Long-context language understanding on L-Eval benchmark leaderboard · SOTA2 Research

Long-context language understanding on L-Eval

58.28CourseraNTK

Evaluation Results

Method	Links
NTK2025.02		58.28	—	—	—	59.38	1.11	39.59
NTK2025.02		57.7	80	63.86	81.04	64.06	5.56	58.7
GALI2025.02		56.54	74	65.35	84.06	66.41	8.89	59.21
YaRN2025.02		56.4	81	59.4	79.18	64.06	5.56	57.6
YaRN2025.02		55.96	—	—	—	62.5	5.56	41.34
SelfExtend2025.02		55.38	—	—	—	64.06	5.56	41.67
SelfExtend2025.02		55.23	79	64.36	79.18	67.97	5.56	58.55
YaRN2025.02		55.23	—	—	—	67.19	7.78	43.4
GALI2025.02		54.65	—	—	—	65.63	6.67	42.32
SelfExtend2025.02		54.51	80	64.36	77.7	67.97	5.56	58.35
ChunkLlama2025.02		54.36	—	—	—	64.06	5.56	41.33
Dyn-NTK2025.02		54.36	—	—	—	64.06	6.67	41.7
GALI2025.02		54.17	74	65.35	84.06	68.75	7.78	59.1
GALI2025.02		54.17	—	—	—	66.41	7.78	42.79
Dyn-NTK2025.02		54.07	75	64.36	82.16	67.97	1.11	57.44
SelfExtend2025.02		53.92	—	—	—	65.63	3.33	40.96
Original2025.02		53.34	75	59.41	81.41	60.94	4.44	55.76
ChunkLlama2025.02		53.34	—	—	—	61.72	5.56	40.21
ChunkLlama2025.02		53.2	75	63.37	81.04	63.28	2.22	56.35
Original2025.02		53.05	—	—	—	60.16	4.44	39.22
YaRN2025.02		53.05	73	59.41	79.55	68.75	5.56	56.55
NTK2025.02		52.91	82	61.39	79.93	67.19	2.22	57.6
ChunkLlama2025.02		52.62	77	63.37	81.04	60.16	3.33	56.25
Dyn-NTK2025.02		52.33	76	63.86	82.16	71.88	3.33	58.26
NTK2025.02		52.03	—	—	—	42.97	0	31.67
Dyn-NTK2025.02		52.03	—	—	—	52.34	2.22	35.53