Update README.md
Browse files
README.md
CHANGED
|
@@ -19,111 +19,141 @@ tags:
|
|
| 19 |
|
| 20 |
ModernBERT multi-task fine-tuned on tasksource NLI tasks, including MNLI, ANLI, SICK, WANLI, doc-nli, LingNLI, FOLIO, FOL-NLI, LogicNLI, Label-NLI and all datasets in the below table).
|
| 21 |
This is the equivalent of an "instruct" version.
|
|
|
|
| 22 |
|
| 23 |
-
Test accuracy at 100k training steps. 215k steps version coming december 24th.
|
| 24 |
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
| glue/
|
| 29 |
-
| glue/
|
| 30 |
-
|
|
| 31 |
-
|
|
| 32 |
-
|
|
| 33 |
-
|
|
| 34 |
-
|
|
| 35 |
-
|
|
| 36 |
-
|
|
| 37 |
-
|
|
| 38 |
-
|
|
| 39 |
-
|
|
| 40 |
-
|
|
| 41 |
-
|
|
| 42 |
-
|
|
| 43 |
-
|
|
| 44 |
-
|
|
| 45 |
-
|
|
| 46 |
-
|
|
| 47 |
-
|
|
| 48 |
-
|
|
| 49 |
-
|
|
| 50 |
-
|
|
| 51 |
-
|
|
| 52 |
-
|
|
| 53 |
-
|
|
| 54 |
-
|
|
| 55 |
-
|
|
| 56 |
-
|
|
| 57 |
-
|
|
| 58 |
-
|
|
| 59 |
-
|
|
| 60 |
-
|
|
| 61 |
-
|
|
| 62 |
-
|
|
| 63 |
-
|
|
| 64 |
-
|
|
| 65 |
-
|
|
| 66 |
-
|
|
| 67 |
-
|
|
| 68 |
-
|
|
| 69 |
-
|
|
| 70 |
-
|
|
| 71 |
-
|
|
| 72 |
-
|
|
| 73 |
-
|
|
| 74 |
-
|
|
| 75 |
-
|
|
| 76 |
-
|
|
| 77 |
-
|
|
| 78 |
-
|
|
| 79 |
-
|
|
| 80 |
-
|
|
| 81 |
-
|
|
| 82 |
-
|
|
| 83 |
-
|
|
| 84 |
-
|
|
| 85 |
-
|
|
| 86 |
-
|
|
| 87 |
-
|
|
| 88 |
-
|
|
| 89 |
-
|
|
| 90 |
-
|
|
| 91 |
-
|
|
| 92 |
-
|
|
| 93 |
-
|
|
| 94 |
-
|
|
| 95 |
-
|
|
| 96 |
-
|
|
| 97 |
-
|
|
| 98 |
-
|
|
| 99 |
-
|
|
| 100 |
-
|
|
| 101 |
-
|
|
| 102 |
-
|
|
| 103 |
-
|
|
| 104 |
-
|
|
| 105 |
-
|
|
| 106 |
-
|
|
| 107 |
-
|
|
| 108 |
-
|
|
| 109 |
-
|
|
| 110 |
-
|
|
| 111 |
-
|
|
| 112 |
-
|
|
| 113 |
-
|
|
| 114 |
-
|
|
| 115 |
-
|
|
| 116 |
-
|
|
| 117 |
-
|
|
| 118 |
-
|
|
| 119 |
-
|
|
| 120 |
-
|
|
| 121 |
-
|
|
| 122 |
-
|
|
| 123 |
-
|
|
| 124 |
-
|
|
| 125 |
-
|
|
| 126 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 127 |
|
| 128 |
# Usage
|
| 129 |
|
|
|
|
| 19 |
|
| 20 |
ModernBERT multi-task fine-tuned on tasksource NLI tasks, including MNLI, ANLI, SICK, WANLI, doc-nli, LingNLI, FOLIO, FOL-NLI, LogicNLI, Label-NLI and all datasets in the below table).
|
| 21 |
This is the equivalent of an "instruct" version.
|
| 22 |
+
The model was trained for 200k steps on an Nvidia A30 GPU.
|
| 23 |
|
|
|
|
| 24 |
|
| 25 |
+
|
| 26 |
+
| test_name | test_accuracy |
|
| 27 |
+
|:--------------------------------------|----------------:|
|
| 28 |
+
| glue/mnli | 0.87 |
|
| 29 |
+
| glue/qnli | 0.93 |
|
| 30 |
+
| glue/rte | 0.85 |
|
| 31 |
+
| glue/mrpc | 0.87 |
|
| 32 |
+
| glue/qqp | 0.9 |
|
| 33 |
+
| glue/cola | 0.86 |
|
| 34 |
+
| glue/sst2 | 0.96 |
|
| 35 |
+
| super_glue/boolq | 0.64 |
|
| 36 |
+
| super_glue/cb | 0.89 |
|
| 37 |
+
| super_glue/multirc | 0.82 |
|
| 38 |
+
| super_glue/wic | 0.67 |
|
| 39 |
+
| super_glue/axg | 0.89 |
|
| 40 |
+
| anli/a1 | 0.66 |
|
| 41 |
+
| anli/a2 | 0.49 |
|
| 42 |
+
| anli/a3 | 0.44 |
|
| 43 |
+
| sick/label | 0.93 |
|
| 44 |
+
| sick/entailment_AB | 0.91 |
|
| 45 |
+
| snli | 0.83 |
|
| 46 |
+
| scitail/snli_format | 0.94 |
|
| 47 |
+
| hans | 1 |
|
| 48 |
+
| WANLI | 0.74 |
|
| 49 |
+
| recast/recast_ner | 0.87 |
|
| 50 |
+
| recast/recast_sentiment | 0.99 |
|
| 51 |
+
| recast/recast_verbnet | 0.88 |
|
| 52 |
+
| recast/recast_megaveridicality | 0.88 |
|
| 53 |
+
| recast/recast_verbcorner | 0.94 |
|
| 54 |
+
| recast/recast_kg_relations | 0.91 |
|
| 55 |
+
| recast/recast_factuality | 0.94 |
|
| 56 |
+
| recast/recast_puns | 0.96 |
|
| 57 |
+
| probability_words_nli/reasoning_1hop | 0.99 |
|
| 58 |
+
| probability_words_nli/usnli | 0.72 |
|
| 59 |
+
| probability_words_nli/reasoning_2hop | 0.98 |
|
| 60 |
+
| nan-nli | 0.85 |
|
| 61 |
+
| nli_fever | 0.78 |
|
| 62 |
+
| breaking_nli | 0.99 |
|
| 63 |
+
| conj_nli | 0.74 |
|
| 64 |
+
| fracas | 0.86 |
|
| 65 |
+
| dialogue_nli | 0.93 |
|
| 66 |
+
| mpe | 0.74 |
|
| 67 |
+
| dnc | 0.92 |
|
| 68 |
+
| recast_white/fnplus | 0.82 |
|
| 69 |
+
| recast_white/sprl | 0.9 |
|
| 70 |
+
| recast_white/dpr | 0.68 |
|
| 71 |
+
| robust_nli/IS_CS | 0.79 |
|
| 72 |
+
| robust_nli/LI_LI | 0.99 |
|
| 73 |
+
| robust_nli/ST_WO | 0.85 |
|
| 74 |
+
| robust_nli/PI_SP | 0.74 |
|
| 75 |
+
| robust_nli/PI_CD | 0.8 |
|
| 76 |
+
| robust_nli/ST_SE | 0.81 |
|
| 77 |
+
| robust_nli/ST_NE | 0.86 |
|
| 78 |
+
| robust_nli/ST_LM | 0.87 |
|
| 79 |
+
| robust_nli_is_sd | 1 |
|
| 80 |
+
| robust_nli_li_ts | 0.89 |
|
| 81 |
+
| add_one_rte | 0.94 |
|
| 82 |
+
| paws/labeled_final | 0.95 |
|
| 83 |
+
| pragmeval/pdtb | 0.64 |
|
| 84 |
+
| lex_glue/scotus | 0.55 |
|
| 85 |
+
| lex_glue/ledgar | 0.8 |
|
| 86 |
+
| dynasent/dynabench.dynasent.r1.all/r1 | 0.81 |
|
| 87 |
+
| dynasent/dynabench.dynasent.r2.all/r2 | 0.75 |
|
| 88 |
+
| cycic_classification | 0.9 |
|
| 89 |
+
| lingnli | 0.84 |
|
| 90 |
+
| monotonicity-entailment | 0.97 |
|
| 91 |
+
| scinli | 0.8 |
|
| 92 |
+
| naturallogic | 0.96 |
|
| 93 |
+
| dynahate | 0.78 |
|
| 94 |
+
| syntactic-augmentation-nli | 0.92 |
|
| 95 |
+
| autotnli | 0.94 |
|
| 96 |
+
| defeasible-nli/atomic | 0.81 |
|
| 97 |
+
| defeasible-nli/snli | 0.78 |
|
| 98 |
+
| help-nli | 0.96 |
|
| 99 |
+
| nli-veridicality-transitivity | 0.98 |
|
| 100 |
+
| lonli | 0.97 |
|
| 101 |
+
| dadc-limit-nli | 0.69 |
|
| 102 |
+
| folio | 0.66 |
|
| 103 |
+
| tomi-nli | 0.48 |
|
| 104 |
+
| puzzte | 0.6 |
|
| 105 |
+
| temporal-nli | 0.92 |
|
| 106 |
+
| counterfactually-augmented-snli | 0.79 |
|
| 107 |
+
| cnli | 0.87 |
|
| 108 |
+
| boolq-natural-perturbations | 0.66 |
|
| 109 |
+
| equate | 0.63 |
|
| 110 |
+
| logiqa-2.0-nli | 0.52 |
|
| 111 |
+
| mindgames | 0.96 |
|
| 112 |
+
| ConTRoL-nli | 0.67 |
|
| 113 |
+
| logical-fallacy | 0.37 |
|
| 114 |
+
| cladder | 0.87 |
|
| 115 |
+
| conceptrules_v2 | 1 |
|
| 116 |
+
| zero-shot-label-nli | 0.82 |
|
| 117 |
+
| scone | 0.98 |
|
| 118 |
+
| monli | 1 |
|
| 119 |
+
| SpaceNLI | 1 |
|
| 120 |
+
| propsegment/nli | 0.88 |
|
| 121 |
+
| FLD.v2/default | 0.91 |
|
| 122 |
+
| FLD.v2/star | 0.76 |
|
| 123 |
+
| SDOH-NLI | 0.98 |
|
| 124 |
+
| scifact_entailment | 0.84 |
|
| 125 |
+
| AdjectiveScaleProbe-nli | 0.99 |
|
| 126 |
+
| resnli | 1 |
|
| 127 |
+
| semantic_fragments_nli | 0.99 |
|
| 128 |
+
| dataset_train_nli | 0.94 |
|
| 129 |
+
| nlgraph | 0.94 |
|
| 130 |
+
| ruletaker | 0.99 |
|
| 131 |
+
| PARARULE-Plus | 1 |
|
| 132 |
+
| logical-entailment | 0.86 |
|
| 133 |
+
| nope | 0.44 |
|
| 134 |
+
| LogicNLI | 0.86 |
|
| 135 |
+
| contract-nli/contractnli_a/seg | 0.87 |
|
| 136 |
+
| contract-nli/contractnli_b/full | 0.79 |
|
| 137 |
+
| nli4ct_semeval2024 | 0.67 |
|
| 138 |
+
| biosift-nli | 0.92 |
|
| 139 |
+
| SIGA-nli | 0.53 |
|
| 140 |
+
| FOL-nli | 0.8 |
|
| 141 |
+
| doc-nli | 0.77 |
|
| 142 |
+
| mctest-nli | 0.87 |
|
| 143 |
+
| natural-language-satisfiability | 0.9 |
|
| 144 |
+
| idioms-nli | 0.81 |
|
| 145 |
+
| lifecycle-entailment | 0.78 |
|
| 146 |
+
| MSciNLI | 0.85 |
|
| 147 |
+
| hover-3way/nli | 0.88 |
|
| 148 |
+
| seahorse_summarization_evaluation | 0.73 |
|
| 149 |
+
| missing-item-prediction/contrastive | 0.79 |
|
| 150 |
+
| Pol_NLI | 0.89 |
|
| 151 |
+
| synthetic-retrieval-NLI/count | 0.64 |
|
| 152 |
+
| synthetic-retrieval-NLI/position | 0.89 |
|
| 153 |
+
| synthetic-retrieval-NLI/binary | 0.91 |
|
| 154 |
+
| babi_nli | 0.97 |
|
| 155 |
+
| gen_debiased_nli | 0.91 |
|
| 156 |
+
|
| 157 |
|
| 158 |
# Usage
|
| 159 |
|