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 |
|---|---|---|---|---|---|---|---|---|---|---|
0d43e91b-59e7-48e8-bfdc-6bf87edb7d5c | easy | Dandandan | 2026-08-04 15:11:56.433905+00:00 | succeeded | d00930beb96165096a97c995307eb36ccb0f4983bb76149382df2b37e74265d9 | 17,735 | null | """Multi-start learned spectral recurrence for One Layer Deeper.
Every coefficient that determines a prediction is randomly initialized and
updated end to end. Shared candidate transitions are unrolled to the prompt's
training depth; self-calibration and an integer-boundary loss favor a reusable
functional root withou... | {
"id": "0d43e91b-59e7-48e8-bfdc-6bf87edb7d5c",
"created_at": "2026-08-04 15:11:56.433905+00:00",
"db_md5": "c798cc643d6416345564a0ee81abb7fe",
"submitter": "Daniël Heres",
"github_login": "Dandandan",
"run_id": "105350b2-862c-4bca-852b-c7cc3340bc04",
"tier": "easy",
"dataset_id": "e1",
"status": "suc... | {
"score": {
"mean_loss": 1.1271511084487429e-05,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.1280336366326082e-05,
"example_count": 100,
... |
0d4c495f-ef3c-4743-97d0-13a083baa890 | easy | Maxwells-Demons | 2026-08-11 10:44:16.436931+00:00 | succeeded | 18417ae9b7e9322a284d59153dc69e090b2a9f7ca44d077de2975fc12310aee0 | 38,339 | null | """One shared digit operator, its two free neighbours of x, and two learned maps.
The backbone is unchanged from v47/v96: the state IS the digit one-hot, one
application of the operator is
(state digits, N digits) -> next state digit logits
and the straight-through code of those logits is the whole state handed to... | {
"id": "0d4c495f-ef3c-4743-97d0-13a083baa890",
"created_at": "2026-08-11 10:44:16.436931+00:00",
"db_md5": "8065efdb953888adbc468a29ff75e148",
"submitter": "Maxwells_Ayakashi",
"github_login": "Maxwells-Demons",
"run_id": "70426456-441b-46aa-b1ea-361971a2e46b",
"tier": "easy",
"dataset_id": "e3",
"st... | {
"score": {
"mean_loss": 2.2892194342780505,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.4068750005960464
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.7629606166104583,
"example_count": 800,... |
0d5318fc-a6b4-48d7-8de2-73b00e53510d | easy | oupadhyay | 2026-08-15 22:14:15.029736+00:00 | succeeded | 5aae6360b6bca923364b08ed06f1fc95a142e70713575d970f9799ac8192033f | 3,995 | null | """Conditional optimizer phase 2: lr10_warm8."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import nn
from benchmark import ModelSpec,OptimizerSpec,OptimizerBundle,Submission,assert_model_state
W,D,H=4,64,32
class C:
def __init__(s,vocab_size,max_seq_len):s.vocab... | {
"id": "0d5318fc-a6b4-48d7-8de2-73b00e53510d",
"created_at": "2026-08-15 22:14:15.029736+00:00",
"db_md5": "0899fae7af272f7abc33265e3a5db13f",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "9313a844-baf3-4011-995e-e0d6ffc5a680",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.412807687788338,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.012916666691501936
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.313384321238428,
"example_count": 600,... |
0d5330b5-ccd3-49e1-a5f6-ac2bd9b4196f | easy | rteehas | 2026-08-19 11:57:09.713452+00:00 | succeeded | 366a9358ec4fc99f2a19c283fc17d88e74743ac006b6338c450ed4803e5e6cd0 | 56,138 | null | """Anonymous marginal-program submission.
Mechanism
---------
The model is a differentiable probabilistic program. Its forward pass
computes, exactly, the answer-token predictions of a small structured latent
hypothesis space:
(prompt segmentation = which token types head the three fields)
x (field-role permutat... | {
"id": "0d5330b5-ccd3-49e1-a5f6-ac2bd9b4196f",
"created_at": "2026-08-19 11:57:09.713452+00:00",
"db_md5": "78db9f1fb62e5e57530e1866657d5193",
"submitter": "Ryan Teehan",
"github_login": "rteehas",
"run_id": "d024cb16-ee9f-43e8-9d34-2f41dfa846a4",
"tier": "easy",
"dataset_id": "e2",
"status": "succee... | {
"score": {
"mean_loss": 2.833213448524475,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8332133293151855,
"example_count": 300,
"exac... |
0d5551b9-972d-4245-b9cd-0265a49f9e8d | easy | DDanlov | 2026-08-16 00:23:33.494079+00:00 | succeeded | 7125587d6b6455f86770c1209641d04880c686b3c6394aaea9c59106d0ec1d1d | 16,322 | null | """
Production Port-Hamiltonian RoPE Dynamic Submission
Config: Run2_Config_freq1 (dim=512, stages=8, dt=0.08, lr=0.005, wd=0.05, update_freq=1)
"""
from __future__ import annotations
import math
from typing import Optional, Tuple
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim impo... | {
"id": "0d5551b9-972d-4245-b9cd-0265a49f9e8d",
"created_at": "2026-08-16 00:23:33.494079+00:00",
"db_md5": "42a8f1fd4c34fbfb99521f465405c2d4",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "adafae12-0e3c-4ff3-b72e-6245ec894009",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.2151466488998626,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2493968266542717,
"example_count": 600,
"e... |
0d5682dc-85ac-4a9e-b0ac-1d1d89344203 | easy | KaustubhKumar05 | 2026-08-27 03:29:35.712483+00:00 | succeeded | f146348424a120672a72b8c45d01fa900e0356501aa61d811973428dace34880 | 15,454 | null | """abacus_infomax — candidate B2: B, plus a loss that makes the digit prior costly.\n\nSee infomax_loss below for the motivation (§44). Architecture is candidate B\nunchanged, so a difference in result is attributable to the loss alone.\n\nOriginal B docstring follows.\n\nabacus_modsquare — candidate B: digit-ALIGNED a... | {
"id": "0d5682dc-85ac-4a9e-b0ac-1d1d89344203",
"created_at": "2026-08-27 03:29:35.712483+00:00",
"db_md5": "ab9046ee9f32cb887d1b68445ae016fa",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "ab927878-309e-4b10-9ec8-87f1a6fee117",
"tier": "easy",
"dataset_id": "e3",
"status": "succee... | {
"score": {
"mean_loss": 9.088882860307047,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.010625
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.179273040042888,
"example_count": 800,
"... |
0d5ddff0-bdb8-45f4-9d11-d7dacbda2d62 | easy | kushasareen | 2026-08-27 03:18:13.272300+00:00 | succeeded | 628e33a63f8adebbfae3a8a64d0324c41f4616bdfaaf1a03e749b14e25ddecdc | 9,984 | null | """hosted_dc_muon_reuse3: hosted_dc_muon plus 3 optimizer updates per fetched batch.
Each number's token is a learned function of its digits: every place has
its own embedding table, the place embeddings are summed and passed
through a small MLP. Unseen numbers therefore get a meaningful vector.
The transformer works ... | {
"id": "0d5ddff0-bdb8-45f4-9d11-d7dacbda2d62",
"created_at": "2026-08-27 03:18:13.272300+00:00",
"db_md5": "b78e4633164dd08ba8cc769e84ec77f7",
"submitter": "Kusha Sareen",
"github_login": "kushasareen",
"run_id": "835f8896-8a7e-49cf-a8d2-adc62af9f9f4",
"tier": "easy",
"dataset_id": "e2",
"status": "s... | {
"score": {
"mean_loss": 3.006168320775032,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.4477083459496498
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.661386966705322,
"example_count": 300,
... |
0d6050df-9d7c-4963-a779-657640425c97 | easy | yashkant | 2026-08-26 07:40:53.199350+00:00 | succeeded | c8e34ad53261d56e75bc7f6ae99336de2f0b7a820527041192e576ec06b3ca46 | 24,378 | null | """One Layer Deeper HARD submission v7: prime-agnostic repeated-squaring digit machine.
The entire model is three tied digit cells (~86K params):
mul: (digit, digit, carry 0..8) -> digit, carry [schoolbook partial products]
add: (digit, digit, carry 0/1) -> digit, carry [partial accumulation, N-doubling]
... | {
"id": "0d6050df-9d7c-4963-a779-657640425c97",
"created_at": "2026-08-26 07:40:53.199350+00:00",
"db_md5": "df22b886f008029388f951288ce89607",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "09252077-0d50-4255-9045-c8fe9fa33ce6",
"tier": "easy",
"dataset_id": "e5",
"status": "succeed... | {
"score": {
"mean_loss": 2.372094305450315,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.003333333277453979
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.3770697116851807,
"example_count": 600... |
0d64cbb9-e73a-42f3-969a-e2b7b02af93c | easy | DDanlov | 2026-08-17 19:37:43.806485+00:00 | succeeded | 981200ba68a894dc7a2ea4d4e9e792cac2d28c7f822b3a1181359af6065083c8 | 24,905 | 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": "0d64cbb9-e73a-42f3-969a-e2b7b02af93c",
"created_at": "2026-08-17 19:37:43.806485+00:00",
"db_md5": "424930e6a2fdef2521a97067ae159b6e",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "d6f24654-21af-407d-9b0f-f9605e14cd9e",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 236.78188113312223,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.09333333387970924
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 247.69953497132258,
"example_count": 100... |
0d66cc84-b1b0-45be-b971-069af8bc7b21 | easy | briheet | 2026-08-07 02:52:34.667372+00:00 | succeeded | cfaea3bd3324b8211876a7cb563ccfbb2a0787243398d4f8bdeed40100dccc7b | 4,021 | 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 = 64
NUM_HEADS = 4
c... | {
"id": "0d66cc84-b1b0-45be-b971-069af8bc7b21",
"created_at": "2026-08-07 02:52:34.667372+00:00",
"db_md5": "3858dae8973f4a67d99da915c331bbb7",
"submitter": "Aizenn",
"github_login": "briheet",
"run_id": "573a3037-816c-452b-8247-7d7ff7648b7d",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",... | {
"score": {
"mean_loss": 2.458151340484619,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.021666666492819786
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.5407660007476807,
"example_count": 100... |
0d684443-8763-40f8-8e78-776327983ea2 | easy | Ranoobaba | 2026-08-17 21:48:12.191332+00:00 | succeeded | 28f2d4657e8cf2c0e6c59f3da1e43a0d7e3ffb63e5ad8c6a15395294f61e7aae | 5,902 | null | """Looped transformer, iteration count conditioned on T read from the prompt.
Core idea: learn one latent "squaring step" operator and apply it T times.
Depth generalization comes from the loop structure itself; the model never
needs more parameters for deeper problems, only more iterations.
"""
from __future__ impor... | {
"id": "0d684443-8763-40f8-8e78-776327983ea2",
"created_at": "2026-08-17 21:48:12.191332+00:00",
"db_md5": "8e78491032b65cc834d993f269876021",
"submitter": "Syed Rayyan Ali",
"github_login": "Ranoobaba",
"run_id": "48a3c7f8-127e-4bd8-a80a-e56b1168d94e",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 6.8325890305630885,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.046666666095455486
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.798558235168457,
"example_count": 100... |
0d76dd56-606f-4071-ab93-5dde4a904ac8 | easy | oupadhyay | 2026-08-07 14:24:12.220471+00:00 | succeeded | 785208f7407418a5cd06a40890295126bd6218fdd3f0e1367de455e0515f3bdf | 6,608 | null | """Frozen E1 bidirectional decimal scan, packaged for the full remote budget."""
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, sp... | {
"id": "0d76dd56-606f-4071-ab93-5dde4a904ac8",
"created_at": "2026-08-07 14:24:12.220471+00:00",
"db_md5": "22d73edd34933fc65747ac11e08b7c99",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "9a795cbc-ae8b-4569-b194-1f4084e4e61c",
"tier": "easy",
"dataset_id": "e1",
"status": "s... | {
"score": {
"mean_loss": 3.4003335178047047,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.154999996026357
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.728949785232544,
"example_count": 100,
... |
0d7c9d96-590f-4e71-8501-4458acd09678 | easy | chad-atexpedient | 2026-08-09 20:15:57.383021+00:00 | succeeded | b97e39010bcefe6d24f8bd2bb5495e9d5e53c5327f82f965c46c0efd139c88e9 | 10,027 | null | """BL5A: BL1 post-400 LR 1e-4 continuation."""
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
IMMUTABLE_... | {
"id": "0d7c9d96-590f-4e71-8501-4458acd09678",
"created_at": "2026-08-09 20:15:57.383021+00:00",
"db_md5": "2dcbd32afea71272821b5401c12cfb5e",
"submitter": "chad-atexpedient",
"github_login": "chad-atexpedient",
"run_id": "42407c79-81d1-4fee-9f6c-c09b3f9e322a",
"tier": "easy",
"dataset_id": "e5",
"st... | {
"score": {
"mean_loss": 3.0261201566544313,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004583333333333333
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.0054520961949645,
"example_count": 60... |
0d8f1d3a-ef4f-4995-bb47-86d598688236 | easy | blake-camp-surge | 2026-08-04 02:45:46.861492+00:00 | succeeded | 1f6d683f6df9da32e71a97e3b45a7d07e90376fc53946272fed8e27340e01de1 | 10,022 | null | """Vector-state virtual neurons driven by one shared QKV-generating DAN."""
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,
... | {
"id": "0d8f1d3a-ef4f-4995-bb47-86d598688236",
"created_at": "2026-08-04 02:45:46.861492+00:00",
"db_md5": "a7555d7372198e1bb4d1160cca1958ee",
"submitter": "Blake Camp",
"github_login": "blake-camp-surge",
"run_id": "522a574b-0942-4ece-9d39-78ba02355f6e",
"tier": "easy",
"dataset_id": "e1",
"status":... | {
"score": {
"mean_loss": 1.9885045886039734,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.06333333067595959
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.046456813812256,
"example_count": 100,... |
0d900593-8321-462f-8176-1120242d197b | easy | sapient-sapiens | 2026-08-18 21:08:05.304635+00:00 | succeeded | 7aed751b1cfc6173c5e35b6fc1f460cdf1e800e43be133d0f2641edf285ae55a | 31,886 | null | """Depth-quantized recurrence with a mixture-over-depth objective.
The recurrent state is a bank of right-aligned digit slots. Every operator
application is followed by a soft quantization back onto the token simplex, and
the digit logits produced by that quantization are the answer logits at that
depth. There is no s... | {
"id": "0d900593-8321-462f-8176-1120242d197b",
"created_at": "2026-08-18 21:08:05.304635+00:00",
"db_md5": "d9f6a64017acd32e0b42549dbb060586",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "1c8860d2-6493-4daf-9d35-2fda12ec9451",
"tier": "easy",
"dataset_id": "e9",
"status": "su... | {
"score": {
"mean_loss": 3.7805960178375244,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05555555783212185
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.6899538040161133,
"example_count": 90,... |
0d948da2-2ff6-458a-8177-8331080f36df | easy | erdavis0 | 2026-08-27 02:03:56.878150+00:00 | succeeded | 795df3f9652030e991e92998273fd7a4939c775c6b6c6c9395c3028f971d4e8b | 31,304 | null | """Typed-prefix reversible transition B4: oracle-weighted step training.
Public delimiter syntax routes opaque N, X, and T fields into three strict
functional roles. X initializes only an evolving auxiliary tape, N initializes
only an invariant environment lane, and T reaches only a learned monotone
prefix controller... | {
"id": "0d948da2-2ff6-458a-8177-8331080f36df",
"created_at": "2026-08-27 02:03:56.878150+00:00",
"db_md5": "86ddb36d905f79e81813767c4d60fc8f",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "9afec2fb-18fc-41bc-aa07-1484a123d743",
"tier": "easy",
"dataset_id": "e9",
"status": "succeeded",... | {
"score": {
"mean_loss": 4.2018725872039795,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03444444481283426
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.507784843444824,
"example_count": 90,
... |
0d9726af-bdac-412d-93c2-77ced9450ae7 | easy | KaustubhKumar05 | 2026-08-25 15:56:34.919697+00:00 | succeeded | cede8f877bf90dc9d38af3879dc3680a73b6693631c3476085830cfc30a955d0 | 16,027 | null | """final_v1: culmination candidate. v8 (N-FiLM) + Muon momentum 0.5 + tail-EMA.
A small MLP embeds the modulus digits (mean-pooled over the N span) and
emits per-channel scale/shift applied to the cell input and a bias on the
update gate, every iteration. Hypernetwork-lite: the executor is modulated
by N rather than h... | {
"id": "0d9726af-bdac-412d-93c2-77ced9450ae7",
"created_at": "2026-08-25 15:56:34.919697+00:00",
"db_md5": "4c081115900db9bcea126ae905aeee44",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "65b428ec-3bd0-46a9-8de9-6e3ce9e03f0b",
"tier": "easy",
"dataset_id": "e8",
"status": "succee... | {
"score": {
"mean_loss": 6.717129230499268,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.11145276203751564
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.288947105407715,
"example_count": 85,
... |
0d976b03-659f-40d1-a1dc-afa962e159e7 | easy | alirezashirvani-jr | 2026-08-30 21:45:39.373898+00:00 | succeeded | 3fa6488bbbb697ef739fb15cafcce44b37ecb85477b348af20ab3ea038d5545c | 24,483 | null | from __future__ import annotations
import math
import time
from contextlib import nullcontext
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import ModelSpec, OptimizerBundle, OptimizerSpec, Submission, TokenLossBatch, assert_model_state
DIGIT_CARDINALITY = 10
STEP_FIELD_W... | {
"id": "0d976b03-659f-40d1-a1dc-afa962e159e7",
"created_at": "2026-08-30 21:45:39.373898+00:00",
"db_md5": "41fd7a05980039664511de3256ab2ebb",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "e88fd121-3798-4ee4-8d46-c1ba1633520c",
"tier": "easy",
"dataset_id": "e4",
... | {
"score": {
"mean_loss": 2.566459583202157,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.13812500004967054
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.770136544047509,
"example_count": 1200,... |
0d97cc96-55fa-4b79-8ade-44aeffb80d6a | easy | karanganesan | 2026-08-06 04:56:46.693322+00:00 | succeeded | 425b4b9587cc591a58d9cf1f3c301e0a58e38c878c8147476971c26ad2fddaca | 25,599 | 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": "0d97cc96-55fa-4b79-8ade-44aeffb80d6a",
"created_at": "2026-08-06 04:56:46.693322+00:00",
"db_md5": "d43c88c41571fcfac018f884a7bf6458",
"submitter": "Karan Ganesan",
"github_login": "karanganesan",
"run_id": "65375fbc-3d37-4f23-8292-a53071012ebb",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 8.230953693389893,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.014999999664723873
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.63919448852539,
"example_count": 100,
... |
0d9b572a-ed04-4bef-bb41-085e03b914af | easy | AdityaVG13 | 2026-08-15 12:05:30.617439+00:00 | succeeded | ca2013c127003658d7b4498e9ed63313ba686e427a79bdf1193341752bded30f | 109,685 | null | """AEGIS V13 v13_margin_esc: mix carry only when the publish cell is weak.
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 trivale... | {
"id": "0d9b572a-ed04-4bef-bb41-085e03b914af",
"created_at": "2026-08-15 12:05:30.617439+00:00",
"db_md5": "35533502910769b0cc7db8d54fe7b72a",
"submitter": "AdityaG",
"github_login": "AdityaVG13",
"run_id": "83528396-eaf0-40ad-bf07-f9b227c2eccb",
"tier": "easy",
"dataset_id": "e1",
"status": "succeed... | {
"score": {
"mean_loss": 1.7015369534492493,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.6299999952316284
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.8381575345993042,
"example_count": 100,... |
0d9bf37e-f042-4be5-9b61-d70efedf5dc6 | easy | KaustubhKumar05 | 2026-08-29 04:50:34.333654+00:00 | succeeded | 9e78834bc8d29c8c666699d106924b822ed8f4db3a92a423b145a273b131d35d | 14,473 | null | """op2_t1w -- stock cell_op2 + T=1 loss upweighting (the scored rung).
Forked from cell_op2. ORIGINAL DOC BELOW.
cell_op2 -- ONE shared squaring operator with internal depth, applied T times.
WHY THIS SHAPE (verified on the real data, not assumed):
answer(T=t) == x squared t times mod N 20000 / 20000 e... | {
"id": "0d9bf37e-f042-4be5-9b61-d70efedf5dc6",
"created_at": "2026-08-29 04:50:34.333654+00:00",
"db_md5": "c77647aaa38a99a3fe9be44d986ad436",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "cf218db4-5b6b-4c8b-bb8d-4b991f65fdcf",
"tier": "easy",
"dataset_id": "e5",
"status": "succee... | {
"score": {
"mean_loss": 3.57368244204711,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007083333345750968
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.6631608867324523,
"example_count": 600,... |
0d9e317f-0ee4-41da-9967-274d8adbb80e | easy | alirezashirvani-jr | 2026-08-30 00:16:44.382941+00:00 | succeeded | 672293a76323da62e296a3ee3f10c6ca38ac913874438ad7921545a2384944f0 | 31,682 | null | from __future__ import annotations
import math
import time
from contextlib import nullcontext
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_model_state,
)
MODEL_... | {
"id": "0d9e317f-0ee4-41da-9967-274d8adbb80e",
"created_at": "2026-08-30 00:16:44.382941+00:00",
"db_md5": "310e8a85d3554c9bd27bdd0854d43589",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "0e62db59-fac9-4aaa-b35a-0cc34d98210c",
"tier": "easy",
"dataset_id": "e6",
... | {
"score": {
"mean_loss": 1.0221661031246185,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.7362612783908844
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.7963672280311584,
"example_count": 60,
... |
0d9e8c37-0b6b-4619-8967-239dff7f52bd | easy | AFeinfield | 2026-08-19 23:28:24.616815+00:00 | succeeded | bec664ceff21c02f9ef50d16d4b06dd144235cd6140767c58eab8e7aeb2632d5 | 21,653 | null | """092: INSTRUMENTED BASE — two measurement fixes, no architecture change.
(1) Step-indexed LR schedule with a fixed screening step budget. The old
schedule decayed on WALL CLOCK, so throughput jitter changed the
LR-per-step trajectory: identical files on e2 gave 1470 vs 1025 steps
and train acc 0.725 vs 0... | {
"id": "0d9e8c37-0b6b-4619-8967-239dff7f52bd",
"created_at": "2026-08-19 23:28:24.616815+00:00",
"db_md5": "f9ddcb1b28d32530a5a54b8d59a29bba",
"submitter": "AFeinfield",
"github_login": "AFeinfield",
"run_id": "a982b1ba-2efa-4660-a031-9a8046b2bb74",
"tier": "easy",
"dataset_id": "e2",
"status": "succ... | {
"score": {
"mean_loss": 2.2234326601028442,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.015416666865348816
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2189977169036865,
"example_count": 30... |
0d9fd884-912e-4309-bd0d-c3482625c75f | easy | lpbb | 2026-08-19 12:56:30.696338+00:00 | succeeded | b27db0f1e82ec54df611c5b185297a86733e346a9ceeeeee09f0132824d6e2f0 | 9,874 | null | """S0 + RoPE + abacus embeddings + dropout, with a looped block: instead of
`num_layers` independent layers, a block of `num_layers` unique layers is run
`num_repeats` times with shared weights (e.g. 12 layers x 4 repeats = depth 48
but only 12 layers' worth of parameters). Weight tying across repeats forces
the block ... | {
"id": "0d9fd884-912e-4309-bd0d-c3482625c75f",
"created_at": "2026-08-19 12:56:30.696338+00:00",
"db_md5": "871553bdf97f0489aa1d43e6ac1007bf",
"submitter": "Lpbb",
"github_login": "lpbb",
"run_id": "44037eba-3588-4cb8-9240-fd48283cdb9a",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
"s... | {
"score": {
"mean_loss": 4.367848873138428,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03499999828636646
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.677396297454834,
"example_count": 100,
... |
0da47fe2-0630-47da-8b9f-c2e15ce145de | easy | alirezashirvani-jr | 2026-08-23 21:03:44.306212+00:00 | failed | da07d0759da095689fb93dfae23fa3a70b6ac1fda0d719474c820a1e0d71a3bb | 48,613 | null | """Looped digit machine whose transition is assembled by learning.
The recurrent state is the decimal digit string of the current value; one
loop application applies one learned transition chosen by a gate over an
enumerated polynomial family (all mod N): pure powers x^k for k in 1..9,
x^2+c and x^3+c for c in 1..9, t... | {
"id": "0da47fe2-0630-47da-8b9f-c2e15ce145de",
"created_at": "2026-08-23 21:03:44.306212+00:00",
"db_md5": "b3880863c0af7a7fbe5064fd497cfdf3",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "c221c6a4-b736-4c0c-88af-02d8bb68ae4c",
"tier": "easy",
"dataset_id": "e5",
... | null
|
0dbb03bc-d34a-4f17-bd4b-3bd8836f00cb | easy | erdavis0 | 2026-08-29 01:38:14.567623+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": "0dbb03bc-d34a-4f17-bd4b-3bd8836f00cb",
"created_at": "2026-08-29 01:38:14.567623+00:00",
"db_md5": "be2bf3989baac5e04978d28d9e6dbbb4",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "2a16cdb5-1427-4b43-a693-6b73c9c16bec",
"tier": "easy",
"dataset_id": "e10",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.98395699262619,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.14062626846134663
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.4765042066574097,
"example_count": 125,
... |
0dbe7f41-2707-4452-96e1-cec629b8b7c1 | easy | sapient-sapiens | 2026-08-29 21:52:34.169035+00:00 | failed | 420ad9fc237668c4fc9d7abb5803bd98356d239461cb68575c1b8956ac6dca83 | 15,632 | null | """Three-loop one-layer Transformer for iterated modular squaring.
The learned arithmetic operator is one tied pre-norm Transformer block. Every
square makes four Horner-digit calls and every digit makes nine micro-calls.
N and the complete operand are immutable during a square; scratch resets at a
digit boundary; an... | {
"id": "0dbe7f41-2707-4452-96e1-cec629b8b7c1",
"created_at": "2026-08-29 21:52:34.169035+00:00",
"db_md5": "f95030f1b3169b674112f35778cdb482",
"submitter": "Ertondy",
"github_login": "sapient-sapiens",
"run_id": "92f1285e-82fd-4331-bbb4-711f38c33be5",
"tier": "easy",
"dataset_id": "e5",
"status": "fa... | null
|
0dc30c76-3f2f-4c6a-a3a4-8f575618435d | easy | jordanrubin | 2026-08-02 20:27:37.356435+00:00 | succeeded | 16a52d22982dfef736863488be48d497f494793eb9aba7fdf8b93962885bc8fc | 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": "0dc30c76-3f2f-4c6a-a3a4-8f575618435d",
"created_at": "2026-08-02 20:27:37.356435+00:00",
"db_md5": "1c62e1fd584dfb5f0181cb2641199911",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "101bb88e-8f07-4ea2-b5a5-35b5f89e35be",
"tier": "easy",
"dataset_id": "e1",
"status": "s... | {
"score": {
"mean_loss": 4.7323095281535785,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.19666666040817896
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.861814712410543,
"example_count": 100,... |
0dc5e0ce-ef3a-45c9-acfc-323ca18af029 | easy | DDanlov | 2026-08-16 00:43:20.204637+00:00 | succeeded | e064ec24660bc3cc6615f88f18eb6c657c2c8b0cb9fd1bcb02a03e4291889896 | 16,290 | null | """
Production Port-Hamiltonian RoPE Dynamic Submission
Config ID: 11 | Sweep: update_freq=5 | lr=0.015, wd=0.03
"""
from __future__ import annotations
import math
from typing import Optional, Tuple
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Optimizer
try:
from benc... | {
"id": "0dc5e0ce-ef3a-45c9-acfc-323ca18af029",
"created_at": "2026-08-16 00:43:20.204637+00:00",
"db_md5": "c784d4257d15dbec6dd77247af7d2a8e",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "13423676-962e-458d-9e27-440282168b85",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.252287304211348,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00875
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.330157012682859,
"example_count": 600,
"e... |
0dc749ee-3c6f-443c-9cda-cd61d6813aad | easy | Ranoobaba | 2026-08-16 10:05:27.001928+00:00 | succeeded | 5785d1fec0c4478e7825c565216b36912a798d8c454ad8e71cee23d0e7d2fc5d | 5,723 | null | """Looped transformer, iteration count conditioned on T read from the prompt.
Core idea: learn one latent "squaring step" operator and apply it T times.
Depth generalization comes from the loop structure itself; the model never
needs more parameters for deeper problems, only more iterations.
"""
from __future__ impor... | {
"id": "0dc749ee-3c6f-443c-9cda-cd61d6813aad",
"created_at": "2026-08-16 10:05:27.001928+00:00",
"db_md5": "714239d629fe3ff0d9fe24f6c9b1d7a5",
"submitter": "Syed Rayyan Ali",
"github_login": "Ranoobaba",
"run_id": "860d9ab5-5369-4759-b234-7c9bc926b0f8",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 5.864157996570083,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05166666691501935
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.80423641204834,
"example_count": 100,
... |
0dc85739-d022-4dc7-8e6d-9deb5c5d9d90 | easy | viridale | 2026-08-04 22:42:18.937798+00:00 | succeeded | 05a9cb419b508cb5f3a6757e3aedfcb5a79aecf9c4ec2543edd83ee56e54eefd | 9,748 | null | """Clean product-quantized autonomous recurrent token model.
hypothesis: v174 can memorize E1 but does not transfer, so a shared hard categorical
recurrent state may impose the missing compact-program bias and prevent continuous
trajectory drift without supplying any task semantics.
axis: arch
target: Easy e1 ... | {
"id": "0dc85739-d022-4dc7-8e6d-9deb5c5d9d90",
"created_at": "2026-08-04 22:42:18.937798+00:00",
"db_md5": "018d9b479a88ca30a97e7844dc95c8f4",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "cfc73ba8-ec25-487a-969d-cc00cd93fae9",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 3.905505657196045,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0283333333209157
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.8559000492095947,
"example_count": 100,
... |
0dcaa761-6880-42d1-92d2-08795c0f6083 | easy | erdavis0 | 2026-08-29 01:42:11.467540+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": "0dcaa761-6880-42d1-92d2-08795c0f6083",
"created_at": "2026-08-29 01:42:11.467540+00:00",
"db_md5": "be2bf3989baac5e04978d28d9e6dbbb4",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "02a4efdf-66e5-4915-979b-d948ad0ef41e",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded",... | {
"score": {
"mean_loss": 4.322494607831026,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.020000000009313228
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.337114849058479,
"example_count": 800,... |
0dcc7a77-c0d7-42e2-b3d9-366269de76e5 | easy | liam-gb | 2026-08-11 16:54:18.980844+00:00 | succeeded | 51092a03dac4589b3952d22bca69d237700882100bc8598ae88aa56ca04d5e9b | 12,431 | null | """rns_pm_gather: rns_critical_pm with the readout replaced by a direct
slot gather.
Single change vs rns_critical_pm: the token transformer is DELETED. The
pooled loop state feeds a per-distance digit head (three vocab rows, one
per distance-from-units), and each output position p reads the digit at
distance prompt_l... | {
"id": "0dcc7a77-c0d7-42e2-b3d9-366269de76e5",
"created_at": "2026-08-11 16:54:18.980844+00:00",
"db_md5": "f103f404f55ab6ba8e1489a2fa9a39a0",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "e0ca106f-17b7-4e78-9ee6-5471f620e227",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 4.534654945135117,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.305000021122396
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.095727920532227,
"example_count": 100,
... |
0dcde078-4a20-49c5-8d20-42ab1a26eac5 | easy | KaustubhKumar05 | 2026-08-27 12:06:47.722637+00:00 | succeeded | 0e9f25a4cc7592f3adc331bfdb9a2f3566e06eb44239586933b609e7350a1085 | 14,701 | null | """abacus_sine — SINUSOIDAL activation instead of GELU.
§72: TWENTY candidates in this project and every single one used GELU -- the
activation was inherited from TransformerEncoderLayer and never once varied.
Salah & Yevick 2411.05353 report that grokking on modular arithmetic is
CONTROLLED by the activation function... | {
"id": "0dcde078-4a20-49c5-8d20-42ab1a26eac5",
"created_at": "2026-08-27 12:06:47.722637+00:00",
"db_md5": "445fe6eb1b47727ceb55b42fc155d11d",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "5fff3007-6d08-46db-bd29-726c596b2690",
"tier": "easy",
"dataset_id": "e3",
"status": "succee... | {
"score": {
"mean_loss": 3.1024995022279507,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.008125
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.093213333118649,
"example_count": 800,
... |
0dcf0a60-bcb7-4277-99bf-31335f48f840 | easy | lpbb | 2026-08-14 15:47:19.481057+00:00 | succeeded | f1659afe669f86256d580cd36d6f38fc24d14fc626196d28186773fd402c4eec | 12,706 | null | """CLEAN S16 — CYCLE-CONSISTENCY TOKEN INTERFACE on the S14 base (Muon kept).
S12's diagnosis: continuous state smuggles example identity across squarings.
S13's annealed lattice-snap failed because it enforced PROXIMITY to the token
lattice with no requirement that the snapped state still computes. Here the
boundary b... | {
"id": "0dcf0a60-bcb7-4277-99bf-31335f48f840",
"created_at": "2026-08-14 15:47:19.481057+00:00",
"db_md5": "cc422b809aac4c652a36dcd9f7e274ad",
"submitter": "Lpbb",
"github_login": "lpbb",
"run_id": "5d782558-0de0-46ea-9f22-9c5fdc25f698",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded",
"s... | {
"score": {
"mean_loss": 4.911688355151448,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.013125000000000001
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.520560732397353,
"example_count": 800,... |
0dd911a3-2560-4348-b087-1d0e0322c958 | easy | yashkant | 2026-08-14 16:12:23.990989+00:00 | succeeded | 20b6688f7b4e4b727395cfd99a07b398e63a90f4995736f5c766ef7a4fb8fd79 | 31,750 | null | """Round1701 Dykstra width16/hidden32 architecture pair on the D416 MemGLU carrier.
The fixed Round1141 carrier first produces its ordinary pooled representation.
A compact primal state and two Dykstra correction memories are initialized
from that immutable representation. Every parsed public-T cycle applies two
cond... | {
"id": "0dd911a3-2560-4348-b087-1d0e0322c958",
"created_at": "2026-08-14 16:12:23.990989+00:00",
"db_md5": "78be01b213383e39de0c9a468834e2be",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "44c2e0b8-bdd2-4f1a-9c32-745392046d91",
"tier": "easy",
"dataset_id": "e3",
"status": "succeed... | {
"score": {
"mean_loss": 2.1199149487645377,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.015625000158324836
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.117059137623602,
"example_count": 800... |
0de870c9-89c6-4ce8-8b6a-f7426383821b | easy | anirudh-chakravarthy | 2026-08-12 04:19:03.645666+00:00 | succeeded | aa2b45a1ee78cbeb03bac2a02dbf9da09b4f83b4a83c754f293c63bf42e00bc0 | 15,374 | 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": "0de870c9-89c6-4ce8-8b6a-f7426383821b",
"created_at": "2026-08-12 04:19:03.645666+00:00",
"db_md5": "923dde918fac56a3cfe44475b3185293",
"submitter": "Anirudh S Chakravarthy",
"github_login": "anirudh-chakravarthy",
"run_id": "9de1b093-d548-4d22-a233-9bf20641fe7f",
"tier": "easy",
"dataset_id": "... | {
"score": {
"mean_loss": 3.7385438296910642,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.012500000062088172
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.51290549932574,
"example_count": 600,... |
0df97b0e-df68-4f81-837b-870e412ab95f | easy | nikhilkumar26 | 2026-08-29 16:44:24.169225+00:00 | succeeded | 0799daf82e64000f420d6641d0e8892468b5cdfb6787752e97d065387148a94c | 7,620 | null | """
One Layer Deeper Easy E1:
digit-wise learned square F, composed T times, T=1-first loss.
x and N stay digit sequences. Shared F is one learned transition.
T only chooses how many times F is applied. No modular arithmetic
is implemented in Python.
"""
from __future__ import annotations
import torch
import torch.n... | {
"id": "0df97b0e-df68-4f81-837b-870e412ab95f",
"created_at": "2026-08-29 16:44:24.169225+00:00",
"db_md5": "5a2aade1f6ce1d464c3f31892d1df4f7",
"submitter": "Nikhil Kumar",
"github_login": "nikhilkumar26",
"run_id": "e0e9b684-285d-4066-a23d-a3c874c58c0d",
"tier": "easy",
"dataset_id": "e1",
"status": ... | {
"score": {
"mean_loss": 1.9316502213478088,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.07666666433215141
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.9020460844039917,
"example_count": 100... |
0dfb8fba-8f5e-458d-9cac-c797eaa84ffb | easy | viridale | 2026-08-02 10:03:53.127704+00:00 | succeeded | 707f29955d8b371eaa3a01206823f5989a819057482cf0d59783d28dc7d6ad33 | 32,526 | null | """One-step-trained active table with cumulative evidence commitment.
hypothesis: only T=1 rows enter the catalogue-open likelihood, so executing its
random hard table to T=2/4/8/16 during training is wasted work and produces
no valid evidence. Train the table with exactly one application on every
batch. A... | {
"id": "0dfb8fba-8f5e-458d-9cac-c797eaa84ffb",
"created_at": "2026-08-02 10:03:53.127704+00:00",
"db_md5": "9d68c07fceecf5026660100a0f20aa9c",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "803ae71b-f9fb-4768-8b50-32b66d886a46",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 0.003014546020381487,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.0030145461205393076,
"example_count": 100,
... |
0dfc085c-4694-4a67-830a-8e0f17ba1840 | easy | khushidahi | 2026-08-05 23:49:23.785978+00:00 | succeeded | 3d4b72539e0ea63888b67fd8853b68d15eff7d1312ed5c47c8720dc4d9461bd5 | 20,286 | null | """R21 hierarchical value-and-scratch 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 ta... | {
"id": "0dfc085c-4694-4a67-830a-8e0f17ba1840",
"created_at": "2026-08-05 23:49:23.785978+00:00",
"db_md5": "3806a76d0bbd27306ea79fb36e5af490",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "66ae0e07-94f8-4b77-bb7b-54508ca22a9a",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.203776717185974,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.012916667386889458
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1964240074157715,
"example_count": 600... |
0e006cd2-774d-43d9-96ab-ba84006c56a0 | easy | DDanlov | 2026-08-13 19:53:59.300509+00:00 | failed | 939f58d818c3f953053dfcbc068030b31142e50f323a87b6d7f84ab5df0f2dd5 | 26,904 | null | from __future__ import annotations
import math
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim.optimizer import Optimizer
try:
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
assert_model_state,
)
except ... | {
"id": "0e006cd2-774d-43d9-96ab-ba84006c56a0",
"created_at": "2026-08-13 19:53:59.300509+00:00",
"db_md5": "e3de2d9e8ccead933e2b8372c4ddba6b",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "ac05b6a8-dd4b-4474-8217-ae8c3bd350d7",
"tier": "easy",
"dataset_id": "e5",
"status": "failed",
... | null
|
0e066ad6-b2dd-481d-958c-b80b3ccd3d80 | easy | yashkant | 2026-08-10 20:28:12.215150+00:00 | succeeded | 8e3ad45dde3bc081bffc8263034b1e05b611d49e69651cefad2f7bc53a779744 | 31,137 | null | """Round1282 shared continued-fraction denominator-recursion pair.
The complete Round1141 D416 MemGLU carrier remains byte-identical. Its pooled
representation seeds a positive terminal denominator. Every shared call adds
one tied continued-fraction level, ``P(x) + softplus(beta)/(state + epsilon)``.
The matched contr... | {
"id": "0e066ad6-b2dd-481d-958c-b80b3ccd3d80",
"created_at": "2026-08-10 20:28:12.215150+00:00",
"db_md5": "60c097b303e1d8abc2a017917010a304",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "9002e70a-811b-4d31-8e6f-9eb7ac03b7d8",
"tier": "easy",
"dataset_id": "e2",
"status": "succeed... | {
"score": {
"mean_loss": 3.974305033683777,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.25958334282040596
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.961348056793213,
"example_count": 300,
... |
0e0c4a3b-66e6-4192-9baf-cb83fb6a142a | easy | ddoan | 2026-08-14 11:04:15.401420+00:00 | succeeded | 4b5e9d4891d24ab7ec629c0b1414e860f449590b277456fedae81d697d039ccd | 10,413 | null | """Place-aligned active memory with hard-gated diagonal CGRU updates."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
BatchReuseContext,
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatc... | {
"id": "0e0c4a3b-66e6-4192-9baf-cb83fb6a142a",
"created_at": "2026-08-14 11:04:15.401420+00:00",
"db_md5": "cf77a5c3ed3605f339d3ff420a9a3eae",
"submitter": "Doug Doan",
"github_login": "ddoan",
"run_id": "66479b2e-a8e8-4524-8188-fec5bdbd9a12",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.148338584680732,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.01687500013038516
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.149608883679189,
"example_count": 800,
... |
0e14ad13-1ec2-4ef4-8659-498421fde6b9 | easy | rougee | 2026-08-21 00:52:27.105944+00:00 | failed | 930f30400bb8afd2fd7a2464389c5e64d84aaddff2d1ec9032839282ed4a628d | 9,687 | null | """Submission for One Layer Deeper: Prelude -> Looped Hierarchical Dual-Stride Core -> Coda.
Architectural Innovations:
1. ModulusScaleEquivariantNorm: Scale-invariant normalization across residue fields Z_N*.
2. RotaryEmbedding: RoPE attention for causal sequence token processing.
3. LoopedAlgebraicCore: Hierarchical... | {
"id": "0e14ad13-1ec2-4ef4-8659-498421fde6b9",
"created_at": "2026-08-21 00:52:27.105944+00:00",
"db_md5": "5cb165a0700a7ebbf0ed35614bc6bb8e",
"submitter": "Rougee",
"github_login": "rougee",
"run_id": "34be6fa8-c60b-4007-88bd-67484e7511cf",
"tier": "easy",
"dataset_id": "e1",
"status": "failed",
"... | null
|
0e15cb0e-dd0e-4ab1-8abf-02c50ab43af6 | easy | 0Chris5R | 2026-08-01 07:52:12.881784+00:00 | succeeded | 5eb6a6ad6663ebf56f403eafc24d3b3b392215979e2910cbf9120747ccd0338b | 23,576 | null | """Shared-digit Neural GPU with late full-objective EMA evaluation.
The model aligns decimal digits by significance, forms a learned low-rank
second-order feature map of the current state, and evolves a two-dimensional
workspace with the original convolutional-GRU equations. Each arithmetic
transition applies the sam... | {
"id": "0e15cb0e-dd0e-4ab1-8abf-02c50ab43af6",
"created_at": "2026-08-01 07:52:12.881784+00:00",
"db_md5": "d1ecdef9362744b111123e441811f4a8",
"submitter": "Chris ",
"github_login": "0Chris5R",
"run_id": "6f1b320d-ab48-4f16-8db8-ff7bc9096bd5",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.643582759607373,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0333333333581686
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.5610785345325557,
"example_count": 600,
... |
0e16253e-6802-44d1-b7ff-f9322f186980 | easy | erdavis0 | 2026-08-24 09:23:17.542495+00:00 | succeeded | 568002a3578811fc9fcac082e03001dcd0e4ce92af578c0bd68ad148422483c8 | 7,802 | null | """Prompt-conditioned quadratic depth ladder with shared transition weights."""
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_sta... | {
"id": "0e16253e-6802-44d1-b7ff-f9322f186980",
"created_at": "2026-08-24 09:23:17.542495+00:00",
"db_md5": "b5d2b0f1a47b6c1c06b4888fde557d42",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "aa0a3a3c-5ec1-4a3b-acce-29ebf6540358",
"tier": "easy",
"dataset_id": "e9",
"status": "succeeded",... | {
"score": {
"mean_loss": 8.165979385375977,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.030000001192092896
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 8.245534896850586,
"example_count": 90,
... |
0e1cdad8-2001-471b-b51c-7e891f043a2a | easy | khushidahi | 2026-08-09 05:22:31.483924+00:00 | succeeded | eaece6c41c6fbc168171cfa99f3d5db1a094cc87cab44547c4e0ad7a78535a47 | 23,845 | 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": "0e1cdad8-2001-471b-b51c-7e891f043a2a",
"created_at": "2026-08-09 05:22:31.483924+00:00",
"db_md5": "f7b43a5876d9b346016283b8892b474b",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "0f881c48-ac80-42a8-be8f-bc34e8c1d3ac",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.1955083217304034,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.010416666691501935
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1946511268615723,
"example_count": 60... |
0e1d4cc2-270c-4ab0-9b41-5962726e4c5e | easy | adambrzosko | 2026-08-30 00:04:11.962068+00:00 | succeeded | e02785a30b7e65ac5d54132e66a3941c6143ffb5c5e57063a4b5b1034d65baa4 | 10,073 | null | """One Layer Deeper — looped-transformer starter.
Defaults below reflect the CPU-scale ablations in report.md. They are proxy
results on a guessed tokenization, not H100 results: re-run the sweeps before
trusting them. MAX_STEPS in particular must match the evaluator's real step
count or the LR schedule never leaves w... | {
"id": "0e1d4cc2-270c-4ab0-9b41-5962726e4c5e",
"created_at": "2026-08-30 00:04:11.962068+00:00",
"db_md5": "188b37bd412680a3e8b05e0a69c5aae1",
"submitter": "Adam Brzosko",
"github_login": "adambrzosko",
"run_id": "f56cd27e-acd1-404b-a840-9854bc942146",
"tier": "easy",
"dataset_id": "e6",
"status": "s... | {
"score": {
"mean_loss": 3.9862749576568604,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03851351607590914
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.718006134033203,
"example_count": 60,
... |
0e21577a-f463-410a-94ea-d88312fac496 | easy | alirezashirvani-jr | 2026-08-30 00:17:57.781587+00:00 | succeeded | 672293a76323da62e296a3ee3f10c6ca38ac913874438ad7921545a2384944f0 | 31,682 | null | from __future__ import annotations
import math
import time
from contextlib import nullcontext
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_model_state,
)
MODEL_... | {
"id": "0e21577a-f463-410a-94ea-d88312fac496",
"created_at": "2026-08-30 00:17:57.781587+00:00",
"db_md5": "310e8a85d3554c9bd27bdd0854d43589",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "99f24559-eaed-4ea4-9c55-48fe064c9e36",
"tier": "easy",
"dataset_id": "e7",
... | {
"score": {
"mean_loss": 1.2479238212108612,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.759358286857605
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.6740640997886658,
"example_count": 85,
... |
0e382b8e-1d7b-4744-876b-4e518ea2fb80 | easy | viridale | 2026-08-02 02:01:17.327907+00:00 | succeeded | 69eda608bd09be067289519cc05c2ac26cbd4581d281bce2e55ee613b563c2da | 23,540 | null | """Scripted graph-pure residual program beam with cumulative evidence.
hypothesis: v080 is mechanically Hard-ready but three prompt-derived values
still cross to Python control flow: digit-column liveness, maximum parsed
depth, and maximum answer width. Remove those Rule-8/Rule-9 exposures while
preserving... | {
"id": "0e382b8e-1d7b-4744-876b-4e518ea2fb80",
"created_at": "2026-08-02 02:01:17.327907+00:00",
"db_md5": "df0a807e854c741e3d65cf5ea098ca9b",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "267fe4e5-c416-4dc1-af56-78879bef4bfa",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 0.003014546214626436,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.00301454615599774,
"example_count": 800,
"... |
0e387be5-c0f5-4b95-8248-12ca65ef73b8 | easy | nikolageorgiev2000 | 2026-08-04 08:43:22.372003+00:00 | succeeded | fe4c052cf2410ed697f85a9fe1a83bd491557858255b038412af6fd79fe09c6e | 8,615 | null | """T-composed categorical-digit reasoner with fully learned pair/reduction maps."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
asser... | {
"id": "0e387be5-c0f5-4b95-8248-12ca65ef73b8",
"created_at": "2026-08-04 08:43:22.372003+00:00",
"db_md5": "1ef5988b8f2d3469a7e0a8c2c679c946",
"submitter": "Nikola Georgiev",
"github_login": "nikolageorgiev2000",
"run_id": "c462a486-0644-4200-8db8-6d3d2a5e59c0",
"tier": "easy",
"dataset_id": "e2",
"s... | {
"score": {
"mean_loss": 1.5703808963298798,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.7533333599567413
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.7113973498344421,
"example_count": 300,... |
0e42b0e7-1d9a-48cf-9163-a2f854289a24 | easy | KaustubhKumar05 | 2026-08-27 06:36:19.236527+00:00 | succeeded | dbcc71c4c0a38e23cc4e64c5be18cc668bb50ce47a094b8b5436f5fa84606a14 | 14,044 | null | """abacus_gromov — Gromov-style: NO attention, plain MLP, MSE loss, no regularisation.
§62: Gromov 2301.02679 groks modular arithmetic with a TWO-LAYER FULLY-CONNECTED
net, vanilla GD, MSE loss and NO regularisation, and derives closed-form weights.
His actual architecture does NOT port: it one-hot encodes operands o... | {
"id": "0e42b0e7-1d9a-48cf-9163-a2f854289a24",
"created_at": "2026-08-27 06:36:19.236527+00:00",
"db_md5": "cfe87322d620a601e25a918010f5954e",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "8b05381b-fa68-4f04-86a8-f701f4073b37",
"tier": "easy",
"dataset_id": "e3",
"status": "succee... | {
"score": {
"mean_loss": 2.7450764260166256,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0025
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.745061397452188,
"example_count": 800,
"e... |
0e450d5e-2e44-4d1c-a32c-fa30eeaaeebc | easy | gauravmishra | 2026-08-09 04:35:59.134563+00:00 | succeeded | bc9c95ea0d9e00ef0d8be60ce8c1f84146fa8a4ed2e6b45a9cad657db1442016 | 4,268 | null | """Basic single-pass Transformer with PyTorch AdamW."""
from __future__ import annotations
import torch
import torch.nn.functional as F
from torch import Tensor, nn
from benchmark import (
ModelSpec,
OptimizerBundle,
OptimizerSpec,
Submission,
TokenLossBatch,
assert_model_state,
)
D_MODEL = ... | {
"id": "0e450d5e-2e44-4d1c-a32c-fa30eeaaeebc",
"created_at": "2026-08-09 04:35:59.134563+00:00",
"db_md5": "ebb82a0cd79e9fd25402c2ba119a7aec",
"submitter": "Gaurav Mishra",
"github_login": "gauravmishra",
"run_id": "7460df3c-c847-48a5-97a5-1e826c4e6cfe",
"tier": "easy",
"dataset_id": "e5",
"status": ... | {
"score": {
"mean_loss": 2.2396655873272957,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0050000000248352685
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2633331923206828,
"example_count": 6... |
0e4cafa8-f502-491f-a530-2ebd07ca0744 | easy | liam-gb | 2026-08-13 13:08:31.869956+00:00 | succeeded | c98994ef0a17cbbc563e971c85fd060c426e82414551fdb5ce599d6575496cbc | 15,106 | null | """rns_pm_gather_ct_widet_pt: widet + per-prime temperature (16-prime bank).
The wide readout is what makes long answers reachable; the trim is what pays
for it, by not running loop iterations whose update is multiplied by zero.
Base docstring:
Single change vs rns_pm_gather_ct: the readout carries one
distance-from... | {
"id": "0e4cafa8-f502-491f-a530-2ebd07ca0744",
"created_at": "2026-08-13 13:08:31.869956+00:00",
"db_md5": "f909714a31387e09614b38df91ccc3f4",
"submitter": "liam-gb",
"github_login": "liam-gb",
"run_id": "c8f18961-0ac0-48a4-ae7a-baa035c4836a",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 5.2047900557518005,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.486666664481163
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.168439865112305,
"example_count": 100,
... |
0e5318b8-b351-4311-8d04-7eada9c1c8cb | easy | yashkant | 2026-08-11 22:22:45.400369+00:00 | succeeded | 36c44133e7753e3fc0d780348e6df5b50afef09a65d62a38736e4282cab2a777 | 31,765 | null | """Round1702 three-projections-per-T Dykstra recursion on the D416 carrier.
Each parsed public-T cycle applies an A-B-A sequence through one shared learned
smooth projection, reinjecting the immutable anchor on every call and updating
the corresponding Dykstra correction memory after each projection. All active
cycle... | {
"id": "0e5318b8-b351-4311-8d04-7eada9c1c8cb",
"created_at": "2026-08-11 22:22:45.400369+00:00",
"db_md5": "af932eb3600971a2c1c83c57fa57dc87",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "0d84f7ec-932a-4298-8849-90c40c998261",
"tier": "easy",
"dataset_id": "e2",
"status": "succeed... | {
"score": {
"mean_loss": 3.6496225595474243,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.24187501147389412
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.477901935577393,
"example_count": 300,... |
0e5366dd-9f6b-422c-be3e-61281dd48ee2 | easy | 0Chris5R | 2026-08-01 07:01:43.223666+00:00 | succeeded | f8f97cab63d72d718cac2f878297ee1d90dcc56024fd4bbc2b9e40058e96b101 | 19,793 | null | """Matched-depth control for the progressive shared Neural GPU.
The model aligns decimal digits by significance, forms a learned low-rank
second-order feature map of the current state, and evolves a two-dimensional
workspace with the original convolutional-GRU equations. Each arithmetic
transition applies the same tw... | {
"id": "0e5366dd-9f6b-422c-be3e-61281dd48ee2",
"created_at": "2026-08-01 07:01:43.223666+00:00",
"db_md5": "e614166ef3d194f46feaeae3c6b7853e",
"submitter": "Chris ",
"github_login": "0Chris5R",
"run_id": "db6c9a96-5d53-47e4-81ca-bed5d543c987",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.3462907413415564,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009166666691501936
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.29059334506903,
"example_count": 600,... |
0e58d3a7-cb29-45d5-b7a8-5a68a274bddc | easy | jayyala | 2026-08-16 14:06:19.267725+00:00 | succeeded | 3dcd8842faf5788ca6c5c4f18d4905239efc63c51abcd606900ce6c6fae313d0 | 14,806 | null | """A generic recurrent program learner with a canonical decimal state.
The model is deliberately unaware of the public benchmark's arithmetic rule.
It represents the current value as right-aligned categorical digit slots and
learns one transition from the complete prompt. The same transition is reused
exactly T times... | {
"id": "0e58d3a7-cb29-45d5-b7a8-5a68a274bddc",
"created_at": "2026-08-16 14:06:19.267725+00:00",
"db_md5": "4a303cf4e487005f10bc3776ee331d0b",
"submitter": "jayyala",
"github_login": "jayyala",
"run_id": "5a1fef64-217e-44c0-b593-49c79470af64",
"tier": "easy",
"dataset_id": "e6",
"status": "succeeded"... | {
"score": {
"mean_loss": 1.8454302549362183,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04031531699001789
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.8218445777893066,
"example_count": 60,... |
0e5d69fc-e58b-4f26-a1fa-d17361c445e1 | easy | viridale | 2026-08-13 07:52:45.041855+00:00 | succeeded | 5e6d0114baac23e7daef0c3cd2150994bbd0486ed799a5ae24cb78ed158d59e8 | 16,910 | null | """Gradient-balanced two-sided observable conjugacy.
hypothesis: v536 was mechanically finite but its causal probe measured observer
gradient 183.70 versus forward 0.98 and reverse 0.12, then loss rose 25.48 ->
32.85 in 24 updates. Under the evaluator's one global clip, the exact FoNE
boundary starves the ... | {
"id": "0e5d69fc-e58b-4f26-a1fa-d17361c445e1",
"created_at": "2026-08-13 07:52:45.041855+00:00",
"db_md5": "8a4a9afe95f1c28b358c591c34ebafaf",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "fa781ab8-42f9-4f42-b339-42c98194c356",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 17.96664333343506,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 18.663246154785156,
"example_count": 100,
"exac... |
0e5ea1d4-3124-4f34-abc1-5db3450205d0 | easy | chad-atexpedient | 2026-08-16 05:21:46.507880+00:00 | succeeded | c077f06491b4f6907abe5344cff3e1b1a978e3f551ea8fb65e013f7d3687eeb0 | 21,211 | null | """X55: Hybrid continuation-label + state consistency.
Combines X54's continuation-label supervision with X18's state consistency loss.
- Continuation: supervise A at step a+b with B's label z when A.output == B.input.
- Consistency: match T=2 intermediate state (step 1) with T=1 final state when
T=2.output == T=1.i... | {
"id": "0e5ea1d4-3124-4f34-abc1-5db3450205d0",
"created_at": "2026-08-16 05:21:46.507880+00:00",
"db_md5": "f3dce11b71372326d95dafe825fff938",
"submitter": "chad-atexpedient",
"github_login": "chad-atexpedient",
"run_id": "0ed6dc64-c9a5-4462-9f91-641000399821",
"tier": "easy",
"dataset_id": "e5",
"st... | {
"score": {
"mean_loss": 2.1719716100374136,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009583333333333334
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.176617054661293,
"example_count": 600... |
0e5eeded-d676-4667-aacf-47aee04d2a2e | easy | isaac0804 | 2026-08-06 16:10:31.920492+00:00 | succeeded | 3880e4271aa101cdc03b33be86b2afdeb120c3e6be4a0b4fb2b24170f2c323dd | 6,795 | null | """Internal register/scratchpad tokens: extra learnable positions concatenated
to the hidden-state sequence purely inside the model's own forward pass, never
touching the evaluator-owned `input_ids`/`attention_mask`/label layout.
Motivation (see docs/findings.md "weight-tied looped transformer" and
"SupCon contrastive... | {
"id": "0e5eeded-d676-4667-aacf-47aee04d2a2e",
"created_at": "2026-08-06 16:10:31.920492+00:00",
"db_md5": "c8a40b9fc6f922604c7343d3be295f5d",
"submitter": "Isaac Yong",
"github_login": "isaac0804",
"run_id": "b2942135-a3ee-420d-a537-78c8940dcb6d",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 5.121029615402222,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.01833333307877183
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.258674144744873,
"example_count": 100,
... |
0e6211c7-5e4e-47b5-b8a2-9e8b21e448af | easy | jordanrubin | 2026-08-02 17:41:57.772179+00:00 | succeeded | be6dbd0119ea82d5994215d1c87a7738a1c5ffc3c220da3d19edd9e9821cb856 | 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": "0e6211c7-5e4e-47b5-b8a2-9e8b21e448af",
"created_at": "2026-08-02 17:41:57.772179+00:00",
"db_md5": "5350863000bca2e6cd873ca226ca62cf",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "3671e6fe-4f05-488d-8487-26c70ec5ac96",
"tier": "easy",
"dataset_id": "e1",
"status": "s... | {
"score": {
"mean_loss": 2.3033451890947205,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.036666665847102806
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.38424542412829,
"example_count": 100,... |
0e64c516-e960-44af-be16-5cbce0906693 | easy | himalalps | 2026-08-07 01:20:23.007713+00:00 | succeeded | 13ae7aae52b8072fee6c5d817f21d5d3131bc20409dba0756684b451de490ae3 | 64,666 | 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": "0e64c516-e960-44af-be16-5cbce0906693",
"created_at": "2026-08-07 01:20:23.007713+00:00",
"db_md5": "4c34f8791cba5629553bd26f49f2d0e6",
"submitter": "Haoyu Tang",
"github_login": "himalalps",
"run_id": "0a60a8ba-5613-4fc8-ba97-26e9cf11f152",
"tier": "easy",
"dataset_id": "e5",
"status": "succe... | {
"score": {
"mean_loss": 5.751823769496368,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00541666670391957
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.535733699798584,
"example_count": 600,
... |
0e6bfa18-a21e-4d77-8045-007b10bf2b1f | easy | khushidahi | 2026-08-03 02:57:53.951182+00:00 | succeeded | 5623a476ea66d1524ca2481331e5d04e0a2bf564cd74701b88fed7f9b2c3e652 | 14,072 | 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": "0e6bfa18-a21e-4d77-8045-007b10bf2b1f",
"created_at": "2026-08-03 02:57:53.951182+00:00",
"db_md5": "f45a98ebab815a3c8b409f9bd41fecae",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "2ca67879-c6c0-49f5-a4a3-9f878c9604d4",
"tier": "easy",
"dataset_id": "e4",
"status": "succ... | {
"score": {
"mean_loss": 2.0795282125473022,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.006979166530072689
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.0770676136016846,
"example_count": 12... |
0e9ee27b-8e0d-47a1-af1e-c6c6d362f2bf | easy | shirvani-jr | 2026-08-29 00:58:07.392683+00:00 | succeeded | dbab70dcfb586df77ada7e5fbc12658b700d1cf70ad28351ad38816ec67ca5e2 | 48,764 | null | """Adaptive learned discrete recurrent fabric for One Layer Deeper.
A gradient-trained, operation-free recurrent computational machine. The
model learns, from the competition's endpoint supervision: how to parse the
prompt into a bank of categorical digit registers, a reusable transition
circuit applied serially, a l... | {
"id": "0e9ee27b-8e0d-47a1-af1e-c6c6d362f2bf",
"created_at": "2026-08-29 00:58:07.392683+00:00",
"db_md5": "d91712d41ecaa1db19f8bb5424f5a92f",
"submitter": "Ali",
"github_login": "shirvani-jr",
"run_id": "2c3526f2-0d98-4a45-923f-d1dc0f2f60e8",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.2587177753448486,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005000000004656613
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.256140947341919,
"example_count": 800... |
0ea1612f-5f9a-4dfe-93b2-52e9136dc7e9 | easy | erdavis0 | 2026-08-25 02:18:29.808025+00:00 | succeeded | 150d173c75338db2a7113c981ff2f114e97e4da3a7ca5424065633b1be6b8905 | 26,183 | 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": "0ea1612f-5f9a-4dfe-93b2-52e9136dc7e9",
"created_at": "2026-08-25 02:18:29.808025+00:00",
"db_md5": "fad6ce63bd264638f733d308596ed7c1",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "128940b5-4e3c-4189-bcd6-341b3d09bc3d",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded",... | {
"score": {
"mean_loss": 6.094151087293698,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.011250000012417634
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.141720151687417,
"example_count": 600,... |
0ea2a6fb-3043-4c1d-a6c3-f2b0b93bf6bb | easy | alirezashirvani-jr | 2026-08-29 05:39:03.084089+00:00 | succeeded | 52d3e8ea3655dd1b49603334ba1b505f43685cd8ec86c58dfe84c472ef64c8fc | 39,524 | null | """Adaptive learned discrete recurrent fabric for One Layer Deeper.
A gradient-trained, operation-free recurrent computational machine. The
public prompt grammar only routes literal digit tokens into canonical,
tail-aligned categorical registers. A reusable learned transition circuit
is applied serially, the public ... | {
"id": "0ea2a6fb-3043-4c1d-a6c3-f2b0b93bf6bb",
"created_at": "2026-08-29 05:39:03.084089+00:00",
"db_md5": "038694af6bb07b235d91146bab29234b",
"submitter": "alirezashirvani-jr",
"github_login": "alirezashirvani-jr",
"run_id": "9a54b0e9-8615-4b99-b635-dfec3e898bcb",
"tier": "easy",
"dataset_id": "e5",
... | {
"score": {
"mean_loss": 2.9616739254209907,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.04333333447575569
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.9954497814178467,
"example_count": 600... |
0ea9395c-ea09-4f3f-810e-51a7be0db3d3 | easy | arthurfeeney | 2026-08-07 06:54:55.128970+00:00 | succeeded | fbd305ad1f7d4b6763b424368a9e401ef1dbcb1b5e48f794ae4356200806a119 | 10,983 | null | r"""
Notes
1. goal is basically learn y = G(x_0, T, N).
G is always a recurrence `x_t+1 = f(x_t, t) mod N`.
Shouldn't be possible in general since f isn't determinable from a dataset...
I.e., multiple f can generate the same training dataset.
2. can't really use any info on structure of f...
"""
from __future... | {
"id": "0ea9395c-ea09-4f3f-810e-51a7be0db3d3",
"created_at": "2026-08-07 06:54:55.128970+00:00",
"db_md5": "ec99215dedf9a1ba3dd2e9adea91f980",
"submitter": "Arthur",
"github_login": "arthurfeeney",
"run_id": "80d3592e-58bd-4769-8c12-8a014017bd5e",
"tier": "easy",
"dataset_id": "e1",
"status": "succee... | {
"score": {
"mean_loss": 6.864308595657349,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.024999999441206455
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.8725457191467285,
"example_count": 100... |
0eaa0bf1-4c97-4ebd-9318-25691de4cac5 | easy | shirvani-jr | 2026-08-29 03:45:04.385372+00:00 | succeeded | 6344b15549efbfde2a8ad8981edb0209e96f0f6357a403ac08aeba5dede3f5fb | 48,708 | null | """Adaptive learned discrete recurrent fabric for One Layer Deeper.
A gradient-trained, operation-free recurrent computational machine. The
model learns, from the competition's endpoint supervision: how to parse the
prompt into a bank of categorical digit registers, a reusable transition
circuit applied serially, a l... | {
"id": "0eaa0bf1-4c97-4ebd-9318-25691de4cac5",
"created_at": "2026-08-29 03:45:04.385372+00:00",
"db_md5": "3d7f6fd312a2021ec8783a4a79cbf295",
"submitter": "Ali",
"github_login": "shirvani-jr",
"run_id": "2121f81e-e4e6-43e1-a363-3c5273187187",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 0.7930187582969666,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.8466666638851166
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.22954928874969482,
"example_count": 100... |
0eac0a90-bc71-48c8-8b75-5769c03be34a | easy | AFeinfield | 2026-08-19 23:26:58.989037+00:00 | succeeded | bec664ceff21c02f9ef50d16d4b06dd144235cd6140767c58eab8e7aeb2632d5 | 21,653 | null | """092: INSTRUMENTED BASE — two measurement fixes, no architecture change.
(1) Step-indexed LR schedule with a fixed screening step budget. The old
schedule decayed on WALL CLOCK, so throughput jitter changed the
LR-per-step trajectory: identical files on e2 gave 1470 vs 1025 steps
and train acc 0.725 vs 0... | {
"id": "0eac0a90-bc71-48c8-8b75-5769c03be34a",
"created_at": "2026-08-19 23:26:58.989037+00:00",
"db_md5": "f9ddcb1b28d32530a5a54b8d59a29bba",
"submitter": "AFeinfield",
"github_login": "AFeinfield",
"run_id": "210b97da-7bac-41cf-9207-4b22f26edaef",
"tier": "easy",
"dataset_id": "e2",
"status": "succ... | {
"score": {
"mean_loss": 2.2222890853881836,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.020416666753590107
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.218306303024292,
"example_count": 300... |
0eae6155-366d-478d-b859-d55aecfc6d60 | easy | khushidahi | 2026-08-03 05:35:20.302952+00:00 | succeeded | 32f2689391f266b51f8fbd6ac0f4cc47fea914fd89ee5c6716769db3944fd536 | 15,250 | null | """R6 structured decimal-workspace model for One Layer Deeper.
The model does not implement multiplication, modular reduction, or the public
recurrence. It uses the public tokenizer structure to build learned, right-
aligned decimal tapes for N, X, and T, then applies a shared neural transition
to a mutable work tape.... | {
"id": "0eae6155-366d-478d-b859-d55aecfc6d60",
"created_at": "2026-08-03 05:35:20.302952+00:00",
"db_md5": "6ddeb0cc0d466c4649cfa5bbd1dde031",
"submitter": "khushidahi",
"github_login": "khushidahi",
"run_id": "37e32b15-ce5b-4048-ac9e-ee7fe0cad2aa",
"tier": "easy",
"dataset_id": "e5",
"status": "succ... | {
"score": {
"mean_loss": 2.1977500915527344,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00791666703298688
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.193415641784668,
"example_count": 600,... |
0eb25e3f-c154-4dfc-94e8-16d2e7cb91eb | easy | KaustubhKumar05 | 2026-08-26 11:38:08.594546+00:00 | succeeded | 1e95b4e423fc3edc97a5a053c77bfe4420a708f7a9cd90f30985f79277743cb7 | 12,947 | null | """stationary — candidate C: force a behaviourally stationary operator.
E7: the accuracy cliff sits EXACTLY at max(train T) in every row of the sweep
(train to 4 -> dead at 8; train to 16 -> dead at 32). Training with more recurrence
only relocates the cliff. So depth is being memorised, not iterated.
E8: measured wh... | {
"id": "0eb25e3f-c154-4dfc-94e8-16d2e7cb91eb",
"created_at": "2026-08-26 11:38:08.594546+00:00",
"db_md5": "9bb03043f0ff3a20dbdbd26ba03faddd",
"submitter": "koz",
"github_login": "KaustubhKumar05",
"run_id": "541a19a5-0eee-4c4f-be74-e978122bfcc7",
"tier": "easy",
"dataset_id": "e5",
"status": "succee... | {
"score": {
"mean_loss": 6.505451732362642,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0066666666915019355
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.394349498064528,
"example_count": 600... |
0eb5633f-3f6a-4518-b4a0-95a8aa3a18eb | easy | MarcoF1 | 2026-08-14 18:31:39.581027+00:00 | succeeded | 3a54ec1185b960b719b3c8e326ab6a949a319d0a10119f7aacd6d29cc8885206 | 14,397 | null | """Residue-clock model for fixed-N generalization.
Representation: parsed integer values of x and N feed (a) exact residue-phase
features over a bank of small integer periods, (b) learned double-hashed value
embeddings (per-residue identity), and T feeds cyclic phase features
(T mod 2..48) plus saturating channels — p... | {
"id": "0eb5633f-3f6a-4518-b4a0-95a8aa3a18eb",
"created_at": "2026-08-14 18:31:39.581027+00:00",
"db_md5": "a9b51a846cde6d4e47e065988e59f349",
"submitter": "Marco Fleming",
"github_login": "MarcoF1",
"run_id": "798444b8-d872-4422-bd21-de9604d0935c",
"tier": "easy",
"dataset_id": "e2",
"status": "succ... | {
"score": {
"mean_loss": 2.5172633199445302,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.004375
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.486201893300268,
"example_count": 300,
... |
0ebbbee0-9bcc-4293-b6fb-97dd8376777f | easy | jordanrubin | 2026-08-10 18:48:24.784451+00:00 | succeeded | d66d3596d565e364e36b2e65e3e2d2e8d5bb226368b36c4f9dd3d714e86dfa54 | 19,332 | null | """Weight-tied MLP-cell loop over soft digit states (mlploop).
Cell = MLP over [expected digits, pairwise digit products, RNS residue
simplices (fixed differentiable mixing over Z_p), N digits]. The cell that
learns one-step modular squaring from direct pairs (52% unseen at 14k rows,
day-20 screen) inside the exact-T ... | {
"id": "0ebbbee0-9bcc-4293-b6fb-97dd8376777f",
"created_at": "2026-08-10 18:48:24.784451+00:00",
"db_md5": "881e1109fb87d3faf946c8d57e92a5d3",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "89f9721e-79db-44d9-8952-398cd96df4ec",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.9533371615354107,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007499999832361937
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1601416123822132,
"example_count": 60... |
0ebde377-7ccc-40e0-ab7d-4f35d987521f | easy | yashkant | 2026-08-19 02:18:08.115243+00:00 | succeeded | ad0e49905c7159b14bd0e0c3106de2476a1411c1fa65583bccbfbe81e739a7db | 46,509 | null | """Round2202 factorizes one tiny recurrent correction into residue and carry.
The measured correction-reset official carrier stays intact. A pooled
four-value/four-carry state runs one shared transition twice per parsed-T
cycle and reinjects its immutable input halves on every call. Carry is
learned from signed resi... | {
"id": "0ebde377-7ccc-40e0-ab7d-4f35d987521f",
"created_at": "2026-08-19 02:18:08.115243+00:00",
"db_md5": "d9b55e46056c0a257fcac86c624152c0",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "de5a4f3b-1b89-4b2a-a5f8-55fc5d359842",
"tier": "easy",
"dataset_id": "e5",
"status": "succeed... | {
"score": {
"mean_loss": 2.145979039984897,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009583333345750967
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.14909050924361,
"example_count": 600,
... |
0ec486af-a06b-4667-814c-9b3bba1f85d3 | easy | Susuguro | 2026-08-15 05:05:45.613962+00:00 | succeeded | 0bfb4a8175cf801ce0e52cb05fe4da140714082f556dcdb92e992cf0eb37c1e7 | 22,890 | null | """v18: canonical-digit-interface architecture with a strict ablation ladder.
Removals (evidence-based):
- FiLM-from-N-summary: the "learned constant" anti-pattern — any compression
of N into learned modulation is a per-modulus lookup and cannot survive
unseen N. N stays in slots as an operand.
- Downward scan: ne... | {
"id": "0ec486af-a06b-4667-814c-9b3bba1f85d3",
"created_at": "2026-08-15 05:05:45.613962+00:00",
"db_md5": "3b5aba680f48273ab38f2d614194550b",
"submitter": "Aqua",
"github_login": "Susuguro",
"run_id": "d5c8083f-d972-42e5-8ed0-6cc442ab03ea",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded",
... | {
"score": {
"mean_loss": 4.256998777389526,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0800000000745058
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.560972213745117,
"example_count": 100,
... |
0eccd538-61c2-409a-b1c6-0f4b0ec13e40 | easy | jordanrubin | 2026-08-11 19:18:40.022808+00:00 | succeeded | 0f93ed4d07f351beca91ef2e290a0c44323a16a8090f3dc0d883cf7155952d62 | 21,702 | null | """Weight-tied MLP-cell loop over soft digit states (mlploop).
Cell = MLP over [expected digits, pairwise digit products, RNS residue
simplices (fixed differentiable mixing over Z_p), N digits]. The cell that
learns one-step modular squaring from direct pairs (52% unseen at 14k rows,
day-20 screen) inside the exact-T ... | {
"id": "0eccd538-61c2-409a-b1c6-0f4b0ec13e40",
"created_at": "2026-08-11 19:18:40.022808+00:00",
"db_md5": "318216f9e8e3f7335c8009581f1e16cc",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "8f650258-aa19-4b0b-b072-7e2a56a67629",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.4795435428350294,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.007083333175008495
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1565830588608046,
"example_count": 60... |
0ed0ae33-b537-43d2-b65d-1bdacb4eb2f2 | easy | yashkant | 2026-08-19 01:43:42.816467+00:00 | succeeded | 862146a1e9acbe92c8744760279022cd13aebb4d9dfb3fb9bb96711be0e3c16d | 41,871 | null | """Round2223 two-independent-branch low-capacity ensemble.
This source keeps the complete Round1829 nonlinear branch architecture,
SAM/Compositional-Muon optimizer, EMA, loss, and evaluator contract. It
instantiates two independently initialized width-288 branches and always
returns their symmetric arithmetic-mean lo... | {
"id": "0ed0ae33-b537-43d2-b65d-1bdacb4eb2f2",
"created_at": "2026-08-19 01:43:42.816467+00:00",
"db_md5": "277a40ae4c484e54a3b1e41decf19bf8",
"submitter": "Yash Kant",
"github_login": "yashkant",
"run_id": "aa1bb0c0-b26e-4e71-991f-392857d62cc2",
"tier": "easy",
"dataset_id": "e3",
"status": "succeed... | {
"score": {
"mean_loss": 2.127884109027957,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.01500000006519258
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1276747098127533,
"example_count": 800,... |
0ed23134-3cb2-4ca1-ad04-4faba6d1640a | easy | oupadhyay | 2026-08-22 17:25:20.652153+00:00 | succeeded | fff2a60223927b53360269d84500c1e70f6d549920d4599ea9b8a3d88078c7a1 | 9,191 | null | """D64 one-sweep chunkwise DeltaNet transition for the Easy benchmark."""
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
PAD, N_MARK, X_MARK, T_MAR... | {
"id": "0ed23134-3cb2-4ca1-ad04-4faba6d1640a",
"created_at": "2026-08-22 17:25:20.652153+00:00",
"db_md5": "bffe57f85c49622e4f2a91bf8b13c5a6",
"submitter": "Ojasw Upadhyay",
"github_login": "oupadhyay",
"run_id": "03abe6a9-e375-4515-ba70-de3b7e068e24",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 2.309630301897146,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0033333333457509677
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2227704642599475,
"example_count": 60... |
0ee07a2f-9923-46e0-bb61-f47b583f842b | easy | chad-atexpedient | 2026-08-15 22:44:26.328110+00:00 | failed | 90043a1dbf9a2ed80d83c687aa9d14f885c8fee8e4b86af798f704133e42bb80 | 21,606 | null | """X52: Continuation-label supervision.
Uses supplied endpoint-to-input chains to create valid longer-horizon final-label loss.
For rows A: (N, x, T=a) -> y and B: (N, y, T=b) -> z, the valid relation is:
f^(a+b)(x) = f^b(y) = z
Continue A's recurrent rollout to step a+b, decode, and supervise with B's label z.
Uses ... | {
"id": "0ee07a2f-9923-46e0-bb61-f47b583f842b",
"created_at": "2026-08-15 22:44:26.328110+00:00",
"db_md5": "dc90114ee0c0203141cb4f195bc99774",
"submitter": "chad-atexpedient",
"github_login": "chad-atexpedient",
"run_id": "c794c744-fc44-43a7-803e-0f1cc1df78e7",
"tier": "easy",
"dataset_id": "e5",
"st... | null
|
0ee91aa3-4a4e-4fa6-8c92-2ff2e992558b | easy | sankalp1999 | 2026-08-04 17:12:03.368522+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": "0ee91aa3-4a4e-4fa6-8c92-2ff2e992558b",
"created_at": "2026-08-04 17:12:03.368522+00:00",
"db_md5": "1d834f16da97d60fe96aa31d6bab0bff",
"submitter": "Sankalp ",
"github_login": "sankalp1999",
"run_id": "193cb127-04ba-44a0-a9e7-948c70d2aef4",
"tier": "easy",
"dataset_id": "e3",
"status": "succe... | {
"score": {
"mean_loss": 0.0,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.0,
"example_count": 800,
"exact_accuracy": 1.0,
"... |
0eeae409-3f81-4847-b528-ebba48925b43 | easy | jordanrubin | 2026-08-01 20:42:08.987226+00:00 | succeeded | 2c6adbc91db77091eaa1fdb53b2b15fdc6e05306608a2440be5e6a2aa2591b8a | 18,183 | null | """Looped bidirectional selective-SSM (Mamba-style), pure PyTorch.
The statefulness bet: input-dependent SSM dynamics along the token axis,
with the block weight-tied and applied NUM_LOOPS times in depth with input
injection — recurrence in depth is where the task's T-iteration lives.
The selective scan is computed i... | {
"id": "0eeae409-3f81-4847-b528-ebba48925b43",
"created_at": "2026-08-01 20:42:08.987226+00:00",
"db_md5": "35fec302dbad6ab1de27b5092ede0b6a",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "0d8b3da4-5953-4b83-9b85-c1ef0d389d57",
"tier": "easy",
"dataset_id": "e1",
"status": "s... | {
"score": {
"mean_loss": 5.033940723114775,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05166666644314925
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.163424968719482,
"example_count": 100,
... |
0eee4a80-b2d6-4ccf-88e1-f22c0425d7b9 | easy | erdavis0 | 2026-08-29 01:30:01.424683+00:00 | succeeded | c7f78ef5748d23ee6c570a69dda0f9e2f1d45cb5b22c279e1ef01c84faeb2f27 | 17,575 | null | """Sealed N-conditioned circulant operator over an X-initialized latent tape.
Public delimiters only route opaque categories into independent right-aligned
N, X, and T tapes. N alone compiles a small bank of learned cyclic place
kernels and a continuous program vector; X alone initializes the recurrent
state; T is co... | {
"id": "0eee4a80-b2d6-4ccf-88e1-f22c0425d7b9",
"created_at": "2026-08-29 01:30:01.424683+00:00",
"db_md5": "c929d07e2f2fc815762284c89335e51b",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "c740befd-12ae-4881-91a5-46b56e003695",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded",... | {
"score": {
"mean_loss": 9.43320462590237,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.01750000001396984
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.64528476103645,
"example_count": 800,
... |
0eee809a-3455-4bde-91f3-d49174057367 | easy | viridale | 2026-08-04 22:30:50.993512+00:00 | succeeded | a020fb7e222c75c95bf040cd0838c3b6fca09229dd5046368013ac0e8163ce95 | 7,140 | null | """Clean small-init recurrence with AdamW learning rate 3e-3.
hypothesis: v176 improved E1 held-out loss and exactness but left sampled train exact
near zero at lr1e-3; a threefold LR should escape underfit while uniform AdamW
decay retains more transfer than v174's memorizing Muon/AdamW split.
axis: opt
targe... | {
"id": "0eee809a-3455-4bde-91f3-d49174057367",
"created_at": "2026-08-04 22:30:50.993512+00:00",
"db_md5": "391c2883a3b342cfe8c9cce2823d6fa2",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "b257c074-4a14-4863-8ea0-22e27251f19a",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 2.1813528537750244,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0833333320915699
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1975598335266113,
"example_count": 100,... |
0ef086ed-e317-4f4f-bc37-faf29d563aca | easy | paulpister | 2026-08-05 07:49:18.494175+00:00 | succeeded | cf007729ae18722ebe2ab1868ca7a408c25b0263b791259fc9c5f8951d4d7b16 | 22,563 | null | """Loop-T tied recurrence v20: v19 with phase-gated EMA (harvest window only).
Design:
- Structural features computed from input_ids (field segment, digit place
value from the little end, reverse position from sequence end). Sinusoidal
place/revpos features so longer OOD-N numbers extrapolate smoothly.
- T is pars... | {
"id": "0ef086ed-e317-4f4f-bc37-faf29d563aca",
"created_at": "2026-08-05 07:49:18.494175+00:00",
"db_md5": "d157afe2736d490746f2cc0cf1a61ad9",
"submitter": "paulv",
"github_login": "paulpister",
"run_id": "bbbeec36-6d22-48a7-b2f5-f01c9c7a629b",
"tier": "easy",
"dataset_id": "e2",
"status": "succeeded... | {
"score": {
"mean_loss": 6.499026536941528,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.26270834915339947
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 10.996399879455566,
"example_count": 300,... |
0ef83922-8194-4cf1-9ae2-6443f0295275 | easy | viridale | 2026-08-07 05:31:48.380790+00:00 | succeeded | 5f4614bfc6d30c7cd5751911b32a9e908413c52924e58e89c2459ee28eda594f | 16,630 | null | """Additive CoFoNE sidecar with frequency-preserving learned interaction.
hypothesis: scalar-value FoNE and exact unary bits tied the clean basin because neither
exposes interaction among numeral coefficients; a DFT of each decimal coefficient
vector followed by learned same-frequency cross-run communication s... | {
"id": "0ef83922-8194-4cf1-9ae2-6443f0295275",
"created_at": "2026-08-07 05:31:48.380790+00:00",
"db_md5": "24f8907742f88e4cae91f158e2454244",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "7a86c142-9025-4667-b0a2-1668460196c4",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 2.1858965158462524,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0833333320915699
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.207343339920044,
"example_count": 100,
... |
0f152225-ec4a-4d94-8f34-b695a0df467d | easy | chuk-uzowihe | 2026-08-12 19:32:54.785706+00:00 | succeeded | be4aeeaebdefd6a2d80b99bd083c521d3ea7462084a53e9d4f1fc84787856f32 | 25,826 | null | """v10: dense fixed-loop recurrence with deep supervision.
Lineage: FPRM structure (alpha/beta-tied residual scaling, pre-rms_norm,
input injection every iteration) + the v8 2D digit lattice / RoPE task
encoding. Three deliberate departures from the v9 MoEUT line:
1. **Full BPTT.** The loop runs a fixed N_ITER iterat... | {
"id": "0f152225-ec4a-4d94-8f34-b695a0df467d",
"created_at": "2026-08-12 19:32:54.785706+00:00",
"db_md5": "bbdcb493fc80484f768b14839e663ad5",
"submitter": "Chuk Uzowihe",
"github_login": "chuk-uzowihe",
"run_id": "6a5b9508-fe28-4573-97b2-8239dd6fa0ba",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 9.390588760375977,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.038333332166075706
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 7.0962371826171875,
"example_count": 100... |
0f1aa6e3-eee1-43a3-9f07-5eff64ee1ebf | easy | viridale | 2026-08-09 02:18:04.680839+00:00 | succeeded | a47b1774684407e36cf8a291ae371db4be4df1ddd14e057c50b3e1db3f5f9f1d | 15,778 | null | """Depth-headed endpoint-collision relation with one-step diffusion.
hypothesis: v378's shared collision head reached 45.17%, while restricting its labels
to literal T=1 fell to 35.00%. Deep collision data therefore supplies useful
representation learning, but F^t(x)=F^t(y) does not imply F(x)=F(y). Share th... | {
"id": "0f1aa6e3-eee1-43a3-9f07-5eff64ee1ebf",
"created_at": "2026-08-09 02:18:04.680839+00:00",
"db_md5": "5d25d6d95367d6c3ccd11125facf80b0",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "d433eb9c-61bf-4162-901b-18d9926cf458",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 2.1898869276046753,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.2750000199303031
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 3.4081151485443115,
"example_count": 100,... |
0f24dbc1-78b5-4ec8-a67c-74ff638ce467 | easy | DDanlov | 2026-08-22 21:52:00.798436+00:00 | succeeded | c41b9bc88897264536a66e57936a3fd4f77e28f7b0c0e44c19a4654014c30bcd | 30,754 | null | """
Unified Hybrid Masked Deliberation Model (85.3M Parameters) - Variant 1 (With Cross-Attention to z_0)
Architecture:
- Scale: dim=1536, num_heads=24, d_ff=6144 (Exactly 85,263,360 parameters)
- Hybrid Prefix-Bidirectional + Causal Target Attention Mask
- Token Layout: [Prompt (L_p), 4 Pause Registers, Target Tokens ... | {
"id": "0f24dbc1-78b5-4ec8-a67c-74ff638ce467",
"created_at": "2026-08-22 21:52:00.798436+00:00",
"db_md5": "5bb097a711b93738e0e9dfb8c092bcb2",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "fcb25c57-3813-49c9-92ae-b5e416f222ab",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.837240037494275,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.0004166666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.8669887433672163,
"example_count": 60... |
0f270c35-9eaf-428e-9df3-1f80ac020e11 | easy | 0Chris5R | 2026-08-07 12:21:08.880744+00:00 | succeeded | faa26a9c621462d4ea392cb9244b0999db90b0c5b096d2b6d626f47c66237c0d | 20,840 | null | """Shared-digit Neural GPU with a compute-matched operator curriculum.
The model aligns decimal digits by significance, forms a learned low-rank
second-order feature map of the current state, and evolves a two-dimensional
workspace with the original convolutional-GRU equations. Each arithmetic
transition applies the ... | {
"id": "0f270c35-9eaf-428e-9df3-1f80ac020e11",
"created_at": "2026-08-07 12:21:08.880744+00:00",
"db_md5": "7b058eeca7f4510c2e71786a84d51c84",
"submitter": "Chris ",
"github_login": "0Chris5R",
"run_id": "a21da273-f6ec-4e0d-b9d8-f8d903ff20f6",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.678040168598365,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03291666666666667
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.6643694387957653,
"example_count": 600,... |
0f2aba10-0c10-4e06-84c7-ad33873400de | easy | jordanrubin | 2026-08-01 20:10:34.071664+00:00 | succeeded | 2c6adbc91db77091eaa1fdb53b2b15fdc6e05306608a2440be5e6a2aa2591b8a | 18,183 | null | """Looped bidirectional selective-SSM (Mamba-style), pure PyTorch.
The statefulness bet: input-dependent SSM dynamics along the token axis,
with the block weight-tied and applied NUM_LOOPS times in depth with input
injection — recurrence in depth is where the task's T-iteration lives.
The selective scan is computed i... | {
"id": "0f2aba10-0c10-4e06-84c7-ad33873400de",
"created_at": "2026-08-01 20:10:34.071664+00:00",
"db_md5": "35fec302dbad6ab1de27b5092ede0b6a",
"submitter": "Jordan Rubin",
"github_login": "jordanrubin",
"run_id": "fc902ddb-2674-4034-a8b6-c2959780b11a",
"tier": "easy",
"dataset_id": "e5",
"status": "s... | {
"score": {
"mean_loss": 4.222771700220909,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005416666545594732
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 4.296317539674819,
"example_count": 600,... |
0f2bdb67-01c2-4d29-848c-9f1acd216a34 | easy | DDanlov | 2026-08-20 22:43:00.400794+00:00 | succeeded | 23aeeadcab7cf58b2b2dd9f07fc94cdaeacd8552e13289865e63e4fe7688c74b | 21,253 | 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": "0f2bdb67-01c2-4d29-848c-9f1acd216a34",
"created_at": "2026-08-20 22:43:00.400794+00:00",
"db_md5": "0c712a7c33c88825b4be8d911194532d",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "e45808b1-6145-4b29-ba62-d245d2b4fa7a",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.6953452826908277,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.002083333358168602
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.6956395945206886,
"example_count": 60... |
0f3822a2-2078-47bd-9180-418abf15bb37 | easy | ArkinDharawat | 2026-08-13 16:50:01.523881+00:00 | succeeded | 7e10238e818cf02e6aec379e12647ead4b59719541c9eff4832bbfe188507377 | 7,630 | 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": "0f3822a2-2078-47bd-9180-418abf15bb37",
"created_at": "2026-08-13 16:50:01.523881+00:00",
"db_md5": "09f5f1a01c8bd0b978803585267d8688",
"submitter": "Arkin Dharawat",
"github_login": "ArkinDharawat",
"run_id": "7cbb9f34-02dc-4083-8c38-39860abc8b29",
"tier": "easy",
"dataset_id": "e3",
"status"... | {
"score": {
"mean_loss": 2.280804690372701,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.018125000046566128
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.327407026972949,
"example_count": 800,... |
0f450426-6681-4317-bc59-a7ba849a084c | easy | viridale | 2026-08-03 08:59:32.537352+00:00 | succeeded | d6feb67d395c0959fc188fd0ba7f0fb6265ef7372aa49c806c4ade20e3747c8e | 133,578 | null | """Dataset-global source evidence with an alias-safe boundary latch.
hypothesis: v131's exact source support is broad enough, but pruning all
364,608 programs to 512 after one minibatch can discard the globally
correct deep alias or a distinct complete T=1 boundary class.
axis: opt
target: revised h1 Max T at ... | {
"id": "0f450426-6681-4317-bc59-a7ba849a084c",
"created_at": "2026-08-03 08:59:32.537352+00:00",
"db_md5": "5d3ba980c5fafb337e39ded276fe084a",
"submitter": "priormancer",
"github_login": "viridale",
"run_id": "47b319cd-f3ec-447e-8682-6094a8c032cb",
"tier": "easy",
"dataset_id": "e4",
"status": "succe... | {
"score": {
"mean_loss": 0.0030146016029730695,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 1.0
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 0.0030146030333625382,
"example_count": 1200,
... |
0f4de5c3-5c89-42bb-a0a5-bc4ed7973698 | easy | DDanlov | 2026-08-16 00:35:24.592764+00:00 | succeeded | 1b27896e181807108dffcd8b192c6208dbf3d1c5c075d8c09a21c9bb644cbb23 | 16,289 | null | """
Production Port-Hamiltonian RoPE Dynamic Submission
Config ID: 5 | Sweep: update_freq=5 | lr=0.006, wd=0.03
"""
from __future__ import annotations
import math
from typing import Optional, Tuple
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim import Optimizer
try:
from bench... | {
"id": "0f4de5c3-5c89-42bb-a0a5-bc4ed7973698",
"created_at": "2026-08-16 00:35:24.592764+00:00",
"db_md5": "9ced91f731b10babe9f2266f52f073af",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "f635c025-a2aa-46d5-a540-3a35767380f9",
"tier": "easy",
"dataset_id": "e5",
"status": "succeeded"... | {
"score": {
"mean_loss": 2.184695644165171,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.009583333370586235
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2053861564584913,
"example_count": 600... |
0f53de5e-f1ed-4dd5-9c5b-c2be43bc29ed | easy | erdavis0 | 2026-08-19 07:58:37.860732+00:00 | succeeded | e6d75591b123483c27b5a4344aa991e325184d79f5ed87c2a9fa7d7390f5f5bd | 10,552 | null | """Generic learned-slot transducer with a shared low-rank latent operator."""
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,
TokenL... | {
"id": "0f53de5e-f1ed-4dd5-9c5b-c2be43bc29ed",
"created_at": "2026-08-19 07:58:37.860732+00:00",
"db_md5": "fb39c040e0caf73a4ee6e4c2f9f4d564",
"submitter": "Ethan",
"github_login": "erdavis0",
"run_id": "a544e69c-1065-47d5-b20f-90c0e624b926",
"tier": "easy",
"dataset_id": "e3",
"status": "succeeded",... | {
"score": {
"mean_loss": 2.1465449031117876,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.005
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.1435460883827835,
"example_count": 800,
"e... |
0f5d983a-2da0-4ae5-84bb-37456ae36d1d | easy | epicosy | 2026-08-26 21:03:55.329459+00:00 | succeeded | ace2f1895b8e10d235e6e5d2b879cf42f08ac8becaa36d40dd1aefb58b3d8e14 | 5,430 | null | """Latent-state Universal Transformer baseline for the Easy tier.
The prompt stream is encoded once and kept fixed. A single shared block
recurrently updates learned latent/query tokens through cross-attention to the
prompt. The final logits remain aligned with the original input sequence, as
required by the public su... | {
"id": "0f5d983a-2da0-4ae5-84bb-37456ae36d1d",
"created_at": "2026-08-26 21:03:55.329459+00:00",
"db_md5": "8678e29543bec659c082834b004aacdb",
"submitter": "Eduard Pinconschi",
"github_login": "epicosy",
"run_id": "1174ca31-3f4f-4a86-a09a-c1acf52d685a",
"tier": "easy",
"dataset_id": "e6",
"status": "... | {
"score": {
"mean_loss": 2.2395740747451782,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.02342342445626855
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 2.2187247276306152,
"example_count": 60,... |
0f67b934-47a4-4a7e-b2fa-b02e3d4ec668 | easy | chad-atexpedient | 2026-08-04 20:34:39.071386+00:00 | succeeded | fa3102d5530848f410dafdaacb8c0cd190610036c3cd632248ea2a800fc9ec48 | 5,493 | null | """R003: R001 plus learned decimal place from each field's right edge."""
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 = 128... | {
"id": "0f67b934-47a4-4a7e-b2fa-b02e3d4ec668",
"created_at": "2026-08-04 20:34:39.071386+00:00",
"db_md5": "c21d2f15917924619ec5b3c46b2579eb",
"submitter": "chad-atexpedient",
"github_login": "chad-atexpedient",
"run_id": "a3c14db4-c3ce-4812-b8d8-ef0d5f943b43",
"tier": "easy",
"dataset_id": "e5",
"st... | {
"score": {
"mean_loss": 9.257354970364789,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.00625
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 9.352877595499493,
"example_count": 600,
"e... |
0f6ada18-ec64-4e53-8a2e-2f8c4b582e20 | easy | DDanlov | 2026-08-10 07:55:29.480800+00:00 | succeeded | 43b7b166f0eb6e2d4a842db69e77ed373a432e838ab83465e11cd5561fabf9a6 | 16,672 | null | from __future__ import annotations
import sys
import os
import math
import time
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.optim.optimizer import Optimizer
VOCAB_SIZE = 17
DIGIT_OFFSET = 7
def init_cutting_the_skip(module):
"""
SVD-based initialization to ensure initial Jaco... | {
"id": "0f6ada18-ec64-4e53-8a2e-2f8c4b582e20",
"created_at": "2026-08-10 07:55:29.480800+00:00",
"db_md5": "a79713782fef269278196617143c292e",
"submitter": "DDanlov",
"github_login": "DDanlov",
"run_id": "fdadb3f9-31d5-4b3d-9ae4-d373be2fd07d",
"tier": "easy",
"dataset_id": "e1",
"status": "succeeded"... | {
"score": {
"mean_loss": 4.85162752227593,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.05666666368643443
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 5.153428554534912,
"example_count": 100,
... |
0f706388-fa35-473d-ab11-3f8db1ffb6e8 | easy | isaac0804 | 2026-08-07 20:51:56.234087+00:00 | succeeded | bdbd580f3ae050d3ac166990dc8fa4cd2132eed5ff6188c4585eff65d6156ace | 7,761 | null | """Clean variable-depth loop + gated linear layers.
Same base as looped_clean (shared weight-tied block applied T times, NO damped
gate, NO step-embedding -- the two proven single-step learning blockers). The
difference is per-linear-layer gating, motivated by the observation that the
register's soft-write (`softmax(l... | {
"id": "0f706388-fa35-473d-ab11-3f8db1ffb6e8",
"created_at": "2026-08-07 20:51:56.234087+00:00",
"db_md5": "55d34f2a3ae20fb670515f52735b256c",
"submitter": "Isaac Yong",
"github_login": "isaac0804",
"run_id": "6add9fe4-ec29-4e1f-9f10-6d26dffb418b",
"tier": "easy",
"dataset_id": "e1",
"status": "succe... | {
"score": {
"mean_loss": 1.961999773979187,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03333333320915699
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 1.8690788745880127,
"example_count": 100,... |
0f73ca9b-c9e6-4d86-b7a3-6e8e57abf6c0 | easy | Ranoobaba | 2026-08-17 05:43:19.309231+00:00 | succeeded | bdc112fb41da8d2fa65899f2becbcddaa7257461cf25beee33824255d0a99a93 | 5,766 | null | """Looped transformer, iteration count conditioned on T read from the prompt.
Core idea: learn one latent "squaring step" operator and apply it T times.
Depth generalization comes from the loop structure itself; the model never
needs more parameters for deeper problems, only more iterations.
"""
from __future__ impor... | {
"id": "0f73ca9b-c9e6-4d86-b7a3-6e8e57abf6c0",
"created_at": "2026-08-17 05:43:19.309231+00:00",
"db_md5": "0980796e6ce2967cef929380f4d94ee4",
"submitter": "Syed Rayyan Ali",
"github_login": "Ranoobaba",
"run_id": "601cb5cb-2e88-4983-b5bc-cd9ba1389084",
"tier": "easy",
"dataset_id": "e1",
"status": "... | {
"score": {
"mean_loss": 6.326666241572088,
"primary_metric": "mean_exact_accuracy",
"num_measurements": 2,
"mean_exact_accuracy": 0.03166666517655055
},
"seeds": [
{
"seed": 74,
"evaluation": {
"ood": {
"loss": 6.384068012237549,
"example_count": 100,
... |
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.