diff --git a/docs/results/cl-nagoya/ruri-base/summary.json b/docs/results/cl-nagoya/ruri-base/summary.json new file mode 100644 index 0000000..a7c7b05 --- /dev/null +++ b/docs/results/cl-nagoya/ruri-base/summary.json @@ -0,0 +1,62 @@ +{ + "Classification": { + "amazon_counterfactual_classification": { + "macro_f1": 0.7665550732749669 + }, + "amazon_review_classification": { + "macro_f1": 0.5575876111411316 + }, + "massive_intent_classification": { + "macro_f1": 0.8141210121425055 + }, + "massive_scenario_classification": { + "macro_f1": 0.8848812917656395 + } + }, + "Reranking": { + "esci": { + "ndcg@10": 0.9290942178703699 + } + }, + "Retrieval": { + "jagovfaqs_22k": { + "ndcg@10": 0.7455660589538348 + }, + "jaqket": { + "ndcg@10": 0.5012253145754781 + }, + "mrtydi": { + "ndcg@10": 0.3545113073009125 + }, + "nlp_journal_abs_intro": { + "ndcg@10": 0.8689204088388403 + }, + "nlp_journal_title_abs": { + "ndcg@10": 0.9656989703684407 + }, + "nlp_journal_title_intro": { + "ndcg@10": 0.7531306059721564 + } + }, + "STS": { + "jsick": { + "spearman": 0.8231772134744029 + }, + "jsts": { + "spearman": 0.8342848039994751 + } + }, + "Clustering": { + "livedoor_news": { + "v_measure_score": 0.5427223607801758 + }, + "mewsc16": { + "v_measure_score": 0.5404099864321413 + } + }, + "PairClassification": { + "paws_x_ja": { + "binary_f1": 0.6237623762376238 + } + } +} \ No newline at end of file diff --git a/docs/results/cl-nagoya/ruri-large/summary.json b/docs/results/cl-nagoya/ruri-large/summary.json new file mode 100644 index 0000000..e86c46b --- /dev/null +++ b/docs/results/cl-nagoya/ruri-large/summary.json @@ -0,0 +1,62 @@ +{ + "Classification": { + "amazon_counterfactual_classification": { + "macro_f1": 0.8080806321853091 + }, + "amazon_review_classification": { + "macro_f1": 0.5680171450057119 + }, + "massive_intent_classification": { + "macro_f1": 0.8255898596881264 + }, + "massive_scenario_classification": { + "macro_f1": 0.8956410349938264 + } + }, + "Reranking": { + "esci": { + "ndcg@10": 0.9298524733536755 + } + }, + "Retrieval": { + "jagovfaqs_22k": { + "ndcg@10": 0.7667506664925435 + }, + "jaqket": { + "ndcg@10": 0.6173871224245404 + }, + "mrtydi": { + "ndcg@10": 0.3803302462897418 + }, + "nlp_journal_abs_intro": { + "ndcg@10": 0.8712459719069233 + }, + "nlp_journal_title_abs": { + "ndcg@10": 0.9657898747088243 + }, + "nlp_journal_title_intro": { + "ndcg@10": 0.779665053945222 + } + }, + "STS": { + "jsick": { + "spearman": 0.8199959693684533 + }, + "jsts": { + "spearman": 0.8426164139167538 + } + }, + "Clustering": { + "livedoor_news": { + "v_measure_score": 0.5139491572866559 + }, + "mewsc16": { + "v_measure_score": 0.5225025331595674 + } + }, + "PairClassification": { + "paws_x_ja": { + "binary_f1": 0.6228813559322034 + } + } +} \ No newline at end of file diff --git a/docs/results/cl-nagoya/ruri-small/summary.json b/docs/results/cl-nagoya/ruri-small/summary.json new file mode 100644 index 0000000..cb591ea --- /dev/null +++ b/docs/results/cl-nagoya/ruri-small/summary.json @@ -0,0 +1,62 @@ +{ + "Classification": { + "amazon_counterfactual_classification": { + "macro_f1": 0.7991935990685706 + }, + "amazon_review_classification": { + "macro_f1": 0.556129066893332 + }, + "massive_intent_classification": { + "macro_f1": 0.8148895285345188 + }, + "massive_scenario_classification": { + "macro_f1": 0.8787774569382543 + } + }, + "Reranking": { + "esci": { + "ndcg@10": 0.9300177985352138 + } + }, + "Retrieval": { + "jagovfaqs_22k": { + "ndcg@10": 0.736494039429321 + }, + "jaqket": { + "ndcg@10": 0.484437639428696 + }, + "mrtydi": { + "ndcg@10": 0.3342716158897666 + }, + "nlp_journal_abs_intro": { + "ndcg@10": 0.8768878489670099 + }, + "nlp_journal_title_abs": { + "ndcg@10": 0.9716879343439146 + }, + "nlp_journal_title_intro": { + "ndcg@10": 0.7608660955794895 + } + }, + "STS": { + "jsick": { + "spearman": 0.8343927017558587 + }, + "jsts": { + "spearman": 0.8213297790184827 + } + }, + "Clustering": { + "livedoor_news": { + "v_measure_score": 0.5096442244018489 + }, + "mewsc16": { + "v_measure_score": 0.5141045788711239 + } + }, + "PairClassification": { + "paws_x_ja": { + "binary_f1": 0.6211267605633802 + } + } +} \ No newline at end of file diff --git a/docs/results/pkshatech/GLuCoSE-base-ja-v2/summary.json b/docs/results/pkshatech/GLuCoSE-base-ja-v2/summary.json new file mode 100644 index 0000000..7318aab --- /dev/null +++ b/docs/results/pkshatech/GLuCoSE-base-ja-v2/summary.json @@ -0,0 +1,62 @@ +{ + "Classification": { + "amazon_counterfactual_classification": { + "macro_f1": 0.7492232749031491 + }, + "amazon_review_classification": { + "macro_f1": 0.5530707609927811 + }, + "massive_intent_classification": { + "macro_f1": 0.7979144461303402 + }, + "massive_scenario_classification": { + "macro_f1": 0.8683641924034757 + } + }, + "Reranking": { + "esci": { + "ndcg@10": 0.9301469431250418 + } + }, + "Retrieval": { + "jagovfaqs_22k": { + "ndcg@10": 0.6979374757372254 + }, + "jaqket": { + "ndcg@10": 0.6729417850207029 + }, + "mrtydi": { + "ndcg@10": 0.41858579533990486 + }, + "nlp_journal_abs_intro": { + "ndcg@10": 0.9029337913460675 + }, + "nlp_journal_title_abs": { + "ndcg@10": 0.9511153967130517 + }, + "nlp_journal_title_intro": { + "ndcg@10": 0.7580448576047344 + } + }, + "STS": { + "jsick": { + "spearman": 0.849637366944316 + }, + "jsts": { + "spearman": 0.8095684318108997 + } + }, + "Clustering": { + "livedoor_news": { + "v_measure_score": 0.5151536908540161 + }, + "mewsc16": { + "v_measure_score": 0.45782610528001805 + } + }, + "PairClassification": { + "paws_x_ja": { + "binary_f1": 0.623716814159292 + } + } +} diff --git a/docs/results/pkshatech/RoSEtta-base-ja/summary.json b/docs/results/pkshatech/RoSEtta-base-ja/summary.json new file mode 100644 index 0000000..d82af4b --- /dev/null +++ b/docs/results/pkshatech/RoSEtta-base-ja/summary.json @@ -0,0 +1,62 @@ +{ + "Classification": { + "amazon_counterfactual_classification": { + "macro_f1": 0.7005147244958231 + }, + "amazon_review_classification": { + "macro_f1": 0.5263680453119501 + }, + "massive_intent_classification": { + "macro_f1": 0.7983787583297884 + }, + "massive_scenario_classification": { + "macro_f1": 0.8709593192703351 + } + }, + "Reranking": { + "esci": { + "ndcg@10": 0.9268625513429571 + } + }, + "Retrieval": { + "jagovfaqs_22k": { + "ndcg@10": 0.6595934642903105 + }, + "jaqket": { + "ndcg@10": 0.6533452086105761 + }, + "mrtydi": { + "ndcg@10": 0.36731170141136216 + }, + "nlp_journal_abs_intro": { + "ndcg@10": 0.9553567926226499 + }, + "nlp_journal_title_abs": { + "ndcg@10": 0.940828991756893 + }, + "nlp_journal_title_intro": { + "ndcg@10": 0.8163161967769845 + } + }, + "STS": { + "jsick": { + "spearman": 0.8383455453168481 + }, + "jsts": { + "spearman": 0.7895388048564987 + } + }, + "Clustering": { + "livedoor_news": { + "v_measure_score": 0.5861760622672214 + }, + "mewsc16": { + "v_measure_score": 0.4784844036038961 + } + }, + "PairClassification": { + "paws_x_ja": { + "binary_f1": 0.6173974540311173 + } + } +} diff --git a/leaderboard.md b/leaderboard.md index b07dbca..4b05e46 100644 --- a/leaderboard.md +++ b/leaderboard.md @@ -3,38 +3,48 @@ This leaderboard shows the results stored under `docs/results`. The scores are a ## Summary -The summary shows the average scores within each task. +The summary shows the average scores within each task. The average score is the average of scores by dataset. | Model | Avg. | Retrieval | STS | Classification | Reranking | Clustering | PairClassification | |:----------------------------------------------|:----------|:------------|:----------|:-----------------|:------------|:-------------|:---------------------| -| OpenAI/text-embedding-3-large | **73.97** | **74.48** | 82.52 | **77.58** | **93.58** | **53.32** | 62.35 | -| intfloat/multilingual-e5-large | 71.65 | 70.98 | 79.70 | 72.89 | 92.96 | 51.24 | 62.15 | -| OpenAI/text-embedding-3-small | 70.86 | 66.39 | 79.46 | 73.06 | 92.92 | 51.06 | 62.27 | -| pkshatech/GLuCoSE-base-ja | 70.44 | 59.02 | 78.71 | 76.82 | 91.90 | 49.78 | **66.39** | -| intfloat/multilingual-e5-base | 70.12 | 68.21 | 79.84 | 69.30 | 92.85 | 48.26 | 62.26 | -| intfloat/multilingual-e5-small | 69.52 | 67.27 | 80.07 | 67.62 | 93.03 | 46.91 | 62.19 | -| OpenAI/text-embedding-ada-002 | 69.48 | 64.38 | 79.02 | 69.75 | 93.04 | 48.30 | 62.40 | -| cl-nagoya/sup-simcse-ja-base | 68.56 | 49.64 | 82.05 | 73.47 | 91.83 | 51.79 | 62.57 | -| MU-Kindai/Japanese-SimCSE-BERT-large-unsup | 66.89 | 47.38 | 78.99 | 73.13 | 91.30 | 48.25 | 62.27 | -| oshizo/sbert-jsnli-luke-japanese-base-lite | 66.75 | 43.00 | 76.60 | 76.61 | 91.56 | 50.33 | 62.38 | -| cl-nagoya/sup-simcse-ja-large | 66.51 | 37.62 | **83.18** | 73.73 | 91.48 | 50.56 | 62.51 | -| cl-nagoya/unsup-simcse-ja-large | 66.27 | 40.53 | 80.56 | 74.66 | 90.95 | 48.41 | 62.49 | -| MU-Kindai/Japanese-SimCSE-BERT-base-unsup | 66.23 | 46.36 | 77.49 | 73.30 | 91.16 | 46.68 | 62.38 | -| MU-Kindai/Japanese-SimCSE-BERT-large-sup | 65.28 | 40.82 | 78.28 | 73.47 | 90.95 | 45.81 | 62.35 | -| MU-Kindai/Japanese-MixCSE-BERT-base | 65.14 | 42.59 | 77.05 | 72.90 | 91.01 | 44.95 | 62.33 | -| cl-nagoya/unsup-simcse-ja-base | 65.07 | 40.23 | 78.72 | 73.07 | 91.16 | 44.77 | 62.44 | -| MU-Kindai/Japanese-DiffCSE-BERT-base | 64.77 | 41.79 | 75.50 | 73.77 | 90.95 | 44.22 | 62.38 | -| sentence-transformers/LaBSE | 64.70 | 40.12 | 76.56 | 72.66 | 91.63 | 44.88 | 62.33 | -| pkshatech/simcse-ja-bert-base-clcmlp | 64.42 | 37.00 | 76.80 | 71.30 | 91.49 | 47.53 | 62.40 | -| MU-Kindai/Japanese-SimCSE-BERT-base-sup | 64.15 | 41.32 | 74.66 | 72.76 | 90.66 | 43.11 | 62.37 | -| colorfulscoop/sbert-base-ja | 58.85 | 16.52 | 70.42 | 69.07 | 89.97 | 44.81 | 62.31 | -| sentence-transformers/stsb-xlm-r-multilingual | 58.01 | 21.00 | 75.40 | 71.84 | 90.20 | 27.46 | 62.20 | +| OpenAI/text-embedding-3-large | **74.05** | **74.48** | 82.52 | **77.58** | **93.58** | 53.32 | 62.35 | +| cl-nagoya/ruri-large | 73.31 | 73.02 | 83.13 | 77.43 | 92.99 | 51.82 | 62.29 | +| pkshatech/GLuCoSE-base-ja-v2 | 72.23 | 73.36 | 82.96 | 74.21 | 93.01 | 48.65 | 62.37 | +| pkshatech/RoSEtta-base-ja | 72.04 | 73.21 | 81.39 | 72.41 | 92.69 | 53.23 | 61.74 | +| cl-nagoya/ruri-base | 71.91 | 69.82 | 82.87 | 75.58 | 92.91 | **54.16** | 62.38 | +| cl-nagoya/ruri-small | 71.53 | 69.41 | 82.79 | 76.22 | 93.00 | 51.19 | 62.11 | +| intfloat/multilingual-e5-large | 70.90 | 70.98 | 79.70 | 72.89 | 92.96 | 51.24 | 62.15 | +| OpenAI/text-embedding-3-small | 69.18 | 66.39 | 79.46 | 73.06 | 92.92 | 51.06 | 62.27 | +| intfloat/multilingual-e5-base | 68.61 | 68.21 | 79.84 | 69.30 | 92.85 | 48.26 | 62.26 | +| intfloat/multilingual-e5-small | 67.71 | 67.27 | 80.07 | 67.62 | 93.03 | 46.91 | 62.19 | +| pkshatech/GLuCoSE-base-ja | 67.29 | 59.02 | 78.71 | 76.82 | 91.90 | 49.78 | **66.39** | +| OpenAI/text-embedding-ada-002 | 67.21 | 64.38 | 79.02 | 69.75 | 93.04 | 48.30 | 62.40 | +| cl-nagoya/sup-simcse-ja-base | 63.36 | 49.64 | 82.05 | 73.47 | 91.83 | 51.79 | 62.57 | +| MU-Kindai/Japanese-SimCSE-BERT-large-unsup | 61.55 | 47.38 | 78.99 | 73.13 | 91.30 | 48.25 | 62.27 | +| MU-Kindai/Japanese-SimCSE-BERT-base-unsup | 60.83 | 46.36 | 77.49 | 73.30 | 91.16 | 46.68 | 62.38 | +| oshizo/sbert-jsnli-luke-japanese-base-lite | 60.77 | 43.00 | 76.60 | 76.61 | 91.56 | 50.33 | 62.38 | +| cl-nagoya/unsup-simcse-ja-large | 59.58 | 40.53 | 80.56 | 74.66 | 90.95 | 48.41 | 62.49 | +| MU-Kindai/Japanese-MixCSE-BERT-base | 59.03 | 42.59 | 77.05 | 72.90 | 91.01 | 44.95 | 62.33 | +| cl-nagoya/sup-simcse-ja-large | 58.88 | 37.62 | **83.18** | 73.73 | 91.48 | 50.56 | 62.51 | +| MU-Kindai/Japanese-SimCSE-BERT-large-sup | 58.77 | 40.82 | 78.28 | 73.47 | 90.95 | 45.81 | 62.35 | +| MU-Kindai/Japanese-DiffCSE-BERT-base | 58.66 | 41.79 | 75.50 | 73.77 | 90.95 | 44.22 | 62.38 | +| cl-nagoya/unsup-simcse-ja-base | 58.39 | 40.23 | 78.72 | 73.07 | 91.16 | 44.77 | 62.44 | +| sentence-transformers/LaBSE | 58.01 | 40.12 | 76.56 | 72.66 | 91.63 | 44.88 | 62.33 | +| MU-Kindai/Japanese-SimCSE-BERT-base-sup | 57.97 | 41.32 | 74.66 | 72.76 | 90.66 | 43.11 | 62.37 | +| pkshatech/simcse-ja-bert-base-clcmlp | 56.86 | 37.00 | 76.80 | 71.30 | 91.49 | 47.53 | 62.40 | +| sentence-transformers/stsb-xlm-r-multilingual | 48.21 | 21.00 | 75.40 | 71.84 | 90.20 | 27.46 | 62.20 | +| colorfulscoop/sbert-base-ja | 47.38 | 16.52 | 70.42 | 69.07 | 89.97 | 44.81 | 62.31 | ## Retrieval | Model | Avg. | jagovfaqs_22k
(ndcg@10) | jaqket
(ndcg@10) | mrtydi
(ndcg@10) | nlp_journal_abs_intro
(ndcg@10) | nlp_journal_title_abs
(ndcg@10) | nlp_journal_title_intro
(ndcg@10) | |:----------------------------------------------|:----------|:-----------------------------|:----------------------|:----------------------|:-------------------------------------|:-------------------------------------|:---------------------------------------| -| OpenAI/text-embedding-3-large | **74.48** | **72.41** | 48.21 | 34.88 | **99.33** | **96.55** | **95.47** | -| intfloat/multilingual-e5-large | 70.98 | 70.30 | **58.78** | **43.63** | 86.00 | 94.70 | 72.48 | +| OpenAI/text-embedding-3-large | **74.48** | 72.41 | 48.21 | 34.88 | **99.33** | 96.55 | **95.47** | +| pkshatech/GLuCoSE-base-ja-v2 | 73.36 | 69.79 | **67.29** | 41.86 | 90.29 | 95.11 | 75.80 | +| pkshatech/RoSEtta-base-ja | 73.21 | 65.96 | 65.33 | 36.73 | 95.54 | 94.08 | 81.63 | +| cl-nagoya/ruri-large | 73.02 | **76.68** | 61.74 | 38.03 | 87.12 | 96.58 | 77.97 | +| intfloat/multilingual-e5-large | 70.98 | 70.30 | 58.78 | **43.63** | 86.00 | 94.70 | 72.48 | +| cl-nagoya/ruri-base | 69.82 | 74.56 | 50.12 | 35.45 | 86.89 | 96.57 | 75.31 | +| cl-nagoya/ruri-small | 69.41 | 73.65 | 48.44 | 33.43 | 87.69 | **97.17** | 76.09 | | intfloat/multilingual-e5-base | 68.21 | 65.34 | 50.67 | 38.38 | 87.10 | 94.73 | 73.05 | | intfloat/multilingual-e5-small | 67.27 | 64.11 | 49.97 | 36.05 | 85.21 | 95.26 | 72.99 | | OpenAI/text-embedding-3-small | 66.39 | 64.02 | 33.94 | 20.03 | 98.47 | 91.70 | 90.17 | @@ -59,9 +69,14 @@ The summary shows the average scores within each task. ## STS | Model | Avg. | jsick
(spearman) | jsts
(spearman) | |:----------------------------------------------|:----------|:----------------------|:---------------------| -| cl-nagoya/sup-simcse-ja-large | **83.18** | **83.80** | 82.57 | -| OpenAI/text-embedding-3-large | 82.52 | 81.27 | **83.77** | +| cl-nagoya/sup-simcse-ja-large | **83.18** | 83.80 | 82.57 | +| cl-nagoya/ruri-large | 83.13 | 82.00 | **84.26** | +| pkshatech/GLuCoSE-base-ja-v2 | 82.96 | **84.96** | 80.96 | +| cl-nagoya/ruri-base | 82.87 | 82.32 | 83.43 | +| cl-nagoya/ruri-small | 82.79 | 83.44 | 82.13 | +| OpenAI/text-embedding-3-large | 82.52 | 81.27 | 83.77 | | cl-nagoya/sup-simcse-ja-base | 82.05 | 82.83 | 81.27 | +| pkshatech/RoSEtta-base-ja | 81.39 | 83.83 | 78.95 | | cl-nagoya/unsup-simcse-ja-large | 80.56 | 80.15 | 80.98 | | intfloat/multilingual-e5-small | 80.07 | 81.50 | 78.65 | | intfloat/multilingual-e5-base | 79.84 | 81.28 | 78.39 | @@ -85,10 +100,14 @@ The summary shows the average scores within each task. ## Classification | Model | Avg. | amazon_counterfactual
(macro_f1) | amazon_review
(macro_f1) | massive_intent
(macro_f1) | massive_scenario
(macro_f1) | |:----------------------------------------------|:----------|:--------------------------------------|:------------------------------|:-------------------------------|:---------------------------------| -| OpenAI/text-embedding-3-large | **77.58** | 77.90 | **60.44** | **80.91** | **91.08** | +| OpenAI/text-embedding-3-large | **77.58** | 77.90 | **60.44** | 80.91 | **91.08** | +| cl-nagoya/ruri-large | 77.43 | 80.81 | 56.80 | **82.56** | 89.56 | | pkshatech/GLuCoSE-base-ja | 76.82 | **82.44** | 58.07 | 78.85 | 87.94 | | oshizo/sbert-jsnli-luke-japanese-base-lite | 76.61 | 79.95 | 57.48 | 80.26 | 88.75 | +| cl-nagoya/ruri-small | 76.22 | 79.92 | 55.61 | 81.49 | 87.88 | +| cl-nagoya/ruri-base | 75.58 | 76.66 | 55.76 | 81.41 | 88.49 | | cl-nagoya/unsup-simcse-ja-large | 74.66 | 76.79 | 55.37 | 79.13 | 87.36 | +| pkshatech/GLuCoSE-base-ja-v2 | 74.21 | 74.92 | 55.31 | 79.79 | 86.84 | | MU-Kindai/Japanese-DiffCSE-BERT-base | 73.77 | 78.10 | 51.56 | 78.79 | 86.63 | | cl-nagoya/sup-simcse-ja-large | 73.73 | 73.21 | 54.76 | 79.23 | 87.72 | | MU-Kindai/Japanese-SimCSE-BERT-large-sup | 73.47 | 77.25 | 53.42 | 76.83 | 86.39 | @@ -101,6 +120,7 @@ The summary shows the average scores within each task. | intfloat/multilingual-e5-large | 72.89 | 70.66 | 56.54 | 75.78 | 88.59 | | MU-Kindai/Japanese-SimCSE-BERT-base-sup | 72.76 | 76.20 | 52.06 | 77.89 | 84.90 | | sentence-transformers/LaBSE | 72.66 | 73.61 | 51.70 | 76.99 | 88.35 | +| pkshatech/RoSEtta-base-ja | 72.41 | 70.05 | 52.64 | 79.84 | 87.10 | | sentence-transformers/stsb-xlm-r-multilingual | 71.84 | 75.65 | 51.32 | 74.28 | 86.10 | | pkshatech/simcse-ja-bert-base-clcmlp | 71.30 | 67.49 | 50.85 | 79.67 | 87.20 | | OpenAI/text-embedding-ada-002 | 69.75 | 64.42 | 53.13 | 74.57 | 86.89 | @@ -114,9 +134,14 @@ The summary shows the average scores within each task. | OpenAI/text-embedding-3-large | **93.58** | **93.58** | | OpenAI/text-embedding-ada-002 | 93.04 | 93.04 | | intfloat/multilingual-e5-small | 93.03 | 93.03 | +| pkshatech/GLuCoSE-base-ja-v2 | 93.01 | 93.01 | +| cl-nagoya/ruri-small | 93.00 | 93.00 | +| cl-nagoya/ruri-large | 92.99 | 92.99 | | intfloat/multilingual-e5-large | 92.96 | 92.96 | | OpenAI/text-embedding-3-small | 92.92 | 92.92 | +| cl-nagoya/ruri-base | 92.91 | 92.91 | | intfloat/multilingual-e5-base | 92.85 | 92.85 | +| pkshatech/RoSEtta-base-ja | 92.69 | 92.69 | | pkshatech/GLuCoSE-base-ja | 91.90 | 91.90 | | cl-nagoya/sup-simcse-ja-base | 91.83 | 91.83 | | sentence-transformers/LaBSE | 91.63 | 91.63 | @@ -137,13 +162,18 @@ The summary shows the average scores within each task. ## Clustering | Model | Avg. | livedoor_news
(v_measure_score) | mewsc16
(v_measure_score) | |:----------------------------------------------|:----------|:-------------------------------------|:-------------------------------| -| OpenAI/text-embedding-3-large | **53.32** | 57.09 | 49.55 | +| cl-nagoya/ruri-base | **54.16** | 54.27 | **54.04** | +| OpenAI/text-embedding-3-large | 53.32 | 57.09 | 49.55 | +| pkshatech/RoSEtta-base-ja | 53.23 | **58.62** | 47.85 | +| cl-nagoya/ruri-large | 51.82 | 51.39 | 52.25 | | cl-nagoya/sup-simcse-ja-base | 51.79 | 52.67 | 50.91 | -| intfloat/multilingual-e5-large | 51.24 | **57.13** | 45.34 | +| intfloat/multilingual-e5-large | 51.24 | 57.13 | 45.34 | +| cl-nagoya/ruri-small | 51.19 | 50.96 | 51.41 | | OpenAI/text-embedding-3-small | 51.06 | 54.57 | 47.55 | | cl-nagoya/sup-simcse-ja-large | 50.56 | 50.75 | 50.38 | -| oshizo/sbert-jsnli-luke-japanese-base-lite | 50.33 | 46.77 | **53.89** | +| oshizo/sbert-jsnli-luke-japanese-base-lite | 50.33 | 46.77 | 53.89 | | pkshatech/GLuCoSE-base-ja | 49.78 | 49.89 | 49.68 | +| pkshatech/GLuCoSE-base-ja-v2 | 48.65 | 51.52 | 45.78 | | cl-nagoya/unsup-simcse-ja-large | 48.41 | 50.90 | 45.92 | | OpenAI/text-embedding-ada-002 | 48.30 | 49.67 | 46.92 | | intfloat/multilingual-e5-base | 48.26 | 55.03 | 41.49 | @@ -171,18 +201,23 @@ The summary shows the average scores within each task. | pkshatech/simcse-ja-bert-base-clcmlp | 62.40 | 62.40 | | OpenAI/text-embedding-ada-002 | 62.40 | 62.40 | | MU-Kindai/Japanese-SimCSE-BERT-base-unsup | 62.38 | 62.38 | +| cl-nagoya/ruri-base | 62.38 | 62.38 | | oshizo/sbert-jsnli-luke-japanese-base-lite | 62.38 | 62.38 | | MU-Kindai/Japanese-DiffCSE-BERT-base | 62.38 | 62.38 | +| pkshatech/GLuCoSE-base-ja-v2 | 62.37 | 62.37 | | MU-Kindai/Japanese-SimCSE-BERT-base-sup | 62.37 | 62.37 | | MU-Kindai/Japanese-SimCSE-BERT-large-sup | 62.35 | 62.35 | | OpenAI/text-embedding-3-large | 62.35 | 62.35 | | MU-Kindai/Japanese-MixCSE-BERT-base | 62.33 | 62.33 | | sentence-transformers/LaBSE | 62.33 | 62.33 | | colorfulscoop/sbert-base-ja | 62.31 | 62.31 | +| cl-nagoya/ruri-large | 62.29 | 62.29 | | OpenAI/text-embedding-3-small | 62.27 | 62.27 | | MU-Kindai/Japanese-SimCSE-BERT-large-unsup | 62.27 | 62.27 | | intfloat/multilingual-e5-base | 62.26 | 62.26 | | sentence-transformers/stsb-xlm-r-multilingual | 62.20 | 62.20 | | intfloat/multilingual-e5-small | 62.19 | 62.19 | | intfloat/multilingual-e5-large | 62.15 | 62.15 | +| cl-nagoya/ruri-small | 62.11 | 62.11 | +| pkshatech/RoSEtta-base-ja | 61.74 | 61.74 | diff --git a/make_leaderboard.py b/make_leaderboard.py index ff3a330..0e43ccf 100644 --- a/make_leaderboard.py +++ b/make_leaderboard.py @@ -62,7 +62,14 @@ def format_score(score: float) -> str: table_list: list[list[str | float]] = [] for model_signature, dataset_scores in task_results.items(): model_scores = [dataset_scores[k] for k in dataset_keys] - average_score = sum(model_scores) / len(model_scores) + if task_name == SUMMARY_KEY: + scores_by_dataset = [] + for _task_name, _task_results in all_results.items(): + if _task_name != SUMMARY_KEY: + scores_by_dataset.extend(list(_task_results[model_signature].values())) + average_score = sum(scores_by_dataset) / len(scores_by_dataset) + else: + average_score = sum(model_scores) / len(model_scores) table_list.append([model_signature, average_score, *model_scores]) # sort by the average score @@ -97,7 +104,10 @@ def format_score(score: float) -> str: f.write(f"## {task_name}\n") if task_name == SUMMARY_KEY: - f.write("\nThe summary shows the average scores within each task.\n\n") + f.write( + "\nThe summary shows the average scores within each task. " + "The average score is the average of scores by dataset.\n\n" + ) f.write(markdown_table) f.write("\n\n") diff --git a/pyproject.toml b/pyproject.toml index 57e4f74..743b6bd 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -12,7 +12,7 @@ description = "The evaluation scripts for JMTEB (Japanese Massive Text Embedding name = "JMTEB" packages = [{from = "src", include = "jmteb"}] readme = "README.md" -version = "1.3.1" +version = "1.3.2" [tool.poetry.dependencies] python = ">=3.10,<4.0"