List your company

Long-context Understanding on RULER (8K to 256K Contexts)

92.88Performance (8K Context)Llama-3.1-8B-Instruct
73.463278.504183.54588.5859Apr 8, 2026

Evaluation results

MethodLinks
Llama-3.1-8B-InstructBackbone=Llama-3.1-8B-Instruct, Attention Mechanism=Full Attention2026.0492.8892.8389.4670.7980.1272.3483.47
FluxAttn (FA-TA)Backbone=Llama-3.1-8B-Instruct, Attention Mechanism=FluxAttn (FA-TA)2026.0492.7290.5386.4567.7880.6367.0981.5
TriangleMixBackbone=Llama-3.1-8B-Instruct, Attention Mechanism=TriangleMix2026.0492.4490.7686.756878.2564.3980.46
FluxAttn (FA-XA)Backbone=Llama-3.1-8B-Instruct, Attention Mechanism=FluxAttn (FA-XA)2026.0492.4390.8588.2368.5675.8660.879.51
DuoAttentionBackbone=Llama-3.1-8B-Instruct, Attention Mechanism=DuoAttention2026.0491.7186.3585.6562.6562.338.6970.33
FluxAttn (FA-SSA)Backbone=Llama-3.1-8B-Instruct, Attention Mechanism=FluxAttn (FA-SSA), Configuration=Variant 22026.0490.1179.3979.2256.0862.9459.3973.67
Qwen3-8BBackbone=Qwen3-8B, Attention Mechanism=Full Attention2026.0489.6985.6263.2382.3965.8466.7175.74
Qwen3-4BBackbone=Qwen3-4B, Attention Mechanism=Full Attention2026.0487.4986.8260.0570.9853.1943.2766
FluxAttn (FA-TA)Backbone=Qwen3-8B, Attention Mechanism=FluxAttn (FA-TA)2026.0487.4986.1760.85—60.7563.0373.51
TriangleMixBackbone=Qwen3-4B, Attention Mechanism=TriangleMix2026.0487.4285.158.7367.9450.9744.4763.74
PruLongBackbone=Llama-3.1-8B-Instruct, Attention Mechanism=PruLong2026.0486.9676.5570.6554.5248.183059.87
FluxAttn (FA-XA)Backbone=Qwen3-4B, Attention Mechanism=FluxAttn (FA-XA)2026.0486.7984.9459.5268.8251.7743.4363.67
DuoAttentionBackbone=Qwen3-8B, Attention Mechanism=DuoAttention2026.0486.6886.0163.2377.5261.561.9572.41
FluxAttn (FA-XA)Backbone=Qwen3-8B, Attention Mechanism=FluxAttn (FA-XA)2026.0485.8885.5465.0881.9565.0965.3874.65
FluxAttn (FA-TA)Backbone=Qwen3-4B, Attention Mechanism=FluxAttn (FA-TA)2026.0484.2884.5360.5868.651.9151.6465.55
FluxAttn (FA-SSA)Backbone=Qwen3-8B, Attention Mechanism=FluxAttn (FA-SSA)2026.0484.0981.960.5879.364.7465.2773.03
PruLongBackbone=Qwen3-8B, Attention Mechanism=PruLong2026.0483.8580.8660.0577.2562.5461.4970.97
FluxAttn (FA-SSA)Backbone=Qwen3-8B, Attention Mechanism=FluxAttn (FA-SSA), Configuration=Variant 22026.0483.548159.7977.9364.8365.1272.51
FluxAttn (FA-SSA)Backbone=Llama-3.1-8B-Instruct, Attention Mechanism=FluxAttn (FA-SSA)2026.0482.8878.0970.3952.2962.250.7376.75
FluxAttn (FA-SSA)Backbone=Qwen3-4B, Attention Mechanism=FluxAttn (FA-SSA)2026.0481.5882.1158.7372.8952.8156.9166.95
TriangleMixBackbone=Qwen3-8B, Attention Mechanism=TriangleMix2026.0481.0175.6763.4973.7661.5466.8470.47
FluxAttn (FA-SSA)Backbone=Qwen3-4B, Attention Mechanism=FluxAttn (FA-SSA), Configuration=Variant 22026.0480.3680.7556.0871.4959.175667.19
DuoAttentionBackbone=Qwen3-4B, Attention Mechanism=DuoAttention2026.0479.3876.0852.9169.0243.2844.9660.67
PruLongBackbone=Qwen3-4B, Attention Mechanism=PruLong2026.0474.2175.7247.8859.2747.145.6960.25