# Question Answering on SQuAD 2.0 (Factual Accuracy & Relevance) Source: [Question Answering on SQuAD 2.0 (Factual Accuracy & Relevance)]() Collection index: [llms.txt]() Data updated: 2026-05-20T08:34:08.848Z Evaluation of model performance on the SQuAD 2.0 Question Answering dataset, focusing on custom qualitative metrics such as Factual Accuracy, Completeness, Relevance, and Clarity. 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 | Question Answering Evaluation | | Dataset (including any stated split) | SQuAD 2.0 | | Metric | Factual Accuracy | | Unit | SCALAR | | Direction | Higher is better | | Results loaded | 10 of 10 | | 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 - [Factual Accuracy](): Higher is better; unit: SCALAR. - [Completeness](): Higher is better; unit: SCALAR. - [Relevance](): Higher is better; unit: SCALAR. - [Clarity](): Higher is better; unit: SCALAR. - [Binary Accuracy](): Higher is better; unit: PERCENT. ## Recorded results | Method | Factual Accuracy | Result date | Report paper | | --- | --- | --- | --- | | EmbGen | 1.66 | Not recorded | 6a0d4045f984739ed67379f9 | | InstructLab | 1.65 | Not recorded | 6a0d4045f984739ed67379f9 | | InstructLab | 1.65 | Not recorded | 6a0d4045f984739ed67379f9 | | EmbGen | 1.6 | Not recorded | 6a0d4045f984739ed67379f9 | | Knowledge-Instruct | 1.4 | Not recorded | 6a0d4045f984739ed67379f9 | | Knowledge-Instruct | 1.39 | Not recorded | 6a0d4045f984739ed67379f9 | | EntiGraph | 1.37 | Not recorded | 6a0d4045f984739ed67379f9 | | EntiGraph Squad 5M | 1.29 | Not recorded | 6a0d4045f984739ed67379f9 | | Baseline: llama 8B Instruct | 1.28 | Not recorded | 6a0d4045f984739ed67379f9 | | Baseline: llama 8B Instruct | 1.28 | Not recorded | 6a0d4045f984739ed67379f9 | ### EmbGen Factual Accuracy: 1.66. Result ID: 6a0d71f5fbc373592c057d48. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Size":"20M"}. This record does not assert independent reproduction. ### InstructLab Factual Accuracy: 1.65. Result ID: 6a0d71f5fbc373592c057d41. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Size":"5M"}. This record does not assert independent reproduction. ### InstructLab Factual Accuracy: 1.65. Result ID: 6a0d71f5fbc373592c057d46. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Size":"20M"}. This record does not assert independent reproduction. ### EmbGen Factual Accuracy: 1.6. Result ID: 6a0d71f5fbc373592c057d43. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Size":"5M"}. This record does not assert independent reproduction. ### Knowledge-Instruct Factual Accuracy: 1.4. Result ID: 6a0d71f5fbc373592c057d47. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Size":"20M"}. This record does not assert independent reproduction. ### Knowledge-Instruct Factual Accuracy: 1.39. Result ID: 6a0d71f5fbc373592c057d42. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Size":"5M"}. This record does not assert independent reproduction. ### EntiGraph Factual Accuracy: 1.37. Result ID: 6a0d71f5fbc373592c057d45. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Size":"20M"}. This record does not assert independent reproduction. ### EntiGraph Squad 5M Factual Accuracy: 1.29. Result ID: 6a0d71f5fbc373592c057d40. [Reporting paper]() [Method origin paper]() Recorded evaluation configuration: {"Size":"5M"}. This record does not assert independent reproduction. ### Baseline: llama 8B Instruct Factual Accuracy: 1.28. Result ID: 6a0d71f5fbc373592c057d3f. [Reporting paper]() Recorded evaluation configuration: {"Size":"5M"}. This record does not assert independent reproduction. ### Baseline: llama 8B Instruct Factual Accuracy: 1.28. Result ID: 6a0d71f5fbc373592c057d44. [Reporting paper]() Recorded evaluation configuration: {"Size":"20M"}. This record does not assert independent reproduction. Page 1 of 1. This is an index page, not the full collection.