update boards_data and config.yaml

boards_data/en-x/data_tasks/mynew/default.jsonl

@@ -0,0 +1 @@
+{"index":235,"Rank":1,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/nvidia\/NV-Embed-v2\">NV-Embed-v2<\/a>","1":2,"！！！":99999}

config.yaml

@@ -53,7 +53,334 @@ tasks:
     metric: "p-MRR"
     metric_description: "paired mean reciprocal rank (p-MRR)"
     task_description: "Retrieval w/Instructions is the task of finding relevant documents for a query that has detailed instructions."
+  MAIR:
+    icon: "🍋"
+    metric: "NDCG@10"
+    metric_description: "NDCG@10"
+    task_description: "MAIR: A Massive Benchmark for Evaluating Instructed Retrieval. Evaluate your retrieval models on 126 diverse tasks [EMNLP 2024]. https://github.com/sunnweiwei/MAIR"
 boards:
+  ovr:
+    title: "Overall"
+    language_long: "English"
+    has_overall: false
+    acronym: null
+    icon: null
+    special_icons: null
+    credits: "https://arxiv.org/abs/2410.10127"
+    tasks:
+      MAIR:
+        - Competition-Math
+        - ProofWiki_Proof
+        - ProofWiki_Reference
+        - Stacks_Proof
+        - Stacks_Reference
+        - Stein_Proof
+        - Stein_Reference
+        - Trench_Proof
+        - Trench_Reference
+        - TAD
+        - TAS2
+        - StackMathQA
+        - SciDocs
+        - SciFact
+        - LitSearch
+        - FairRanking_2020
+        - APPS
+        - CodeEditSearch
+        - CodeSearchNet
+        - Conala
+        - HumanEval-X
+        - LeetCode
+        - MBPP
+        - RepoBench
+        - TLDR
+        - SWE-Bench-Lite
+        - FoodAPI
+        - HuggingfaceAPI
+        - PytorchAPI
+        - SpotifyAPI
+        - TMDB
+        - TensorAPI
+        - ToolBench
+        - WeatherAPI
+        - AY2
+        - ELI5
+        - Fever
+        - TREx
+        - WnCw
+        - WnWi
+        - WoW
+        - zsRE
+        - ArguAna
+        - CQADupStack
+        - Quora
+        - TopiOCQA
+        - Touche
+        - ACORDAR
+        - CPCD
+        - ChroniclingAmericaQA
+        - NTCIR
+        - PointRec
+        - ProCIS-Dialog
+        - ProCIS-Turn
+        - QuanTemp
+        - WebTableSearch
+        - MISeD
+        - SParC
+        - SParC-SQL
+        - Spider
+        - Spider-SQL
+        - CAsT_2019
+        - CAsT_2020
+        - CAsT_2021
+        - CAsT_2022
+        - Core_2017
+        - Microblog_2011
+        - Microblog_2012
+        - Microblog_2013
+        - Microblog_2014
+        - DD_2015
+        - DD_2016
+        - DD_2017
+        - FairRanking_2021
+        - FairRanking_2022
+        - NeuCLIR-Tech_2023
+        - NeuCLIR_2022
+        - NeuCLIR_2023
+        - ProductSearch_2023
+        - ToT_2023
+        - ToT_2024
+        - ExcluIR
+        - Core17
+        - News21
+        - Robust04
+        - InstructIR
+        - NevIR
+        - IFEval
+        - AILA2019-Case
+        - AILA2019-Statutes
+        - BSARD
+        - BillSum
+        - CUAD
+        - GerDaLIR
+        - LeCaRDv2
+        - LegalQuAD
+        - REGIR-EU2UK
+        - REGIR-UK2EU
+        - TREC-Legal_2011
+        - NFCorpus
+        - Trec-Covid
+        - Monant
+        - CARE
+        - PrecisionMedicine_2017
+        - PrecisionMedicine_2018
+        - PrecisionMedicine_2019
+        - PrecisionMedicine-Article_2019
+        - PrecisionMedicine-Article_2020
+        - CliniDS_2014
+        - CliniDS_2015
+        - CliniDS_2016
+        - ClinicalTrials_2021
+        - ClinicalTrials_2022
+        - ClinicalTrials_2023
+        - Genomics-AdHoc_2004
+        - Genomics-AdHoc_2005
+        - Genomics-AdHoc_2006
+        - Genomics-AdHoc_2007
+        - Apple
+        - ConvFinQA
+        - FinQA
+        - FinanceBench
+        - HC3Finance
+        - TAT-DQA
+        - Trade-the-event
+        - FiQA
+  wb:
+    title: "Web"
+    language_long: "English"
+    has_overall: false
+    acronym: null
+    icon: null
+    special_icons: null
+    credits: "https://arxiv.org/abs/2410.10127"
+    tasks:
+      MAIR:
+        - AY2
+        - ELI5
+        - Fever
+        - TREx
+        - WnCw
+        - WnWi
+        - WoW
+        - zsRE
+        - ArguAna
+        - CQADupStack
+        - Quora
+        - TopiOCQA
+        - Touche
+        - ACORDAR
+        - CPCD
+        - ChroniclingAmericaQA
+        - NTCIR
+        - PointRec
+        - ProCIS-Dialog
+        - ProCIS-Turn
+        - QuanTemp
+        - WebTableSearch
+        - MISeD
+        - SParC
+        - SParC-SQL
+        - Spider
+        - Spider-SQL
+        - CAsT_2019
+        - CAsT_2020
+        - CAsT_2021
+        - CAsT_2022
+        - Core_2017
+        - Microblog_2011
+        - Microblog_2012
+        - Microblog_2013
+        - Microblog_2014
+        - DD_2015
+        - DD_2016
+        - DD_2017
+        - FairRanking_2021
+        - FairRanking_2022
+        - NeuCLIR-Tech_2023
+        - NeuCLIR_2022
+        - NeuCLIR_2023
+        - ProductSearch_2023
+        - ToT_2023
+        - ToT_2024
+        - ExcluIR
+        - Core17
+        - News21
+        - Robust04
+        - InstructIR
+        - NevIR
+        - IFEval
+  acd:
+    title: "Academic"
+    language_long: "English"
+    has_overall: false
+    acronym: null
+    icon: null
+    special_icons: null
+    credits: "https://arxiv.org/abs/2410.10127"
+    tasks:
+      MAIR:
+        - Competition-Math
+        - ProofWiki_Proof
+        - ProofWiki_Reference
+        - Stacks_Proof
+        - Stacks_Reference
+        - Stein_Proof
+        - Stein_Reference
+        - Trench_Proof
+        - Trench_Reference
+        - TAD
+        - TAS2
+        - StackMathQA
+        - SciDocs
+        - SciFact
+        - LitSearch
+        - FairRanking_2020
+  lg:
+    title: "Legal"
+    language_long: "English"
+    has_overall: false
+    acronym: null
+    icon: null
+    special_icons: null
+    credits: "https://arxiv.org/abs/2410.10127"
+    tasks:
+      MAIR:
+        - AILA2019-Case
+        - AILA2019-Statutes
+        - BSARD
+        - BillSum
+        - CUAD
+        - GerDaLIR
+        - LeCaRDv2
+        - LegalQuAD
+        - REGIR-EU2UK
+        - REGIR-UK2EU
+        - TREC-Legal_2011
+  mdc:
+    title: "Medical"
+    language_long: "English"
+    has_overall: false
+    acronym: null
+    icon: null
+    special_icons: null
+    credits: "https://arxiv.org/abs/2410.10127"
+    tasks:
+      MAIR:
+        - NFCorpus
+        - Trec-Covid
+        - Monant
+        - CARE
+        - PrecisionMedicine_2017
+        - PrecisionMedicine_2018
+        - PrecisionMedicine_2019
+        - PrecisionMedicine-Article_2019
+        - PrecisionMedicine-Article_2020
+        - CliniDS_2014
+        - CliniDS_2015
+        - CliniDS_2016
+        - ClinicalTrials_2021
+        - ClinicalTrials_2022
+        - ClinicalTrials_2023
+        - Genomics-AdHoc_2004
+        - Genomics-AdHoc_2005
+        - Genomics-AdHoc_2006
+        - Genomics-AdHoc_2007
+  fnc:
+    title: "Finance"
+    language_long: "English"
+    has_overall: false
+    acronym: null
+    icon: null
+    special_icons: null
+    credits: "https://arxiv.org/abs/2410.10127"
+    tasks:
+      MAIR:
+        - Apple
+        - ConvFinQA
+        - FinQA
+        - FinanceBench
+        - HC3Finance
+        - TAT-DQA
+        - Trade-the-event
+        - FiQA
+  cd:
+    title: "Code"
+    language_long: "English"
+    has_overall: false
+    acronym: null
+    icon: null
+    special_icons: null
+    credits: "https://arxiv.org/abs/2410.10127"
+    tasks:
+      MAIR:
+        - APPS
+        - CodeEditSearch
+        - CodeSearchNet
+        - Conala
+        - HumanEval-X
+        - LeetCode
+        - MBPP
+        - RepoBench
+        - TLDR
+        - SWE-Bench-Lite
+        - FoodAPI
+        - HuggingfaceAPI
+        - PytorchAPI
+        - SpotifyAPI
+        - TMDB
+        - TensorAPI
+        - ToolBench
+        - WeatherAPI
   en:
     title: English
     language_long: "English"