eae481d37450 | easy | jordanrubin | 2026-08-05 19:45:54.179541+00:00 | succeeded | adf82ad877e82a9ada879ce8744f8307e37c891a65c0ab06dc7b7a1e033952e0 | 11,503 | null | """Continuous-state 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 times.
T controls o... | {
"id": "407dd115-9c09-4260-baa4-eae481d37450",
"created_at": "2026-08-05 19:45:54.179541+00:00",
"db_md5": "808b2428831ae3bb60d193e7ca2faca9",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "5c56d119-a2a4-4530-8fdc-26daa88f0536",
"tier": "easy",
"dataset_id": "e2",
"status": "s... | {
"score": {
"mean_loss": 4.853577453429247,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.19020833012958369
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.199588593982515,
"example_count": 300,
... |
407eb7f4-737c-41a3-829c-015ad7654acf | easy | KaustubhKumar05 | 2026-08-26 10:56:11.398583+00:00 | succeeded | ad1a59dbe438196247711f8c8d79ef308d13ceac1d0b58a8502799bb93863774 | 14,154 | null | """Neural-GPU-style conv GRU, v2s: v2 + speedrun pack.
- LR: warmup, hold max to 85% of wall-clock, steep late cliff (2026 schedule theory)
- identity-at-init: cell output projection zero-initialized
- batch reuse x6 (multi-pass theory)
Base (v2):
Changes over v1 (each backed by measured results in the literature):
... | {
"id": "407eb7f4-737c-41a3-829c-015ad7654acf",
"created_at": "2026-08-26 10:56:11.398583+00:00",
"db_md5": "3af77fcc511e8bb1212c68881b9f02a2",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "1b65b5b1-0905-4f7f-ac2b-3fa0c68fa1b9",
"tier": "easy",
"dataset_id": "e6",
"status": "succee... | {
"score": {
"mean_loss": 6.513102293014526,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.23265765979886055
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.380997657775879,
"example_count": 60,
... |
4086cf52-0174-4f7a-b73d-0dbc1043e81f | easy | datathesisai | 2026-08-05 07:17:14.792599+00:00 | succeeded | 615f72ceb38e5e9c6192496f92032f58f2f031d6d3aac7bb17a493a1bd0793c3 | 16,131 | null | """Two-timescale latent recurrence for task-agnostic function composition.
The model separates a slowly changing solution stream ``y`` from a fast
scratch stream ``z``. Both are updated by one weight-tied Transformer cell,
and every update recalls an immutable encoding of the input. There are no
task token IDs, arit... | {
"id": "4086cf52-0174-4f7a-b73d-0dbc1043e81f",
"created_at": "2026-08-05 07:17:14.792599+00:00",
"db_md5": "7eeba9fee9f0f012feb119f7baa9b105",
"submitter": "datathesisai",
"github_login": "datathesisai",
"run_id": "8f6a70e7-49a5-4813-a5a7-1347787027f2",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 17.57243251800537,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.02499999850988388
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 17.73082160949707,
"example_count": 100,
... |
4096eb17-5853-474b-8d6e-cb55a809dd72 | easy | karanganesan | 2026-08-05 11:31:17.808657+00:00 | succeeded | 9096df86ebb64088930c218447a455300be53fc71e669c9034efb240a7712777 | 22,098 | 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": "4096eb17-5853-474b-8d6e-cb55a809dd72",
"created_at": "2026-08-05 11:31:17.808657+00:00",
"db_md5": "2211588332c5dfa70e9b5a238392da80",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "187a559d-f55b-4f43-8991-c64436e0e3ba",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 13.437085151672363,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03666666615754366
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 14.550686836242676,
"example_count": 100... |
409a64c3-1981-4b1b-8192-df9c98ba6ad0 | easy | lpbb | 2026-08-19 11:24:49.237066+00:00 | succeeded | 1e902181180e520aa996b60741f97c8030e27902d1eebdb54347e6cd169005d9 | 8,863 | null | """S0 + RoPE + abacus embeddings + dropout, with a looped block: instead of
`num_layers` independent layers, a block of `num_layers` unique layers is run
`num_repeats` times with shared weights (e.g. 12 layers x 4 repeats = depth 48
but only 12 layers' worth of parameters). Weight tying across repeats forces
the block ... | {
"id": "409a64c3-1981-4b1b-8192-df9c98ba6ad0",
"created_at": "2026-08-19 11:24:49.237066+00:00",
"db_md5": "11edb1f5db9b92d32d4f6fd72539068e",
"submitter": "Lpbb",
"github_login": "lpbb",
"run_id": "14208ad4-203a-4c77-8334-27eb33d5e904",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
"s... | {
"score": {
"mean_loss": 3.304517149925232,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04333333298563957
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.9017813205718994,
"example_count": 100,... |
409f8f15-e0e5-4550-9f0b-f798c7b562ff | easy | karanganesan | 2026-08-07 07:42:26.336924+00:00 | succeeded | 6c11c42a55769e5f047679d82d1acdbca340ebb80eafb09bee4b4f4028e91c94 | 25,600 | null | """Parametric looped-transformer family (P1).
One weight-tied transformer block applied k times in latent space. Config
flags cover four P1 families with one file:
looped recall=0 gated=0 tfilm=0 plain weight-tied loop
looped-recall recall=1 re-inject the input embedding each
... | {
"id": "409f8f15-e0e5-4550-9f0b-f798c7b562ff",
"created_at": "2026-08-07 07:42:26.336924+00:00",
"db_md5": "090fe34270a38663990271bb5dc0b588",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "f173f20c-c1d4-425a-b984-058acc7335d9",
"tier": "easy",
"dataset_id": "e5",
"status": ... | {
"score": {
"mean_loss": 7.05932354927063,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007083333563059568
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.291353702545166,
"example_count": 600,
... |
40adaaea-996e-4144-a871-84e8b1f44dae | easy | chad-atexpedient | 2026-08-05 04:22:47.386231+00:00 | succeeded | e35dcae433ad1f0880c2334fef4f25b85bee0f196c12a464b727f85018798240 | 10,880 | null | """R028F: semantic processor with forced behavior and memory choices."""
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
CH... | {
"id": "40adaaea-996e-4144-a871-84e8b1f44dae",
"created_at": "2026-08-05 04:22:47.386231+00:00",
"db_md5": "9a840b08639778459d83b1f0a432df21",
"submitter": "chad-atexpedient",
"github_login": "chad-atexpedient",
"run_id": "449ac1d5-3e64-4f7c-ba70-31d19be081d9",
"tier": "easy",
"dataset_id": "e3",
"st... | {
"score": {
"mean_loss": 2.2707220179329557,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.018125000046566128
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.270254759914894,
"example_count": 800... |
40addd7c-e03e-4728-9213-6e216d2c8822 | easy | velocizapkar | 2026-08-06 22:08:14.493522+00:00 | succeeded | 0dec51a1d3e38108ddfb55fe78233dc55574ed1d3c6047dbb253285de11c2e0b | 54,197 | null | """Fast packed residual-free Attention-Residual transformer with Muon.
The model is a generic learned sequence model. It contains no parser,
arithmetic primitive, task constant, generated data, label feedback, or answer
construction. Every macro-state is produced by a serial pack of attention and
MLP transformations... | {
"id": "40addd7c-e03e-4728-9213-6e216d2c8822",
"created_at": "2026-08-06 22:08:14.493522+00:00",
"db_md5": "4ab7dd7d0bc4b93c1a8626055730f902",
"submitter": "Aakanksh Zarapkar",
"github_login": "velocizapkar",
"run_id": "513fce79-fe12-45f3-9e31-8c9546f7c367",
"tier": "easy",
"dataset_id": "e1",
"statu... | {
"score": {
"mean_loss": 3.428761601448059,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.053333332762122154
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.6739261150360107,
"example_count": 100... |
40af239c-7579-4705-96f0-e3b1126f1333 | easy | sapient-sapiens | 2026-08-18 10:24:44.802205+00:00 | succeeded | fdf9f2447929e46b6d5026a970b07a90cd784070ad7a411ed7981888d1e05c66 | 24,074 | null | """v43: v40 lineage with the halting controller removed and a fixed unroll.
Diagnosis this arm responds to. Across every stored local run the learned
controller emitted a single constant depth for the whole dataset (1032 of 1037
depth-vs-T cells had exactly zero per-example variance), and forced-depth
sweeps showed lo... | {
"id": "40af239c-7579-4705-96f0-e3b1126f1333",
"created_at": "2026-08-18 10:24:44.802205+00:00",
"db_md5": "0dae543e6b461fc07eb19dbda081f3b7",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "24c716c4-f353-4c63-b84a-f77d15787e40",
"tier": "easy",
"dataset_id": "e9",
"status": "su... | {
"score": {
"mean_loss": 4.82354871572115,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03444444465140502
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.365936279296875,
"example_count": 90,
... |
40b49fd0-0aad-4be3-8079-3dec33917ebb | easy | rishivg | 2026-08-04 00:09:51.168604+00:00 | failed | 9dbcde50a7b885cf7828958a7670c4cc924664b094275bfc0d49a8a9b5c4a023 | 101,642 | null | """Looped transformer with place-value coordinates, Muon, and a clock-paced LR.
Design notes, in rough order of expected value:
1. Weight-tied recurrence with input injection. The task is function
composition of depth ``T``, so one block applied ``LOOPS`` times is a much
better prior than a single feed-forw... | {
"id": "40b49fd0-0aad-4be3-8079-3dec33917ebb",
"created_at": "2026-08-04 00:09:51.168604+00:00",
"db_md5": "74f69085bcee89fce61e7a03052b87eb",
"submitter": "Rishi Gottumukkala",
"github_login": "rishivg",
"run_id": "b31fcd70-e853-4eea-8606-f8ea34268d7c",
"tier": "easy",
"dataset_id": "e1",
"status": ... | null
|
40b6f852-0169-4d9d-82d2-7fd75720f832 | easy | khushidahi | 2026-08-23 20:43:27.447721+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": "40b6f852-0169-4d9d-82d2-7fd75720f832",
"created_at": "2026-08-23 20:43:27.447721+00:00",
"db_md5": "25f8e724d7ec7f02d7c56a4ef3401964",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "d79bc2c2-5223-4de7-af85-0360a2321377",
"tier": "easy",
"dataset_id": "e9",
"status": "fail... | null
|
40bdd115-0ea7-4fa1-bce9-f12af6755f3f | easy | akkolli | 2026-08-05 23:42:29.181188+00:00 | succeeded | 8fb537235d2dd5eae37c1edc4ff3fd3135ccf7bc1bf8dadc0ce5476c65338064 | 59,192 | null | """One learned arithmetic, relative geometry, and affine-content mixture.
The decimal step count drives three independent recurrent trajectories with
shared learned parameters. One applies a two-factor Neural Multiplication
Unit directly, one applies that arithmetic before a learned length-relative
digit transport, a... | {
"id": "40bdd115-0ea7-4fa1-bce9-f12af6755f3f",
"created_at": "2026-08-05 23:42:29.181188+00:00",
"db_md5": "e64f4060f0cb95de2616ae38f90526d3",
"submitter": "Akshay Kolli",
"github_login": "akkolli",
"run_id": "3f7a0567-9f40-43de-b64d-688d24d2e1c8",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 0.10875302780816162,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.1087530255317688,
"example_count": 800,
"ex... |
40c57dc4-df74-48bc-b81e-93ce688f0616 | easy | c-nnr | 2026-08-04 13:19:42.995980+00:00 | succeeded | a917d850399f14f6ab453f56f31b057ec08d15f953d69231e8935e6806a2edd1 | 6,441 | null | """One Layer Deeper E1 final-15 experiment: Calibrated embeddings plus identity-like zero-initialised recurrent updates."""
from __future__ import annotations
import math
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
Optimize... | {
"id": "40c57dc4-df74-48bc-b81e-93ce688f0616",
"created_at": "2026-08-04 13:19:42.995980+00:00",
"db_md5": "6e489ad08f9d6ef1a93a6d2c7e0b2fea",
"submitter": "c-nnr",
"github_login": "c-nnr",
"run_id": "f3b01068-2b29-482a-9a54-d1ee4a58f924",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
... | {
"score": {
"mean_loss": 7.884720802307129,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0566666666418314
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.493170738220215,
"example_count": 100,
... |
40c8f071-dc22-4687-b569-e2fe2dba2fb5 | easy | khushidahi | 2026-08-04 00:25:14.154322+00:00 | succeeded | eadcf6f32447beb3dbd124b69b63e310b497d6bbeea52bfa751ecb0a06ce6344 | 14,961 | 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": "40c8f071-dc22-4687-b569-e2fe2dba2fb5",
"created_at": "2026-08-04 00:25:14.154322+00:00",
"db_md5": "cfabcd6d22eb0194d0fe0ebb74498a87",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "3e2fbd06-934c-4bd1-9fb2-444696db8e7a",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.2199243307113647,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008333333302289248
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2164783477783203,
"example_count": 60... |
40c9548e-8bc1-418e-bc66-b14107abd4ad | easy | jordanrubin | 2026-08-07 17:02:24.781139+00:00 | succeeded | 19e434c7e9fc9d9d601292a0f5c0a4083813bc7790caf87c0582fc1bbcc2cad8 | 16,909 | null | """Autonomous digit-state recurrent Transformer for repeated modular squaring.
The model is deliberately organized around one learned transition:
p_0 = right_aligned_decimal_digits(x)
p_{k+1} = F_theta(p_k, decimal_digits(N))
The same two-block Transformer cell is applied exactly T times. T is used only
as ... | {
"id": "40c9548e-8bc1-418e-bc66-b14107abd4ad",
"created_at": "2026-08-07 17:02:24.781139+00:00",
"db_md5": "671c15c2db39c821999bae96a7defa51",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "5418ec63-c2c0-4df9-acbc-ef136621e994",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.822429433013057,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0058333332029481735
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8490090737695652,
"example_count": 60... |
40d17e44-ea63-4fa7-9b87-b2e0c9ddea5e | easy | oupadhyay | 2026-08-07 10:49:38.321653+00:00 | succeeded | a695e083f021d0d82cea542d70277ea57029593f60e85da61d778bf51f6ac679 | 8,351 | null | """Complete compressed-axial forward path with the known-safe baseline optimizer."""
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, 32, 13162
... | {
"id": "40d17e44-ea63-4fa7-9b87-b2e0c9ddea5e",
"created_at": "2026-08-07 10:49:38.321653+00:00",
"db_md5": "050479d67b4f8a80d3618128c89efe6c",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "bb04b802-be0b-41b0-ac82-c8325329eab9",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.8340627822384015,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0020833333333333333
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.841014304904125,
"example_count": 60... |
40d60c02-321a-4fb7-a8b9-287a03f458b7 | easy | DDanlov | 2026-08-19 21:13:22.094223+00:00 | failed | d7142e45e24e54bd75fc181ee8b566aad1d3ac369dfaa24baaf30dce86a4a6d4 | 17,727 | null | """
Dynamic Submission for One Layer Deeper API Benchmark
Config: dim=640, num_heads=32, d_ff_mult=4, fixed_trec=2000
"""
from __future__ import annotations
import math
from typing import Any, Optional, Tuple
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Optimizer
try:
... | {
"id": "40d60c02-321a-4fb7-a8b9-287a03f458b7",
"created_at": "2026-08-19 21:13:22.094223+00:00",
"db_md5": "a1666f91f8472872f6fb0f09455fef88",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "32005a1d-a010-4dad-8938-139dbd8e5911",
"tier": "easy",
"dataset_id": "e5",
"status": "failed",
... | null
|
40d712a1-50cf-4fb7-b3f5-4cffe340a740 | easy | msnow0 | 2026-08-08 11:22:48.122337+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": "40d712a1-50cf-4fb7-b3f5-4cffe340a740",
"created_at": "2026-08-08 11:22:48.122337+00:00",
"db_md5": "b9182f4cd6fb33d3bf6b28cdaca6d3e9",
"submitter": "msnow0",
"github_login": "msnow0",
"run_id": "9516ba4b-44bd-421e-9275-64528989612b",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
... | {
"score": {
"mean_loss": 4.759247303009033,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03499999828636646
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.777778625488281,
"example_count": 100,
... |
40df4db8-d8cb-4a05-a62b-d3ac37488a40 | easy | Dandandan | 2026-08-07 03:19:21.405881+00:00 | failed | 663f7a3030189fa6ea323b6c79088ca2b45a31b017463cb4d8d35f17ba437eb4 | 99,005 | null | """Generalized exact recursive learner for One Layer Deeper.
The learned transition is a population of ordinary modular neural units with
several generic arithmetic activations. A single learned evidence selector
chooses one transition, which is then reused for every input-requested outer
step. The population covers ... | {
"id": "40df4db8-d8cb-4a05-a62b-d3ac37488a40",
"created_at": "2026-08-07 03:19:21.405881+00:00",
"db_md5": "c497bd48c4cc50e352b982a9782c6b41",
"submitter": "Daniël Heres",
"github_login": "Dandandan",
"run_id": "64cf4a69-8411-4423-8a98-f88cf5425e33",
"tier": "easy",
"dataset_id": "e1",
"status": "fai... | null
|
40e025ac-2d06-4049-8c3d-85c0db8a5d60 | easy | shan23chen | 2026-08-10 18:37:44.105901+00:00 | succeeded | 42d0dc002b41e5f3552fcbcd945546e6b13b63417ad63dc28223ef5020669fb7 | 39,363 | null | """Dual-root boundary-reset recurrent register Transformer.
One learned transition is applied once per requested outer step. The
transition receives an immutable decimal N register and the current learned
boundary state, but never receives T or a separate copy of the original X.
Within each outer application, scratch... | {
"id": "40e025ac-2d06-4049-8c3d-85c0db8a5d60",
"created_at": "2026-08-10 18:37:44.105901+00:00",
"db_md5": "d2ae3589b082618d4cb2e44c3ef40793",
"submitter": "Shan Chen",
"github_login": "shan23chen",
"run_id": "02b62b2b-788a-4e23-89b3-95f5ab663c1b",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 2.259348750114441,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.013124999590218067
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.251192808151245,
"example_count": 800,... |
40e4388d-2413-479a-ae0a-f41af778bc85 | easy | pculaf | 2026-08-15 19:16:43.605920+00:00 | succeeded | d30a152df0cc060b42368a10b0823f75fe35022cb5e48682c54514c57f7e7540 | 10,278 | null | """Muon-optimized recurrent Transformer for the E6 activation/depth sweep."""
from __future__ import annotations
from collections.abc import Iterable
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submissio... | {
"id": "40e4388d-2413-479a-ae0a-f41af778bc85",
"created_at": "2026-08-15 19:16:43.605920+00:00",
"db_md5": "b1584e8b1f7d3636952ec47e8d1acbb7",
"submitter": "Pavle Culafic",
"github_login": "pculaf",
"run_id": "c08f31a6-bb05-4f7d-8499-a7c394d754a8",
"tier": "easy",
"dataset_id": "e6",
"status": "succe... | {
"score": {
"mean_loss": 1.884991705417633,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03513513784855604
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.8552610874176025,
"example_count": 60,
... |
40e8288d-01fb-4504-8842-768a69b1cbcf | easy | DDanlov | 2026-08-09 00:18:31.149065+00:00 | failed | 00e4c444b4b925c37c62fc3a0e532173bfc92a48c01aee80b32eca5f96e1d834 | 15,981 | null | from __future__ import annotations
import sys
import os
import math
import time
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim.optimizer import Optimizer
VOCAB_SIZE = 17
DIGIT_OFFSET = 7
class RohanShampoo(Optimizer):
"""Self-contained RohanShampoo optimizer with eigenvalue ma... | {
"id": "40e8288d-01fb-4504-8842-768a69b1cbcf",
"created_at": "2026-08-09 00:18:31.149065+00:00",
"db_md5": "dcdbe623973f92497df0c0573b76113d",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "201bd776-e850-4ac7-8e82-11a0f6506261",
"tier": "easy",
"dataset_id": "e5",
"status": "failed",
... | null
|
40e93a26-750b-4c61-b35e-b483bb409ca2 | easy | erdavis0 | 2026-08-26 04:17:28.898027+00:00 | succeeded | 34958bd1fecf39e0cd177cec37f4a5b763f56ecda181924fc21973bf094acd2c | 30,362 | null | """Latent-role continuous REV C3: sequence mixture plus role balance.
Public delimiter syntax only routes opaque token categories into distinct,
right-aligned N, X, and T tapes. Learned local cells compile separate
interaction, carry, and modulus-conditioned workspaces. N is isolated from
interaction and carry and f... | {
"id": "40e93a26-750b-4c61-b35e-b483bb409ca2",
"created_at": "2026-08-26 04:17:28.898027+00:00",
"db_md5": "63a88d8095c75c1574375cc20d89bc96",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "00ba1dab-dacd-41ab-813a-f59423f34350",
"tier": "easy",
"dataset_id": "e7",
"status": "succeeded",... | {
"score": {
"mean_loss": 5.032892465591431,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.06938502937555313
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.763271331787109,
"example_count": 85,
... |
40ed5c5a-8346-4dc6-9cce-6557e4687203 | easy | benjaminW2025 | 2026-08-17 18:41:41.967611+00:00 | succeeded | df56a7437288d72797152153deb15e0082d6f720ba47c14b242bd3f3177e3290 | 5,813 | 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": "40ed5c5a-8346-4dc6-9cce-6557e4687203",
"created_at": "2026-08-17 18:41:41.967611+00:00",
"db_md5": "0ac10b86f24fb92ef86dce0e8746dbb8",
"submitter": "benjawesome",
"github_login": "benjaminW2025",
"run_id": "0fe22357-41d2-4c6a-8878-c8d8cb3f70be",
"tier": "easy",
"dataset_id": "e5",
"status": "... | {
"score": {
"mean_loss": 2.3414850459663112,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009583333358168602
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.4535747996360198,
"example_count": 60... |
40f1dd1e-e201-4f77-baf4-13e81f9f0897 | easy | oupadhyay | 2026-08-28 01:52:19.923369+00:00 | succeeded | a3a2947ad380d57dd49ed05c925eebe763522eb978e455796777c8a107db8f51 | 4,046 | null | """Literal polynomial cell."""
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
D,H=64,32
class C:
def __init__(s,vocab_size,max_seq_len):s.vocab_size,s.max_seq_len=vocab_size,max_seq_... | {
"id": "40f1dd1e-e201-4f77-baf4-13e81f9f0897",
"created_at": "2026-08-28 01:52:19.923369+00:00",
"db_md5": "7909d3f6791981e4fc068cd6b72d30b4",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "6ee63f78-9a5e-4720-b46c-58cfba2cb462",
"tier": "easy",
"dataset_id": "e7",
"status": "s... | {
"score": {
"mean_loss": 3.2532339096069336,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.02691622171550989
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.399191379547119,
"example_count": 85,
... |
40f6d605-2d41-44ac-b053-22f6aad37fbe | easy | khushidahi | 2026-08-17 00:07:46.907038+00:00 | succeeded | 21e17ccdd4d0b184480f995d2d935c2d9ea760295cad424d366b488737fbfb10 | 27,862 | null | """R178: per-example soft-nonlinearity Neural GPU control.
This rules-safe control keeps the length-equivariant diagonal computation and
global communication mechanism of R162. Each
learned state-digit pair becomes a token that attends over every learned
modulus digit before diagonal pooling. That creates a learned ... | {
"id": "40f6d605-2d41-44ac-b053-22f6aad37fbe",
"created_at": "2026-08-17 00:07:46.907038+00:00",
"db_md5": "4c62f1f306ebe1ef0fd4222f5a55c121",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "a957281d-3f70-4193-98e0-609eaaa647bb",
"tier": "easy",
"dataset_id": "e10",
"status": "suc... | {
"score": {
"mean_loss": 1.5772649049758911,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.591050535440445
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.4297534227371216,
"example_count": 125,
... |
40f738fb-d512-491e-baa3-6c3e602e73ab | easy | oupadhyay | 2026-08-15 22:36:05.490186+00:00 | succeeded | f5d1dc6be186bcf8db4aa55c79bcf5f5c154fd5c1bb1d97b4d38c1a0e85e249b | 4,205 | null | """Conditional structured-loss screen: smooth10."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import nn
from benchmark import ModelSpec,OptimizerSpec,OptimizerBundle,Submission,TokenLossBatch,assert_model_state
W,D,H=4,64,32
class C:
def __init__(s,vocab_size,ma... | {
"id": "40f738fb-d512-491e-baa3-6c3e602e73ab",
"created_at": "2026-08-15 22:36:05.490186+00:00",
"db_md5": "122a1d8206169fb36efde7a560d1e0a2",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "5b4e132e-9ac0-4c28-840b-a097037aa28c",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 120.55294207722251,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0004166666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 118.49327282413773,
"example_count": 6... |
40f95784-e908-4a25-9679-dc96a1d62902 | easy | Cree0618 | 2026-08-30 14:34:21.267437+00:00 | succeeded | fa6ced7934852ef22327372d9bdab8b727e6c933bc930e9d66f889b92169da9b | 14,151 | null | """E040: factorized numeric state with a learned multiplicative transition.
This is an original state-machine model rather than a recurrent prompt
Transformer. The input is split into N and x digit fields, each field is
encoded by a learned digit-by-place product, and x initializes a dedicated
value state. A static ... | {
"id": "40f95784-e908-4a25-9679-dc96a1d62902",
"created_at": "2026-08-30 14:34:21.267437+00:00",
"db_md5": "d15814c5272c43aac64449849a6068a6",
"submitter": "Cree0618",
"github_login": "Cree0618",
"run_id": "1df2a046-60cb-4986-9a8d-24562296b471",
"tier": "easy",
"dataset_id": "e1",
"status": "succeede... | {
"score": {
"mean_loss": 5.208487033843994,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0949999988079071
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.458451271057129,
"example_count": 100,
... |
40fa52ff-b84a-4f4d-80e2-86ef60569e44 | easy | CodeReclaimers | 2026-08-16 18:17:55.492789+00:00 | succeeded | d51cff743155edc84152557d8a53b1adc9e9ce409de337eaea8a9f20a6f55c07 | 14,989 | null | """Direct/recurrent dual path with a learned decimal codec.
The direct path may use the complete prompt. The recurrent path encodes N and
X once, then applies one N-conditioned tied transition per literal time step;
T never enters that transition. Both paths and their bounded learned mixture
are trained end to end f... | {
"id": "40fa52ff-b84a-4f4d-80e2-86ef60569e44",
"created_at": "2026-08-16 18:17:55.492789+00:00",
"db_md5": "8f82293898024d2910a97df5c37daf65",
"submitter": "CodeReclaimers",
"github_login": "CodeReclaimers",
"run_id": "05727702-a6a9-4930-95a0-8d72f4f0a49b",
"tier": "easy",
"dataset_id": "e5",
"status... | {
"score": {
"mean_loss": 2.1587193411799124,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.010833333345750967
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.168390593721193,
"example_count": 600... |
40ffc20d-3913-49c7-ae83-5184fd36b90c | easy | yashkant | 2026-08-14 16:39:48.608102+00:00 | succeeded | 872294eb85bb0c332cca4b2a79e5424d5f3fabbb7eae2ae274bce94a4b61b34e | 47,625 | null | """Round1802 nonlinear carry-transfer prefix scan pair.
The exact Round1792 AdamW5 carrier is retained. A separate answer state and
latent bank of learned nonlinear carry-transfer curves are added at twelve
right-aligned places. Each parsed-T macro runs a four-stage parallel prefix
scan. Its one shared binary opera... | {
"id": "40ffc20d-3913-49c7-ae83-5184fd36b90c",
"created_at": "2026-08-14 16:39:48.608102+00:00",
"db_md5": "605df94b49f9c266ff636b85f4ecece1",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "870ba1cc-951c-4fde-8b66-7970607d25a4",
"tier": "easy",
"dataset_id": "e3",
"status": "succeed... | {
"score": {
"mean_loss": 2.143725327445151,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.010000000018626451
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1465217489427912,
"example_count": 800... |
4101f25e-7438-4383-b176-d6b5372ea509 | easy | pjjajal | 2026-08-07 02:56:37.466530+00:00 | failed | 3db7d10cfee123aa575b97d0f52ef3f1aff5e4c868e01b256603457dad41e7fe | 8,613 | null | """Minimal EqR-style looped Transformer submission.
The embedded input is preserved as ``z0`` and injected into every iteration of
one shared Transformer block. Each iteration is a damped update toward the
block's proposal. Noise, truncation, trajectories, and prelude/coda stacks are
intentionally left out for now.
... | {
"id": "4101f25e-7438-4383-b176-d6b5372ea509",
"created_at": "2026-08-07 02:56:37.466530+00:00",
"db_md5": "7f153e448d03e7d422aea039adf9f9d8",
"submitter": "pjjajal",
"github_login": "pjjajal",
"run_id": "ab683a54-36d9-464d-8f9b-82320f5d5c66",
"tier": "easy",
"dataset_id": "e1",
"status": "failed",
... | null
|
410a6f86-5c18-4e07-a504-b0419b0f7a85 | easy | DDanlov | 2026-08-14 07:53:31.627912+00:00 | succeeded | 55afd208114f497b3dae798f63cbcbc1b80f14e567e3c65870739058beb7de91 | 21,975 | null | import math
from typing import Optional
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Optimizer
try:
from benchmark import Submission, assert_model_state, OptimizerBundle
except ImportError:
try:
from client import Submission, assert_model_state, OptimizerBu... | {
"id": "410a6f86-5c18-4e07-a504-b0419b0f7a85",
"created_at": "2026-08-14 07:53:31.627912+00:00",
"db_md5": "aea8cb5beb7ee6e0de1160092775603d",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "ce9872c1-8954-432f-999a-a69dd6ed4621",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.311524169168238,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.018125000000000002
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.312114728816076,
"example_count": 800,... |
411a849c-4f4a-476b-ae73-812e5524492a | easy | G-AshwinKumar | 2026-08-13 09:37:43.581252+00:00 | succeeded | 1fc4ebc90f9cfb1af1522b253820f51c1d6975d15810a8b7c933c2a06b250b1c | 3,934 | null | """Recurrent Weight-Tied Transformer (TRM style) with K=6 iteration step embeddings."""
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": "411a849c-4f4a-476b-ae73-812e5524492a",
"created_at": "2026-08-13 09:37:43.581252+00:00",
"db_md5": "dca881ecc60a7629abdf9df94f595b2a",
"submitter": "Ashwin Kumar",
"github_login": "G-AshwinKumar",
"run_id": "6a2ba10a-7569-4375-9d44-fb580272a4f3",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 2.201019525527954,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04833333380520344
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.3384015560150146,
"example_count": 100,... |
4126ace7-1551-4589-b387-6e8c0b1c6f60 | easy | DDanlov | 2026-08-11 19:30:30.693601+00:00 | failed | 467e10fb604f246963dc78fbf8e0847654410ccd4448e2f8b8fefb7398ec9456 | 18,961 | null | from __future__ import annotations
import sys
import os
import math
import time
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim.optimizer import Optimizer
VOCAB_SIZE = 17
DIGIT_OFFSET = 7
class RohanShampoo(Optimizer):
"""Self-contained RohanShampoo optimizer with bias-correcte... | {
"id": "4126ace7-1551-4589-b387-6e8c0b1c6f60",
"created_at": "2026-08-11 19:30:30.693601+00:00",
"db_md5": "9665e06f48f13811eb2ef5b581d7181d",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "95ffc995-2b39-4afe-92c2-3cb4033a24ee",
"tier": "easy",
"dataset_id": "e1",
"status": "failed",
... | null
|
412c3ef8-9a97-4041-8d62-75fb6a1ecea8 | easy | viridale | 2026-08-06 06:08:59.886513+00:00 | succeeded | 8cade153e40f470ab4cafc43b068b5982d04977c03d683a40f427b15ad919893 | 16,239 | null | """Clean recurrent triplet edge-attention graph machine.
hypothesis: persistent learned state on every ordered token pair, updated through
differentiable two-edge path composition, preserves relational constraints that
token-vector recurrence compresses into endpoint memorization.
axis: arch
target: one frozen... | {
"id": "412c3ef8-9a97-4041-8d62-75fb6a1ecea8",
"created_at": "2026-08-06 06:08:59.886513+00:00",
"db_md5": "6ad128504919a34e32e3c4415c4239a7",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "5d3d9632-3f3c-4f58-8e5f-c5537f4be860",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 4.223746167735792,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.358554375705435,
"example_count": 100,
"exac... |
4136ac82-d1dd-4804-a1f9-c6b6d9dc50c9 | easy | himalalps | 2026-08-06 08:36:28.999516+00:00 | succeeded | adfdf099f5b6e1eac7d6b58cca1d8d1549edce469b112c6d090e9a5344157af2 | 27,520 | null | """Algorithmically constrained latent recurrence for One Layer Deeper.
One masked cross-attention pass creates two independent representations:
``state_0`` from X only and an immutable context from N only. T is excluded
from both representations and is used solely to choose how many times the
learned recurrent MLP tr... | {
"id": "4136ac82-d1dd-4804-a1f9-c6b6d9dc50c9",
"created_at": "2026-08-06 08:36:28.999516+00:00",
"db_md5": "a6ed2896cc4a4bbebdf23dc99001261e",
"submitter": "Haoyu Tang",
"github_login": "himalalps",
"run_id": "72a47ef1-d470-4c82-9a74-4176cf3473b5",
"tier": "easy",
"dataset_id": "e2",
"status": "succe... | {
"score": {
"mean_loss": 8.029197454452515,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.011250000447034836
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.460752487182617,
"example_count": 300,... |
413c6587-9145-4dea-8912-352e6fa1f18f | easy | sapient-sapiens | 2026-08-17 02:33:51.296194+00:00 | succeeded | 96d137d90e52e00d4bfe7084329085f05efee0981345708cdf647be0670dae02 | 15,120 | null | """Fixed-4 recurrence candidate: v70 source without adaptive halt.
Parent: act64_e6_w1024_dh4_L6_wd1_v70 (width 1024, d_h=4, L=6, WD=1.0).
Primary variable: adaptive ACT over 1--64 (typically hard-halt 2) versus a
fixed four applications of the shared core. The halt controller is removed.
T remains an ordinary prompt ... | {
"id": "413c6587-9145-4dea-8912-352e6fa1f18f",
"created_at": "2026-08-17 02:33:51.296194+00:00",
"db_md5": "a360270b1bd62e5c960d2cc497021af4",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "190fd1f0-4151-49c6-b840-1a986190ce3c",
"tier": "easy",
"dataset_id": "e6",
"status": "su... | {
"score": {
"mean_loss": 5.569543070989113,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.10698198451585061
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.9880523681640625,
"example_count": 60,
... |
413e9558-046d-46be-86fe-52d90a4ad177 | easy | DDanlov | 2026-08-20 11:29:15.496343+00:00 | succeeded | 85266637fabde726c006c33100c9819f3748c1be8a3c97d1e66cc42e3dd3b0c0 | 17,806 | null | from __future__ import annotations
import math
import time
from typing import Any, Optional, Tuple
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Optimizer
try:
from benchmark.api import ModelSpec, OptimizerBundle, OptimizerSpec, Submission, assert_model_state
except Imp... | {
"id": "413e9558-046d-46be-86fe-52d90a4ad177",
"created_at": "2026-08-20 11:29:15.496343+00:00",
"db_md5": "ed830511ef56db04c0ee3a88c5ea5ce5",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "429cce17-6b74-4519-80cf-f5325918b355",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.2880981126080417,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.011250000012417634
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2854679881724542,
"example_count": 60... |
414ac4a5-b5ed-458b-ad62-7fe458747656 | easy | shirvani-jr | 2026-09-01 00:50:35.238715+00:00 | succeeded | b317b19da85760bde75e4d27c457d49e66ac8dcca62bdf058148eb0af50f5c6b | 24,687 | 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
ANSWER_TOKEN = 5
END_TOKEN = 6
DIGIT_OFFSET = 7
FACTOR_COUNT = 12
OUTER... | {
"id": "414ac4a5-b5ed-458b-ad62-7fe458747656",
"created_at": "2026-09-01 00:50:35.238715+00:00",
"db_md5": "9af79d78d1f2438fc32ea2e1db49817a",
"submitter": "Ali",
"github_login": "shirvani-jr",
"run_id": "6e28fb22-29b3-4d7b-8be8-07b5f53d6a7c",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded"... | {
"score": {
"mean_loss": 0.6763964626153561,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.9875000011920929
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.6847489485732637,
"example_count": 800,... |
414ac722-32cd-4bdf-9d42-01cfc0276c4c | easy | erdavis0 | 2026-08-25 02:45:03.598843+00:00 | succeeded | c9f21dbfa5e3f75ce670918d20378f38fca944ca903cc99fda41be09be46d0ae | 26,762 | null | """Reversible augmented-state transducer over aligned learned digit tapes.
Public delimiter syntax only routes opaque token categories into distinct,
right-aligned N, X, and T tapes. Learned local cells compile separate
interaction, carry, and modulus-conditioned workspaces. N is isolated from
interaction and carry ... | {
"id": "414ac722-32cd-4bdf-9d42-01cfc0276c4c",
"created_at": "2026-08-25 02:45:03.598843+00:00",
"db_md5": "73ab5c56d5aea3ccc98fb3f3b89f5955",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "f288248d-f1a5-4694-a90a-7e0b31e1d427",
"tier": "easy",
"dataset_id": "e6",
"status": "succeeded",... | {
"score": {
"mean_loss": 3.013643264770508,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.11554054543375969
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.0643718242645264,
"example_count": 60,
... |
41528715-0744-4dee-9da7-29dc9f200139 | easy | jordanrubin | 2026-08-02 17:36:36.330742+00:00 | succeeded | a98432391baf36fe81642db6d6a41e7ca5af50c210bb231561ca1ca02efb76d5 | 13,993 | 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": "41528715-0744-4dee-9da7-29dc9f200139",
"created_at": "2026-08-02 17:36:36.330742+00:00",
"db_md5": "08327994152c913babb987198cda7f52",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "af338f63-f584-4644-8827-7fe0c4701d08",
"tier": "easy",
"dataset_id": "e1",
"status": "s... | {
"score": {
"mean_loss": 3.444551392268801,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.028333332700033985
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.56556925666866,
"example_count": 100,
... |
416bf934-60ef-4cff-bf7c-c4aa7423bdfc | easy | AFeinfield | 2026-08-29 03:44:44.885019+00:00 | succeeded | 5e1d9653ad95d148176927e070ff7b2ef68ff6a768b2d05919ced269f5d71830 | 16,611 | null | """205: 204 + TABLE DETACH in the training forward. Local isolation
shows drills snap all 7 tables in 10-20 steps; in context the end-CE
gradient noise through hundreds of ST ops inflates Adam moments and
stalls them. Detaching tables in the forward (training only) leaves the
clean drill gradient as their sole teacher;... | {
"id": "416bf934-60ef-4cff-bf7c-c4aa7423bdfc",
"created_at": "2026-08-29 03:44:44.885019+00:00",
"db_md5": "c0ee5124a192087b1b0b10c9b12fe6fe",
"submitter": "AFeinfield",
"github_login": "AFeinfield",
"run_id": "41ff64f7-9fd2-4b0c-ada9-96b8566e641b",
"tier": "easy",
"dataset_id": "e1",
"status": "succ... | {
"score": {
"mean_loss": 2.5456602807391873,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.1366666675110658
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.5671536415370544,
"example_count": 100,... |
416c5190-eb25-40ad-b6c6-8a39b264e189 | easy | viridale | 2026-08-04 21:55:37.970767+00:00 | succeeded | 8c6c6c21b7bf165a9dad30c23b21fe99e3951a28cb45d084a648fadd45045c1c | 7,880 | null | """Clean autonomous recurrence with correlation-aware tied logits.
hypothesis: the tied head's random self-token logit scales as d because the residual
stream contains its input embedding, so v168's 1/sqrt(d) scale still starts at CE
14.7; using 1/d should expose ordinary categorical loss from the first update... | {
"id": "416c5190-eb25-40ad-b6c6-8a39b264e189",
"created_at": "2026-08-04 21:55:37.970767+00:00",
"db_md5": "662ab5d3ac28c203ffc1d076df5c4215",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "b02146da-77f4-4933-8472-2d8c8f4a8f03",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 2.2066688453456385,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.01
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2048228737883675,
"example_count": 800,
"ex... |
4170a7db-54dc-4eb4-83c9-e66d0869aa64 | easy | khushidahi | 2026-08-01 21:32:38.917114+00:00 | succeeded | a42af26df258dd2c7bde64a1f58f4e676087f69c51ce0a8951a8bc459ca6b50d | 4,045 | 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": "4170a7db-54dc-4eb4-83c9-e66d0869aa64",
"created_at": "2026-08-01 21:32:38.917114+00:00",
"db_md5": "c976f95ad4d2c0440f827cac10085dcf",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "a3e98a64-4bfc-41ca-9b80-ce0085836fb0",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.2169210714369436,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007916666741172473
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2204720867054344,
"example_count": 60... |
417191cd-371a-4570-af15-8605dfcee105 | easy | sapient-sapiens | 2026-08-15 14:55:59.671784+00:00 | succeeded | eb8d707f64c3cc6f052c571897c76e97727ac6041bfc52004ec308b5bc5e91d3 | 9,043 | null | """Fixed-T=2 recurrent Transformer with semantic field-relative embeddings.
The learned four-block operator is applied exactly twice with shared weights.
Inputs remain categorical: token, prompt-field identity, and left-to-right
position within the field. No token is converted to a number and no arithmetic
operation ... | {
"id": "417191cd-371a-4570-af15-8605dfcee105",
"created_at": "2026-08-15 14:55:59.671784+00:00",
"db_md5": "2d5b5a1fa21c3242e0a5d64802fce16a",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "1e191e48-1e99-45aa-a7ff-26bd304bcb98",
"tier": "easy",
"dataset_id": "e3",
"status": "su... | {
"score": {
"mean_loss": 2.130320610782653,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0037500000139698386
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.127272536185894,
"example_count": 800... |
417516d7-1e95-467a-a671-cc1f367cda94 | easy | jordanrubin | 2026-08-06 07:59:30.230322+00:00 | succeeded | db0ea04649a5b1e7f6ccee718712f34217561008baaf39d112ff703e707d976e | 14,772 | null | """Global-binding categorical looped Transformer for public Easy e1.
The recurrent boundary is exactly a 16-place by 10-symbol categorical state.
Each tied cell reconstructs a learned D=256 global residue fingerprint from
that entire state, carries it as the first token of a residual Transformer,
injects it into every... | {
"id": "417516d7-1e95-467a-a671-cc1f367cda94",
"created_at": "2026-08-06 07:59:30.230322+00:00",
"db_md5": "7fd0fcf87f1111de121441725dd81d50",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "886734cb-f65a-4d45-ade0-cf8d8ca88fe6",
"tier": "easy",
"dataset_id": "e1",
"status": "s... | {
"score": {
"mean_loss": 1.8756293058395386,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0833333320915699
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.8305025100708008,
"example_count": 100,... |
417650d7-5784-4685-8a93-d5e2f3f3cdbc | easy | LongView-dev | 2026-09-01 03:40:24.528571+00:00 | succeeded | ff6fcaa04596f3ba48cf2ef6a20a7ef117115b3e84b8840a7b07b2f9c1891956 | 5,085 | null | """A compact tied-depth Transformer for the One Layer Deeper benchmark.
The model is deliberately task-agnostic: it is a learned bidirectional sequence
model with a recurrently reused block, not an implementation of modular
arithmetic. A sequence-balanced loss matches the benchmark's exact-example
metric more closely... | {
"id": "417650d7-5784-4685-8a93-d5e2f3f3cdbc",
"created_at": "2026-09-01 03:40:24.528571+00:00",
"db_md5": "0828fbe032f3e50e63317fba943c61c9",
"submitter": "LongView-dev",
"github_login": "LongView-dev",
"run_id": "78cd5242-600a-49df-bee0-c3446206525d",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 2.7901798486709595,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03499999921768904
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.802616596221924,
"example_count": 100,... |
4180db26-ae98-44bc-9fb7-a6927d97773a | easy | lpbb | 2026-08-21 13:41:11.200779+00:00 | succeeded | ff92f3622cb6d6b17c7142912272bee132c8df59f5f038a625b6700caba1c6da | 12,545 | null | """S0 + RoPE + abacus + looped block, with *input-adaptive* depth: instead of
a fixed `num_repeats`, each sample's T is parsed from its own token stream
(the digits after the `T` marker) and the shared block of `num_layers` layers
is applied exactly T times for that sample. One block pass = one modular
squaring step, s... | {
"id": "4180db26-ae98-44bc-9fb7-a6927d97773a",
"created_at": "2026-08-21 13:41:11.200779+00:00",
"db_md5": "e8d3e81bc2440dca95ccbafadc8343f7",
"submitter": "Lpbb",
"github_login": "lpbb",
"run_id": "25e2de04-992b-4a34-ac7b-2cf55f8b3700",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
"s... | {
"score": {
"mean_loss": 3.6808348894119263,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03666666615754366
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.1236443519592285,
"example_count": 100... |
4184ab76-6d1c-4e2e-ad02-d8685c73fb3a | easy | DDanlov | 2026-08-15 20:52:47.516145+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": "4184ab76-6d1c-4e2e-ad02-d8685c73fb3a",
"created_at": "2026-08-15 20:52:47.516145+00:00",
"db_md5": "623970f9d216ffb45b2ef4f8507de941",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "b9ab05f1-defa-48bb-9f4d-9c0404a7baac",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.3072685881011,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.015000000136593977
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.241301476065792,
"example_count": 100,
... |
419969e1-a571-4abf-ba58-64fdb485527c | easy | yashkant | 2026-08-13 00:44:07.700291+00:00 | succeeded | 679bf5db4a2c079269ca10f50716832e3f7b103469960697bd1d80cd098cf4b3 | 48,492 | null | """Round1868 learned shuffle-exchange circuit on the SAM carrier.
The exact Round1829 SAM/Compositional-Muon Dykstra model stays intact. Eight
small rails repeatedly call one shared learned pair transition four times per
parsed-T step, with the immutable carrier anchor reinjected into every call.
The candidate cycles... | {
"id": "419969e1-a571-4abf-ba58-64fdb485527c",
"created_at": "2026-08-13 00:44:07.700291+00:00",
"db_md5": "cb8385924e345f95899182a53ec4f58a",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "98bcbf3a-8218-4d16-8bdb-6df7cea235d5",
"tier": "easy",
"dataset_id": "e3",
"status": "succeed... | {
"score": {
"mean_loss": 2.1319035176484267,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.012500000027939676
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.132585828826866,
"example_count": 800... |
419c6058-bfb3-40cb-8ebe-2ba5daf1c04e | easy | Alan-Qin | 2026-08-26 15:40:42.383559+00:00 | failed | 43292be318074b6ff920dd87415656262f451a54bd98eb5b1409527adc4abc10 | 26,078 | null | """v91h17s = v91h17 + file-hash init seeding (m6 full-ladder lottery tickets).
v91h17 = v91 with LR-cosine horizon 1700 (Medium tiers are 600s hosted ~1.7k steps, NOTES #172; cosine must land inside the real budget).
v91 = v63lf + w-hat auxiliary head on the proven gated MLP (route B v2:
same readout, added CRT-quoti... | {
"id": "419c6058-bfb3-40cb-8ebe-2ba5daf1c04e",
"created_at": "2026-08-26 15:40:42.383559+00:00",
"db_md5": "135361c5370969693b9c64cb366cb95d",
"submitter": "Zeyu Qin",
"github_login": "Alan-Qin",
"run_id": "a582b2ca-2ec1-4080-bade-8fd56be383fd",
"tier": "easy",
"dataset_id": "e6",
"status": "failed",... | null
|
41b28d4d-abeb-4cb0-bada-ba216f7bed74 | easy | Amr-Hegazy1 | 2026-08-04 05:35:02.258180+00:00 | succeeded | fb666bf699a98672d1d07640f63a23781b01f7ccc9b88daebd812a6572d2bd09 | 22,839 | null | """RecurrentGPT v2 — place-value aware looped transformer.
v1 memorized every Easy split perfectly (100% train exact) while scoring below
the label prior on held-out prompts. The diagnosis is not capacity or budget:
the model had no way to represent *which digit it was looking at*, so a lookup
table was the only reach... | {
"id": "41b28d4d-abeb-4cb0-bada-ba216f7bed74",
"created_at": "2026-08-04 05:35:02.258180+00:00",
"db_md5": "59e69d32f8d6055a214912efc2400f8e",
"submitter": "Amr Hegazy",
"github_login": "Amr-Hegazy1",
"run_id": "fbfacc3c-5b3e-4542-88c4-627f28f6942e",
"tier": "easy",
"dataset_id": "e3",
"status": "suc... | {
"score": {
"mean_loss": 2.1381528033685226,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.01562500004656613
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.138206184737864,
"example_count": 800,... |
41b33374-4ed9-443d-80a6-b2efecd8b61d | easy | briheet | 2026-08-05 12:17:02.357396+00:00 | succeeded | ab7ce9cc2824bac4c190c55fc3255e19d864ed1aad60a851f6ec5052c15a98ce | 5,515 | 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
LI... | {
"id": "41b33374-4ed9-443d-80a6-b2efecd8b61d",
"created_at": "2026-08-05 12:17:02.357396+00:00",
"db_md5": "b33bf7406db2455d06d17df3d041044c",
"submitter": "Aizenn",
"github_login": "briheet",
"run_id": "86343572-c545-44e3-9aa9-0dff0cd57665",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",... | {
"score": {
"mean_loss": 3.4001946449279785,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03666666615754366
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.975893020629883,
"example_count": 100,... |
41b940a9-15ac-4794-bae6-26f824573697 | easy | DDanlov | 2026-08-12 20:20:36.252906+00:00 | failed | 3b8f16a61562e98e0141397e9a2edf07c10255f5c5d718e49dccf9156216509f | 25,938 | null | from __future__ import annotations
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim.optimizer import Optimizer
try:
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
)
except ... | {
"id": "41b940a9-15ac-4794-bae6-26f824573697",
"created_at": "2026-08-12 20:20:36.252906+00:00",
"db_md5": "533d90befa5805b66a630e783eb53e98",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "9bcf6e5a-4a25-44ff-a3b2-d89da179b421",
"tier": "easy",
"dataset_id": "e3",
"status": "failed",
... | null
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.