# Length Extrapolation on RULER (8K-256K Contexts) Source: [Length Extrapolation on RULER (8K-256K Contexts)]() Collection index: [llms.txt]() Data updated: 2026-02-26T05:57:49.971Z Evaluation of model performance on the RULER dataset focusing on Length Extrapolation capabilities across context lengths ranging from 8K up to 256K tokens, including derived performance and error metrics. A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. | Scope | Recorded value | | --- | --- | | Task | Length Extrapolation | | Dataset (including any stated split) | RULER | | Metric | Performance @ 8K Context | | Unit | PERCENT | | Direction | Higher is better | | Results loaded | 18 of 18 | | Coverage | All results currently returned for this benchmark | Evaluation settings may differ between results. The original papers and recorded configurations are needed to establish comparability. A shared metric name alone does not establish an identical protocol. ## Metrics - [Performance @ 8K Context](): Higher is better; unit: PERCENT. - [Performance @ 16K Context](): Higher is better; unit: PERCENT. - [Performance @ 32K Context](): Higher is better; unit: PERCENT. - [Performance @ 64K Context](): Higher is better; unit: PERCENT. - [Performance @ 128K Context](): Higher is better; unit: PERCENT. - [Performance @ 256K Context](): Higher is better; unit: PERCENT. - [Overall Performance Score](): Higher is better; unit: PERCENT. - [Omega MSR Score](): Higher is better; unit: SCALAR. ## Recorded results | Method | Performance @ 8K Context | Result date | Report paper | | --- | --- | --- | --- | | Llama-3.1-8B-Instruct | 92.88 | Not recorded | 69784664a2b28e761d87c358 | | Elastic Attention | 92.82 | Not recorded | 69784664a2b28e761d87c358 | | Qwen3-8B | 89.69 | Not recorded | 69784664a2b28e761d87c358 | | InfLLM-V2 | 89.3 | Not recorded | 69784664a2b28e761d87c358 | | Qwen3-4B | 87.49 | Not recorded | 69784664a2b28e761d87c358 | | PruLong | 87.2 | Not recorded | 69784664a2b28e761d87c358 | | DuoAttention | 87.2 | Not recorded | 69784664a2b28e761d87c358 | | Elastic Attention | 86.62 | Not recorded | 69784664a2b28e761d87c358 | | Elastic Attention | 86.56 | Not recorded | 69784664a2b28e761d87c358 | | Elastic Attention | 85.07 | Not recorded | 69784664a2b28e761d87c358 | | PruLong | 83.54 | Not recorded | 69784664a2b28e761d87c358 | | Elastic Attention | 83.35 | Not recorded | 69784664a2b28e761d87c358 | | DuoAttention | 81.03 | Not recorded | 69784664a2b28e761d87c358 | | Elastic Attention | 80.2 | Not recorded | 69784664a2b28e761d87c358 | | InfLLM-V2 | 78.59 | Not recorded | 69784664a2b28e761d87c358 | | PruLong | 77.03 | Not recorded | 69784664a2b28e761d87c358 | | InfLLM-V2 | 77.01 | Not recorded | 69784664a2b28e761d87c358 | | DuoAttention | 76.82 | Not recorded | 69784664a2b28e761d87c358 | ### Llama-3.1-8B-Instruct Performance @ 8K Context: 92.88. Result ID: 699670185698665ec4b723aa. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction. ### Elastic Attention Performance @ 8K Context: 92.82. Result ID: 699670185698665ec4b723af. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Sparsity Config":"FA-XA"}. This record does not assert independent reproduction. ### Qwen3-8B Performance @ 8K Context: 89.69. Result ID: 699670185698665ec4b723a4. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-8B"}. This record does not assert independent reproduction. ### InfLLM-V2 Performance @ 8K Context: 89.3. Result ID: 699670185698665ec4b723ab. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction. ### Qwen3-4B Performance @ 8K Context: 87.49. Result ID: 699670185698665ec4b7239e. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B"}. This record does not assert independent reproduction. ### PruLong Performance @ 8K Context: 87.2. Result ID: 699670185698665ec4b723a7. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-8B"}. This record does not assert independent reproduction. ### DuoAttention Performance @ 8K Context: 87.2. Result ID: 699670185698665ec4b723ac. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction. ### Elastic Attention Performance @ 8K Context: 86.62. Result ID: 699670185698665ec4b723a8. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Qwen3-8B","Sparsity Config":"FA-SSA"}. This record does not assert independent reproduction. ### Elastic Attention Performance @ 8K Context: 86.56. Result ID: 699670185698665ec4b723a3. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B","Sparsity Config":"FA-XA"}. This record does not assert independent reproduction. ### Elastic Attention Performance @ 8K Context: 85.07. Result ID: 699670185698665ec4b723a9. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Qwen3-8B","Sparsity Config":"FA-XA"}. This record does not assert independent reproduction. ### PruLong Performance @ 8K Context: 83.54. Result ID: 699670185698665ec4b723ad. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct"}. This record does not assert independent reproduction. ### Elastic Attention Performance @ 8K Context: 83.35. Result ID: 699670185698665ec4b723a2. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B","Sparsity Config":"FA-SSA"}. This record does not assert independent reproduction. ### DuoAttention Performance @ 8K Context: 81.03. Result ID: 699670185698665ec4b723a6. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-8B"}. This record does not assert independent reproduction. ### Elastic Attention Performance @ 8K Context: 80.2. Result ID: 699670185698665ec4b723ae. [Reporting paper]() [Method origin paper]() [Implementation]() Recorded evaluation configuration: {"Backbone":"Llama-3.1-8B-Instruct","Sparsity Config":"FA-SSA"}. This record does not assert independent reproduction. ### InfLLM-V2 Performance @ 8K Context: 78.59. Result ID: 699670185698665ec4b7239f. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B"}. This record does not assert independent reproduction. ### PruLong Performance @ 8K Context: 77.03. Result ID: 699670185698665ec4b723a1. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B"}. This record does not assert independent reproduction. ### InfLLM-V2 Performance @ 8K Context: 77.01. Result ID: 699670185698665ec4b723a5. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-8B"}. This record does not assert independent reproduction. ### DuoAttention Performance @ 8K Context: 76.82. Result ID: 699670185698665ec4b723a0. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Backbone":"Qwen3-4B"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.