Automated Leaderboard Update

all_data_tasks/4/default.jsonl

@@ -1,4 +1,4 @@
-{"index":198,"Rank":1,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/nvidia\/NV-Retriever-v1\">NV-Retriever-v1<\/a>","Model Size (Million Parameters)":7851,"Memory Usage (GB, fp32)":29.25,"Average":60.9,"ArguAna":68.28,"ClimateFEVER":43.47,"CQADupstackRetrieval":49.36,"DBPedia":50.82,"FEVER":93.15,"FiQA2018":61.18,"HotpotQA":79.12,"MSMARCO":44.89,"NFCorpus":45.06,"NQ":72.44,"QuoraRetrieval":88.78,"SCIDOCS":22.55,"SciFact":81.31,"Touche2020":26.6,"TRECCOVID":86.44}
+{"index":198,"Rank":1,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/nvidia\/NV-Retriever-v1\">NV-Retriever-v1<\/a>","Model Size (Million Parameters)":7111,"Memory Usage (GB, fp32)":26.49,"Average":60.9,"ArguAna":68.28,"ClimateFEVER":43.47,"CQADupstackRetrieval":49.36,"DBPedia":50.82,"FEVER":93.15,"FiQA2018":61.18,"HotpotQA":79.12,"MSMARCO":44.89,"NFCorpus":45.06,"NQ":72.44,"QuoraRetrieval":88.78,"SCIDOCS":22.55,"SciFact":81.31,"Touche2020":26.6,"TRECCOVID":86.44}
 {"index":16,"Rank":2,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/Alibaba-NLP\/gte-Qwen2-7B-instruct\">gte-Qwen2-7B-instruct<\/a>","Model Size (Million Parameters)":7613,"Memory Usage (GB, fp32)":28.36,"Average":60.25,"ArguAna":64.27,"ClimateFEVER":45.88,"CQADupstackRetrieval":46.43,"DBPedia":52.42,"FEVER":95.11,"FiQA2018":62.03,"HotpotQA":73.08,"MSMARCO":45.98,"NFCorpus":40.6,"NQ":67.0,"QuoraRetrieval":90.09,"SCIDOCS":28.91,"SciFact":79.06,"Touche2020":30.57,"TRECCOVID":82.26}
 {"index":54,"Rank":3,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/Linq-AI-Research\/Linq-Embed-Mistral\">Linq-Embed-Mistral<\/a>","Model Size (Million Parameters)":7111,"Memory Usage (GB, fp32)":26.49,"Average":60.19,"ArguAna":69.65,"ClimateFEVER":39.11,"CQADupstackRetrieval":47.27,"DBPedia":51.32,"FEVER":92.42,"FiQA2018":61.2,"HotpotQA":76.24,"MSMARCO":45.21,"NFCorpus":41.62,"NQ":70.63,"QuoraRetrieval":90.27,"SCIDOCS":21.93,"SciFact":78.32,"Touche2020":30.61,"TRECCOVID":87.1}
 {"index":91,"Rank":4,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/Salesforce\/SFR-Embedding-2_R\">SFR-Embedding-2_R<\/a>","Model Size (Million Parameters)":7111,"Memory Usage (GB, fp32)":26.49,"Average":60.18,"ArguAna":62.34,"ClimateFEVER":34.43,"CQADupstackRetrieval":46.11,"DBPedia":51.21,"FEVER":92.16,"FiQA2018":61.77,"HotpotQA":81.36,"MSMARCO":42.18,"NFCorpus":41.34,"NQ":73.96,"QuoraRetrieval":89.58,"SCIDOCS":24.87,"SciFact":85.91,"Touche2020":28.18,"TRECCOVID":87.27}

boards_data/en/data_overall/default.jsonl

@@ -213,7 +213,7 @@
 {"index":185,"Rank":225,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/neuralmagic\/bge-large-en-v1.5-sparse\">bge-large-en-v1.5-sparse<\/a>","Model Size (Million Parameters)":"","Memory Usage (GB, fp32)":"","Embedding Dimensions":"","Max Tokens":"","Average (56 datasets)":"","Classification Average (12 datasets)":"","Clustering Average (11 datasets)":"","PairClassification Average (3 datasets)":85.06,"Reranking Average (4 datasets)":"","Retrieval Average (15 datasets)":"","STS Average (10 datasets)":82.5,"Summarization Average (1 datasets)":""}
 {"index":186,"Rank":226,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/neuralmagic\/bge-small-en-v1.5-quant\">bge-small-en-v1.5-quant<\/a>","Model Size (Million Parameters)":"","Memory Usage (GB, fp32)":"","Embedding Dimensions":"","Max Tokens":"","Average (56 datasets)":"","Classification Average (12 datasets)":73.52,"Clustering Average (11 datasets)":42.63,"PairClassification Average (3 datasets)":85.05,"Reranking Average (4 datasets)":57.86,"Retrieval Average (15 datasets)":"","STS Average (10 datasets)":82.04,"Summarization Average (1 datasets)":""}
 {"index":187,"Rank":227,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/neuralmagic\/bge-small-en-v1.5-sparse\">bge-small-en-v1.5-sparse<\/a>","Model Size (Million Parameters)":"","Memory Usage (GB, fp32)":"","Embedding Dimensions":"","Max Tokens":"","Average (56 datasets)":"","Classification Average (12 datasets)":68.55,"Clustering Average (11 datasets)":38.37,"PairClassification Average (3 datasets)":81.93,"Reranking Average (4 datasets)":54.14,"Retrieval Average (15 datasets)":"","STS Average (10 datasets)":78.53,"Summarization Average (1 datasets)":""}
-{"index":198,"Rank":228,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/nvidia\/NV-Retriever-v1\">NV-Retriever-v1<\/a>","Model Size (Million Parameters)":7851,"Memory Usage (GB, fp32)":29.25,"Embedding Dimensions":4096,"Max Tokens":512,"Average (56 datasets)":"","Classification Average (12 datasets)":"","Clustering Average (11 datasets)":"","PairClassification Average (3 datasets)":"","Reranking Average (4 datasets)":"","Retrieval Average (15 datasets)":60.9,"STS Average (10 datasets)":"","Summarization Average (1 datasets)":""}
+{"index":198,"Rank":228,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/nvidia\/NV-Retriever-v1\">NV-Retriever-v1<\/a>","Model Size (Million Parameters)":7111,"Memory Usage (GB, fp32)":26.49,"Embedding Dimensions":4096,"Max Tokens":512,"Average (56 datasets)":"","Classification Average (12 datasets)":"","Clustering Average (11 datasets)":"","PairClassification Average (3 datasets)":"","Reranking Average (4 datasets)":"","Retrieval Average (15 datasets)":60.9,"STS Average (10 datasets)":"","Summarization Average (1 datasets)":""}
 {"index":202,"Rank":229,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/sdadas\/mmlw-e5-base\">mmlw-e5-base<\/a>","Model Size (Million Parameters)":278,"Memory Usage (GB, fp32)":1.04,"Embedding Dimensions":768,"Max Tokens":514,"Average (56 datasets)":"","Classification Average (12 datasets)":"","Clustering Average (11 datasets)":"","PairClassification Average (3 datasets)":"","Reranking Average (4 datasets)":"","Retrieval Average (15 datasets)":"","STS Average (10 datasets)":"","Summarization Average (1 datasets)":""}
 {"index":203,"Rank":230,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/sdadas\/mmlw-e5-large\">mmlw-e5-large<\/a>","Model Size (Million Parameters)":560,"Memory Usage (GB, fp32)":2.09,"Embedding Dimensions":1024,"Max Tokens":514,"Average (56 datasets)":"","Classification Average (12 datasets)":"","Clustering Average (11 datasets)":"","PairClassification Average (3 datasets)":"","Reranking Average (4 datasets)":"","Retrieval Average (15 datasets)":"","STS Average (10 datasets)":"","Summarization Average (1 datasets)":""}
 {"index":204,"Rank":231,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/sdadas\/mmlw-e5-small\">mmlw-e5-small<\/a>","Model Size (Million Parameters)":118,"Memory Usage (GB, fp32)":0.44,"Embedding Dimensions":384,"Max Tokens":512,"Average (56 datasets)":"","Classification Average (12 datasets)":"","Clustering Average (11 datasets)":"","PairClassification Average (3 datasets)":"","Reranking Average (4 datasets)":"","Retrieval Average (15 datasets)":"","STS Average (10 datasets)":"","Summarization Average (1 datasets)":""}

boards_data/en/data_tasks/Retrieval/default.jsonl

@@ -1,4 +1,4 @@
-{"index":198,"Rank":1,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/nvidia\/NV-Retriever-v1\">NV-Retriever-v1<\/a>","Model Size (Million Parameters)":7851,"Memory Usage (GB, fp32)":29.25,"Average":60.9,"ArguAna":68.28,"ClimateFEVER":43.47,"CQADupstackRetrieval":49.36,"DBPedia":50.82,"FEVER":93.15,"FiQA2018":61.18,"HotpotQA":79.12,"MSMARCO":44.89,"NFCorpus":45.06,"NQ":72.44,"QuoraRetrieval":88.78,"SCIDOCS":22.55,"SciFact":81.31,"Touche2020":26.6,"TRECCOVID":86.44}
+{"index":198,"Rank":1,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/nvidia\/NV-Retriever-v1\">NV-Retriever-v1<\/a>","Model Size (Million Parameters)":7111,"Memory Usage (GB, fp32)":26.49,"Average":60.9,"ArguAna":68.28,"ClimateFEVER":43.47,"CQADupstackRetrieval":49.36,"DBPedia":50.82,"FEVER":93.15,"FiQA2018":61.18,"HotpotQA":79.12,"MSMARCO":44.89,"NFCorpus":45.06,"NQ":72.44,"QuoraRetrieval":88.78,"SCIDOCS":22.55,"SciFact":81.31,"Touche2020":26.6,"TRECCOVID":86.44}
 {"index":16,"Rank":2,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/Alibaba-NLP\/gte-Qwen2-7B-instruct\">gte-Qwen2-7B-instruct<\/a>","Model Size (Million Parameters)":7613,"Memory Usage (GB, fp32)":28.36,"Average":60.25,"ArguAna":64.27,"ClimateFEVER":45.88,"CQADupstackRetrieval":46.43,"DBPedia":52.42,"FEVER":95.11,"FiQA2018":62.03,"HotpotQA":73.08,"MSMARCO":45.98,"NFCorpus":40.6,"NQ":67.0,"QuoraRetrieval":90.09,"SCIDOCS":28.91,"SciFact":79.06,"Touche2020":30.57,"TRECCOVID":82.26}
 {"index":54,"Rank":3,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/Linq-AI-Research\/Linq-Embed-Mistral\">Linq-Embed-Mistral<\/a>","Model Size (Million Parameters)":7111,"Memory Usage (GB, fp32)":26.49,"Average":60.19,"ArguAna":69.65,"ClimateFEVER":39.11,"CQADupstackRetrieval":47.27,"DBPedia":51.32,"FEVER":92.42,"FiQA2018":61.2,"HotpotQA":76.24,"MSMARCO":45.21,"NFCorpus":41.62,"NQ":70.63,"QuoraRetrieval":90.27,"SCIDOCS":21.93,"SciFact":78.32,"Touche2020":30.61,"TRECCOVID":87.1}
 {"index":91,"Rank":4,"Model":"<a target=\"_blank\" style=\"text-decoration: underline\" href=\"https:\/\/huggingface.co\/Salesforce\/SFR-Embedding-2_R\">SFR-Embedding-2_R<\/a>","Model Size (Million Parameters)":7111,"Memory Usage (GB, fp32)":26.49,"Average":60.18,"ArguAna":62.34,"ClimateFEVER":34.43,"CQADupstackRetrieval":46.11,"DBPedia":51.21,"FEVER":92.16,"FiQA2018":61.77,"HotpotQA":81.36,"MSMARCO":42.18,"NFCorpus":41.34,"NQ":73.96,"QuoraRetrieval":89.58,"SCIDOCS":24.87,"SciFact":85.91,"Touche2020":28.18,"TRECCOVID":87.27}