List your company

Long-range Sequence Modeling on Long Range Arena (LRA)

90.53Text AccuracyMEGALODON
63.115670.232877.3584.4672Mar 1, 2021Dec 31, 2021Nov 1, 2022Sep 3, 2023Jul 4, 2024May 5, 2025Mar 7, 2026

Evaluation results

MethodLinks
MEGALODONMode=chunk2024.0490.5387.1196.8997.2162.2391.7487.62—
MEGALODONMode=full2024.0490.4889.4298.1398.1763.7991.7688.63—
Mega# parameters=~600k2023.0690.4390.4496.0197.9863.1491.25——
MEGAMode=full2024.0490.4390.4496.0197.9863.1491.2588.21—
Mega2025.1290.4390.4496.0197.98—91.2593.22—
Mega2023.1190.4390.4496.0197.9863.1491.2588.21—
Mega + SMRMode=Convolution, SMR=true2024.0590.3688.2193.78—59.4590.6484.49—
MEGAMode=chunk2024.0490.1985.894.4193.8158.7690.9785.66—
MegaMode=Convolution, SMR=false2024.0589.9786.8293.4—57.6790.1783.61—
LRU2025.1289.48995.194.2—89.991.52—
LRU2023.1189.48995.194.260.289.986.3—
S52025.1289.318895.3398.56—91.492.52—
S52023.1189.318895.3398.5662.1591.487.46—
SGConv2025.1289.287.9795.4697.83—91.1192.31—
SGConv2023.1189.287.9795.4697.8361.4591.1187.17—
S4 + SMRMode=Convolution, SMR=true2024.0589.0988.9789.01—59.0189.2883.07—
HGRN2023.1188.1488.6992.9297.559.9594.2386.91—
TNN2025.1287.988.249396.1—90.9791.24—
TNN2023.1187.988.249396.161.0490.9786.21—
CDIL-CNNChannels=642022.0187.6164.4991——84.27——
CNNChannels=642022.0187.2935.8555.95——83.33——
MS-SSMBackbone=S42025.1287.2289.1594.997.12—91.0691.89—
S5 + SMRMode=Recurrence, SMR=true2024.0587.0889.3188.05—59.0889.3782.58—
SPADE + SMRMode=Convolution, SMR=true2024.0587.0689.2993.34—59.5289.0183.64—
Deformable2022.0186.9136.5756.14——83.69——
S42025.1286.8288.6594.296.35—90.991.38—
S42023.1186.8288.6594.296.3559.690.986.09—
S4Streaming?=No2026.0286.8288.6594.2—59.690.986.09—
GRUHidden size=128, Layers=12022.0186.744.485.45——77.08——
S42024.0486.5388.4894.0196.0759.190.9485.86—
S4Mode=Convolution, SMR=false2024.0586.4787.285.99—57.0686.7480.69—
SPADEMode=Convolution, SMR=false2024.0586.2988.0592.77—58.7588.6282.9—
LSTMHidden size=128, Layers=12022.0185.832.9961.26——77.18——
MS-SSMBackbone=S62025.1285.789.8387.2487.7—83.2186.73—
DSS softmax2025.1284.885.784.687.8—87.886.13—
DSS_softmax2023.1184.885.784.687.860.687.881.88—
DSSEXP2025.1284.684.984.785.6—87.685.47—
DSSEXP2023.1184.684.984.785.659.787.681.18—
S5Mode=Recurrence, SMR=false2024.0584.287.5187.42—58.2587.9981.07—
S6 + SMRMode=Recurrence, SMR=true2024.0584.1483.2385.8—56.1583.6678.6—
S6Mode=Recurrence, SMR=false2024.0583.5282.9685.54—55.6283.2878.18—
Mamba2025.1282.9869.8269.2667.32—72.1472.3—
DSSEXPScaling=No2025.1282.481.281.3——8666.46—
DSSEXP-NO-SCALE2023.1182.481.281.3—59.38665.03—
ELM Neuron# parameters=~100k-200k2023.0680.349.6271.1577.2946.7784.93——
H-Transformer-1D2021.0778.6946.0568.78—49.5363.9961.41—
S4# parameters=~600k2023.0676.0287.2686.0588.158.3587.09——
Chr.-LSTM# parameters=~100k-200k2023.0675.446.0970.79—44.5582.87——
LUNA2025.1273.4164.3269.52—38.9481.0265.44—
PolaFormeralpha=32025.1273.0642.1570.53—37.3580.560.72—
Griffin2025.1271.7561.1573.3869.53—66.5868.47—
LOTFormer2025.1271.154.169.9—38.580.962.9—
Transformer-LSReference=[54]2022.0168.4————81.95——
cosFormer2025.1267.751.2371.96——83.1568.51—
cosFormer2023.1167.751.2371.96—36.583.1551.76—
Transformer-LSMode=None, SMR=false2024.0566.6269.9847.6—40.381.6861.24—
RFA-GaussianReference=[25]2022.0166————56.1——
Linear Trans.2021.0765.942.3475.3—16.1353.0950.55—
Linear Trans.2022.0265.942.3475.3—16.1353.0950.55—
Linear TransformerReference=[53]2022.0165.942.3475.3——53.09——
Linear Trans.2022.0265.942.3475.3—16.1353.0950.55—
LinearTrans.2025.1265.942.3475.3—16.1353.0950.55—
Linear Trans.Streaming?=No2026.0265.942.3475.3—16.1353.0950.46—
Luna-2562024.0465.7847.8678.55—37.9879.5661.95—
NyströmformerReference=[54]2022.0165.75————81.29——
OmniNetAttention Kernel=Performer, Configuration Group=LRA Baselines2021.0365.6———18.260.948.2—
NyströmformerReference=[55]2022.0165.5241.5870.94——79.56——
NyströmformerStreaming?=No2026.0265.5241.5870.94—37.1579.5657.46—
Performer2021.0365.4———1853.845.7—
Performer2021.0765.442.7777.05—18.0153.8251.41—
Performer2022.0265.442.7777.05—18.0153.8251.41—
PerformerReference=[53]2022.0165.442.7777.05——53.82——
Performer2022.0265.442.7777.05—18.0153.8251.41—
Performer2025.1265.442.7777.05—18.0153.8251.41—
PerformerStreaming?=No2026.0265.442.7777.05—18.0153.8251.18—
TransformerReference=[54]2022.0165.35————82.3——
PerformerReference=[54]2022.0165.21————81.7——
Transformer2024.0465.2142.9471.83—37.1179.1459.24—
FNetStreaming?=No2026.0265.1138.6777.8—35.3359.6154.42—
OmniNetAttention Kernel=Performer, Configuration Group=Transformer Comparison2021.0365.1———37.258.853.7—
TransformerReference=[55]2022.0165.0238.274.16——79.35——
ReformerReference=[55]2022.0164.8843.2969.36——78.64——
ReformerReference=[54]2022.0164.88————78.64——
Nyströmformer (Spec. cond.)Architecture=Spectral conditioned attention2026.0364.840.273.7—37.880.6——
Skyformer2025.1264.740.7770.73—38.6982.0659.39—
Hedgehog2025.1264.640.1574.16—37.1582.2459.66—
Luna-256Streaming?=No2026.0264.5747.3877.72—37.2579.2959.37—
ScatterbrainNumber of parameters=1/8 of full2021.1064.5543.6569.91—38.680.2259.38—
ScatterbrainParameter Budget=1/8 of full2021.1064.5543.6569.91—38.680.2259.38—
Orthoformer-64GFLOPS=0.24, Memory (MB)=3442021.0664.4233.2666.41—33.8778.3655.26—
suGRU(binary new)Streaming?=Yes2026.0264.4280.3284.92—44.9377.5470.43—
Flowformer2022.0264.2943.273.95—38.762.2456.48—
Transformer# parameters=~600k2023.0664.2742.4471.4—36.3757.46——
Transformer2021.0764.2742.4471.4—36.3757.4654.39—
Transformer2022.0264.2742.4471.4—36.3757.4654.39—
TransformerReference=[53]2022.0164.2742.4471.4——57.46——
Transformer2022.0264.2742.4471.4—36.3757.4654.39—
TransformerStreaming?=No2026.0264.2742.4471.4—36.3757.4653.66—
Nyströmformer-128GFLOPS=0.62, Memory (MB)=7452021.0664.1736.1652.32—36.978.6753.64—
suGRU(sigmoid)Streaming?=Yes2026.0264.1756.9273.45—39.4563.2359.44—