{ "selected": "head_lora_rlcd", "seed_run": "artifacts/runs/e4b-v142", "method": "Proper-scoring REINFORCE with Gaussian logit exploration", "gates": { "seed": { "eligible": true, "score": 0.0 }, "head_rlcd": { "eligible": true, "preservation_gate": true, "probability_gate": true, "score": 0.013723702557236433, "mean_nll": 0.2912874264480955, "mean_brier": 0.17143807150110177, "mean_ece": 0.026519458902999775 }, "head_lora_rlcd": { "eligible": true, "preservation_gate": true, "probability_gate": true, "score": 0.02410640078446494, "mean_nll": 0.2851054899756657, "mean_brier": 0.1656363697259328, "mean_ece": 0.02280542785381634 } }, "results": { "seed": { "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.097443699836731, "groups": { "choice:2": { "temperature": 0.7726892229265028, "raw_temperature": 0.4564978778362274, "n": 80, "shrinkage_weight": 0.4 }, "choice:3-5": { "temperature": 1.0735645605398847, "raw_temperature": 1.0707285404205322, "n": 998, "shrinkage_weight": 0.8926654740608229 }, "choice:6-10": { "temperature": 1.0213380776161125, "raw_temperature": 0.9631407260894775, "n": 147, "shrinkage_weight": 0.550561797752809 }, "noul:*": { "temperature": 1.18076753369684, "raw_temperature": 1.1976659297943115, "n": 618, "shrinkage_weight": 0.8373983739837398 }, "score:*": { "temperature": 1.1118863993359898, "raw_temperature": 1.1230534315109253, "n": 157, "shrinkage_weight": 0.5667870036101083 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "sets": { "validation": { "n": 2000, "accuracy": 0.8635, "nll": 0.33747788503553516, "brier": 0.1907344097228753, "ece": 0.016734494030475625, "mean_confidence": 0.8650541846752167, "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.097443699836731, "groups": { "choice:2": { "temperature": 0.7726892229265028, "raw_temperature": 0.4564978778362274, "n": 80, "shrinkage_weight": 0.4 }, "choice:3-5": { "temperature": 1.0735645605398847, "raw_temperature": 1.0707285404205322, "n": 998, "shrinkage_weight": 0.8926654740608229 }, "choice:6-10": { "temperature": 1.0213380776161125, "raw_temperature": 0.9631407260894775, "n": 147, "shrinkage_weight": 0.550561797752809 }, "noul:*": { "temperature": 1.18076753369684, "raw_temperature": 1.1976659297943115, "n": 618, "shrinkage_weight": 0.8373983739837398 }, "score:*": { "temperature": 1.1118863993359898, "raw_temperature": 1.1230534315109253, "n": 157, "shrinkage_weight": 0.5667870036101083 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "by_dataset": { "PolyAI/banking77": { "n": 136, "accuracy": 0.9338235294117647, "nll": 0.18829089664733842, "brier": 0.10488018020936762, "ece": 0.05482878689380254, "mean_confidence": 0.9316614732584533 }, "Rowan/hellaswag": { "n": 153, "accuracy": 0.934640522875817, "nll": 0.20100579900196738, "brier": 0.11092448101729271, "ece": 0.050551106921987575, "mean_confidence": 0.9167824553118812 }, "alisawuffles/WANLI": { "n": 158, "accuracy": 0.7911392405063291, "nll": 0.4983649758763517, "brier": 0.2937829899183501, "ece": 0.06604516147812708, "mean_confidence": 0.7893789567147629 }, "cais/mmlu": { "n": 154, "accuracy": 0.8766233766233766, "nll": 0.4075193296471239, "brier": 0.19814945253592872, "ece": 0.07114563940407395, "mean_confidence": 0.8730270781300284 }, "google/boolq": { "n": 153, "accuracy": 0.869281045751634, "nll": 0.3304804692638724, "brier": 0.19049125629416552, "ece": 0.04868968210968311, "mean_confidence": 0.8886790147014693 }, "google/boolq-negation": { "n": 153, "accuracy": 0.8627450980392157, "nll": 0.29803780987980516, "brier": 0.1722142930618895, "ece": 0.06692373012405595, "mean_confidence": 0.8571316631011714 }, "nvidia/HelpSteer2": { "n": 157, "accuracy": 0.535031847133758, "nll": 0.9802866589491535, "brier": 0.5608620777678718, "ece": 0.1083361166677657, "mean_confidence": 0.5841244707821281 }, "nvidia/HelpSteer2-threshold": { "n": 158, "accuracy": 0.879746835443038, "nll": 0.3300814368538087, "brier": 0.19343432728677312, "ece": 0.06685320712343043, "mean_confidence": 0.832719200019595 }, "nyu-mll/multi_nli": { "n": 157, "accuracy": 0.8662420382165605, "nll": 0.3357255889909567, "brier": 0.19144842998859854, "ece": 0.07937940025025873, "mean_confidence": 0.84814400467903 }, "product-science/xlam-function-calling-60k-raw": { "n": 154, "accuracy": 1.0, "nll": 0.002737168537177378, "brier": 0.0005628128288157573, "ece": 0.0025677526151979472, "mean_confidence": 0.9974322473848021 }, "product-science/xlam-hard": { "n": 154, "accuracy": 0.961038961038961, "nll": 0.105508735974156, "brier": 0.06104789821045359, "ece": 0.03465073526679697, "mean_confidence": 0.9694648733386746 }, "stanfordnlp/snli": { "n": 159, "accuracy": 0.8553459119496856, "nll": 0.3588936702368793, "brier": 0.2055985772370396, "ece": 0.053331098466549334, "mean_confidence": 0.8946507621861104 }, "synthetic/nli-haystack": { "n": 154, "accuracy": 0.8766233766233766, "nll": 0.31448460003064777, "brier": 0.1750220485890215, "ece": 0.058161640709096796, "mean_confidence": 0.8783779376512998 } }, "by_kind": { "choice": { "n": 1225, "accuracy": 0.8906122448979592, "nll": 0.2910091881993051, "brier": 0.16159634624408029, "ece": 0.018170039288851744, "mean_confidence": 0.8901425058257824 }, "noul": { "n": 618, "accuracy": 0.8932038834951457, "nll": 0.2662856133849587, "brier": 0.15446270078834373, "ece": 0.022636081794319064, "mean_confidence": 0.8866929737881164 }, "score": { "n": 157, "accuracy": 0.535031847133758, "nll": 0.9802866589491535, "brier": 0.5608620777678718, "ece": 0.1083361166677657, "mean_confidence": 0.5841244707821281 } }, "by_transform": { "agent.invocation.correct.v1": { "n": 59, "accuracy": 0.9322033898305084, "nll": 0.19396388833805667, "brier": 0.10943394663022984, "ece": 0.06162244788670942, "mean_confidence": 0.9474405292737282 }, "agent.invocation.corrupt_argument.v1": { "n": 47, "accuracy": 1.0, "nll": 0.007110317709528782, "brier": 0.00041410416924144496, "ece": 0.007002635205045669, "mean_confidence": 0.9929973647949544 }, "agent.invocation.wrong_tool.v1": { "n": 48, "accuracy": 0.9583333333333334, "nll": 0.09313106241097557, "brier": 0.06094397035983216, "ece": 0.048730698724587775, "mean_confidence": 0.9734942317008972 }, "commonsense.choice.v2": { "n": 153, "accuracy": 0.934640522875817, "nll": 0.20100579900196738, "brier": 0.11092448101729271, "ece": 0.050551106921987575, "mean_confidence": 0.9167824553118812 }, "intent.choice.v3": { "n": 136, "accuracy": 0.9338235294117647, "nll": 0.18829089664733842, "brier": 0.10488018020936762, "ece": 0.05482878689380254, "mean_confidence": 0.9316614732584533 }, "knowledge.choice.v1": { "n": 154, "accuracy": 0.8766233766233766, "nll": 0.4075193296471239, "brier": 0.19814945253592872, "ece": 0.07114563940407395, "mean_confidence": 0.8730270781300284 }, "nli.choice.v4": { "n": 157, "accuracy": 0.8662420382165605, "nll": 0.3357255889909567, "brier": 0.19144842998859854, "ece": 0.07937940025025873, "mean_confidence": 0.84814400467903 }, "nli.hard.v4": { "n": 158, "accuracy": 0.7911392405063291, "nll": 0.4983649758763517, "brier": 0.2937829899183501, "ece": 0.06604516147812708, "mean_confidence": 0.7893789567147629 }, "nli.haystack.absent.v1": { "n": 9, "accuracy": 1.0, "nll": 0.05771497719817691, "brier": 0.012067315275392806, "ece": 0.05367900265587699, "mean_confidence": 0.946320997344123 }, "nli.haystack.present.v1": { "n": 145, "accuracy": 0.8689655172413793, "nll": 0.33042202489611144, "brier": 0.18513648031193639, "ece": 0.060484279846322944, "mean_confidence": 0.8741607822220901 }, "nli.snli.v1": { "n": 159, "accuracy": 0.8553459119496856, "nll": 0.3588936702368793, "brier": 0.2055985772370396, "ece": 0.053331098466549334, "mean_confidence": 0.8946507621861104 }, "noul.negated_answer.v1": { "n": 153, "accuracy": 0.8627450980392157, "nll": 0.29803780987980516, "brier": 0.1722142930618895, "ece": 0.06692373012405595, "mean_confidence": 0.8571316631011714 }, "noul.v3": { "n": 153, "accuracy": 0.869281045751634, "nll": 0.3304804692638724, "brier": 0.19049125629416552, "ece": 0.04868968210968311, "mean_confidence": 0.8886790147014693 }, "score.coherence.threshold.v1": { "n": 35, "accuracy": 0.9428571428571428, "nll": 0.30217631511124116, "brier": 0.16118863898768074, "ece": 0.13178183862141202, "mean_confidence": 0.8401974422591073 }, "score.coherence.v3": { "n": 32, "accuracy": 0.5625, "nll": 0.9295596550218761, "brier": 0.5722244664793834, "ece": 0.20393101125955582, "mean_confidence": 0.694971589371562 }, "score.complexity.threshold.v1": { "n": 30, "accuracy": 0.9333333333333333, "nll": 0.27167421360500155, "brier": 0.14001424941464696, "ece": 0.082729438940684, "mean_confidence": 0.9072748144467672 }, "score.complexity.v3": { "n": 27, "accuracy": 0.5925925925925926, "nll": 0.7247205337992421, "brier": 0.4520837900539239, "ece": 0.1485392583741082, "mean_confidence": 0.6281530835010387 }, "score.correctness.threshold.v1": { "n": 30, "accuracy": 0.7333333333333333, "nll": 0.44957111608237027, "brier": 0.30260899337008595, "ece": 0.13217389583587646, "mean_confidence": 0.7230223735173543 }, "score.correctness.v3": { "n": 30, "accuracy": 0.4, "nll": 1.2207635447382927, "brier": 0.6627602388461431, "ece": 0.19082429458697636, "mean_confidence": 0.4697169209520022 }, "score.helpfulness.threshold.v1": { "n": 30, "accuracy": 0.8, "nll": 0.41629887875169513, "brier": 0.27174266742852826, "ece": 0.1563175916671753, "mean_confidence": 0.7734385410944621 }, "score.helpfulness.v3": { "n": 36, "accuracy": 0.5277777777777778, "nll": 1.0197049867775705, "brier": 0.5488159826232327, "ece": 0.2696710663537184, "mean_confidence": 0.44966551951236194 }, "score.verbosity.threshold.v1": { "n": 33, "accuracy": 0.9696969696969697, "nll": 0.2257687800253431, "brier": 0.10575891031343913, "ece": 0.07801289088798288, "mean_confidence": 0.9106257955233256 }, "score.verbosity.v3": { "n": 32, "accuracy": 0.59375, "nll": 0.9768548817373812, "brier": 0.5593037003418431, "ece": 0.16308739501982927, "mean_confidence": 0.6946516083553433 }, "tool.choice.single.v3": { "n": 154, "accuracy": 1.0, "nll": 0.002737168537177378, "brier": 0.0005628128288157573, "ece": 0.0025677526151979472, "mean_confidence": 0.9974322473848021 } }, "by_option_bucket": { "2": { "n": 698, "accuracy": 0.9054441260744985, "nll": 0.23576630212333755, "brier": 0.13675923951146768, "ece": 0.020041158684345223, "mean_confidence": 0.8996789226559309 }, "3-5": { "n": 1155, "accuracy": 0.8285714285714286, "nll": 0.41972496986994157, "brier": 0.23527893114692047, "ece": 0.017945458233614507, "mean_confidence": 0.8350021731027794 }, "6-10": { "n": 147, "accuracy": 0.9387755102040817, "nll": 0.17420782985849112, "brier": 0.09703200538811604, "ece": 0.05073268012124669, "mean_confidence": 0.9367685129447859 } }, "by_truncation": { "not_truncated": { "n": 1999, "accuracy": 0.863431715857929, "nll": 0.3375232138199196, "brier": 0.19078194724154862, "ece": 0.016852297354722528, "mean_confidence": 0.865096109905739 }, "truncated": { "n": 1, "accuracy": 1.0, "nll": 0.2468656450510025, "brier": 0.09570690989494324, "ece": 0.2187543511390686, "mean_confidence": 0.7812456488609314 } }, "by_dataset_truncation": { "PolyAI/banking77|not_truncated": { "n": 136, "accuracy": 0.9338235294117647, "nll": 0.18829089664733842, "brier": 0.10488018020936762, "ece": 0.05482878689380254, "mean_confidence": 0.9316614732584533 }, "Rowan/hellaswag|not_truncated": { "n": 153, "accuracy": 0.934640522875817, "nll": 0.20100579900196738, "brier": 0.11092448101729271, "ece": 0.050551106921987575, "mean_confidence": 0.9167824553118812 }, "alisawuffles/WANLI|not_truncated": { "n": 158, "accuracy": 0.7911392405063291, "nll": 0.4983649758763517, "brier": 0.2937829899183501, "ece": 0.06604516147812708, "mean_confidence": 0.7893789567147629 }, "cais/mmlu|not_truncated": { "n": 154, "accuracy": 0.8766233766233766, "nll": 0.4075193296471239, "brier": 0.19814945253592872, "ece": 0.07114563940407395, "mean_confidence": 0.8730270781300284 }, "google/boolq-negation|not_truncated": { "n": 153, "accuracy": 0.8627450980392157, "nll": 0.29803780987980516, "brier": 0.1722142930618895, "ece": 0.06692373012405595, "mean_confidence": 0.8571316631011714 }, "google/boolq|not_truncated": { "n": 153, "accuracy": 0.869281045751634, "nll": 0.3304804692638724, "brier": 0.19049125629416552, "ece": 0.04868968210968311, "mean_confidence": 0.8886790147014693 }, "nvidia/HelpSteer2-threshold|not_truncated": { "n": 157, "accuracy": 0.8789808917197452, "nll": 0.3306114737442724, "brier": 0.19405679491347266, "ece": 0.06588568391313974, "mean_confidence": 0.8330470570333445 }, "nvidia/HelpSteer2-threshold|truncated": { "n": 1, "accuracy": 1.0, "nll": 0.2468656450510025, "brier": 0.09570690989494324, "ece": 0.2187543511390686, "mean_confidence": 0.7812456488609314 }, "nvidia/HelpSteer2|not_truncated": { "n": 157, "accuracy": 0.535031847133758, "nll": 0.9802866589491535, "brier": 0.5608620777678718, "ece": 0.1083361166677657, "mean_confidence": 0.5841244707821281 }, "nyu-mll/multi_nli|not_truncated": { "n": 157, "accuracy": 0.8662420382165605, "nll": 0.3357255889909567, "brier": 0.19144842998859854, "ece": 0.07937940025025873, "mean_confidence": 0.84814400467903 }, "product-science/xlam-function-calling-60k-raw|not_truncated": { "n": 154, "accuracy": 1.0, "nll": 0.002737168537177378, "brier": 0.0005628128288157573, "ece": 0.0025677526151979472, "mean_confidence": 0.9974322473848021 }, "product-science/xlam-hard|not_truncated": { "n": 154, "accuracy": 0.961038961038961, "nll": 0.105508735974156, "brier": 0.06104789821045359, "ece": 0.03465073526679697, "mean_confidence": 0.9694648733386746 }, "stanfordnlp/snli|not_truncated": { "n": 159, "accuracy": 0.8553459119496856, "nll": 0.3588936702368793, "brier": 0.2055985772370396, "ece": 0.053331098466549334, "mean_confidence": 0.8946507621861104 }, "synthetic/nli-haystack|not_truncated": { "n": 154, "accuracy": 0.8766233766233766, "nll": 0.31448460003064777, "brier": 0.1750220485890215, "ece": 0.058161640709096796, "mean_confidence": 0.8783779376512998 } }, "by_kind_truncation": { "choice|not_truncated": { "n": 1225, "accuracy": 0.8906122448979592, "nll": 0.2910091881993051, "brier": 0.16159634624408029, "ece": 0.018170039288851744, "mean_confidence": 0.8901425058257824 }, "noul|not_truncated": { "n": 617, "accuracy": 0.893030794165316, "nll": 0.26631708821208017, "brier": 0.1545579289745567, "ece": 0.02302731426260657, "mean_confidence": 0.886863877070008 }, "noul|truncated": { "n": 1, "accuracy": 1.0, "nll": 0.2468656450510025, "brier": 0.09570690989494324, "ece": 0.2187543511390686, "mean_confidence": 0.7812456488609314 }, "score|not_truncated": { "n": 157, "accuracy": 0.535031847133758, "nll": 0.9802866589491535, "brier": 0.5608620777678718, "ece": 0.1083361166677657, "mean_confidence": 0.5841244707821281 } }, "nli_confusion": { "n": 628, "labels": [ "entailment", "neutral", "contradiction" ], "matrix": { "entailment": { "entailment": 173, "neutral": 23, "contradiction": 3 }, "neutral": { "entailment": 21, "neutral": 203, "contradiction": 15 }, "contradiction": { "entailment": 2, "neutral": 32, "contradiction": 156 } } } }, "schema": { "n": 2000, "accuracy": 0.758, "nll": 0.5204962961420965, "brier": 0.32018979337543463, "ece": 0.07037245657294991, "mean_confidence": 0.8196629058793187, "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.097443699836731, "groups": { "choice:2": { "temperature": 0.7726892229265028, "raw_temperature": 0.4564978778362274, "n": 80, "shrinkage_weight": 0.4 }, "choice:3-5": { "temperature": 1.0735645605398847, "raw_temperature": 1.0707285404205322, "n": 998, "shrinkage_weight": 0.8926654740608229 }, "choice:6-10": { "temperature": 1.0213380776161125, "raw_temperature": 0.9631407260894775, "n": 147, "shrinkage_weight": 0.550561797752809 }, "noul:*": { "temperature": 1.18076753369684, "raw_temperature": 1.1976659297943115, "n": 618, "shrinkage_weight": 0.8373983739837398 }, "score:*": { "temperature": 1.1118863993359898, "raw_temperature": 1.1230534315109253, "n": 157, "shrinkage_weight": 0.5667870036101083 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "by_dataset": { "schema-firewall/PolyAI/banking77": { "n": 46, "accuracy": 0.8043478260869565, "nll": 0.35612511238073435, "brier": 0.22549919875508087, "ece": 0.10526247128196384, "mean_confidence": 0.8808949822964875 }, "schema-firewall/Rowan/hellaswag": { "n": 130, "accuracy": 0.9, "nll": 0.20414897868758125, "brier": 0.11925340885983747, "ece": 0.05373616493665252, "mean_confidence": 0.9159407689021184 }, "schema-firewall/alisawuffles/WANLI": { "n": 209, "accuracy": 0.7655502392344498, "nll": 0.5446630402214148, "brier": 0.32441898222513327, "ece": 0.059890149027536935, "mean_confidence": 0.7769627089135385 }, "schema-firewall/cais/mmlu": { "n": 134, "accuracy": 0.917910447761194, "nll": 0.22831852445149217, "brier": 0.11953794011554492, "ece": 0.049322252620511986, "mean_confidence": 0.909431281596867 }, "schema-firewall/google/boolq": { "n": 36, "accuracy": 0.8888888888888888, "nll": 0.44322766041700057, "brier": 0.22839449674166368, "ece": 0.1561499618821674, "mean_confidence": 0.8758690738015704 }, "schema-firewall/google/boolq-negation": { "n": 60, "accuracy": 0.5166666666666667, "nll": 1.2234002001583577, "brier": 0.7434715281813321, "ece": 0.34772479931513467, "mean_confidence": 0.8540057341257731 }, "schema-firewall/nvidia/HelpSteer2": { "n": 270, "accuracy": 0.6296296296296297, "nll": 0.8983366615242429, "brier": 0.496752599419819, "ece": 0.07374763350795817, "mean_confidence": 0.5930154732531971 }, "schema-firewall/nvidia/HelpSteer2-threshold": { "n": 260, "accuracy": 0.5230769230769231, "nll": 0.8322646450712524, "brier": 0.6063095804116249, "ece": 0.2711401492357254, "mean_confidence": 0.7942170723126485 }, "schema-firewall/nyu-mll/multi_nli": { "n": 200, "accuracy": 0.865, "nll": 0.41195048167137427, "brier": 0.21928038816042317, "ece": 0.061470215469598743, "mean_confidence": 0.8334396223723889 }, "schema-firewall/product-science/xlam-function-calling-60k-raw": { "n": 81, "accuracy": 1.0, "nll": 0.0050883564066314295, "brier": 0.001193973693776232, "ece": 0.004606815032017043, "mean_confidence": 0.9953931849679829 }, "schema-firewall/product-science/xlam-hard": { "n": 211, "accuracy": 0.6303317535545023, "nll": 0.5777726572084735, "brier": 0.400591680919437, "ece": 0.21581395203468354, "mean_confidence": 0.846145705589186 }, "schema-firewall/stanfordnlp/snli": { "n": 208, "accuracy": 0.8846153846153846, "nll": 0.26919846923891994, "brier": 0.15986069018861843, "ece": 0.04227677503457438, "mean_confidence": 0.8858483952398484 }, "schema-firewall/synthetic/nli-haystack": { "n": 155, "accuracy": 0.896774193548387, "nll": 0.28797892007766474, "brier": 0.15715603314421323, "ece": 0.06688037726186936, "mean_confidence": 0.8773749347656004 } }, "by_kind": { "choice": { "n": 1163, "accuracy": 0.8718830610490111, "nll": 0.3188156900246091, "brier": 0.1816513806989808, "ece": 0.026625588907830587, "mean_confidence": 0.8696533281128671 }, "noul": { "n": 567, "accuracy": 0.5855379188712522, "nll": 0.754248758684351, "brier": 0.5202748309605, "ece": 0.23951466205469096, "mean_confidence": 0.8250525809259431 }, "score": { "n": 270, "accuracy": 0.6296296296296297, "nll": 0.8983366615242429, "brier": 0.496752599419819, "ece": 0.07374763350795817, "mean_confidence": 0.5930154732531971 } }, "by_transform": { "schema.firewall.distractor_drop.v1": { "n": 179, "accuracy": 0.9162011173184358, "nll": 0.17996553745532837, "brier": 0.10203871893360637, "ece": 0.041837152822057584, "mean_confidence": 0.9357914165411582 }, "schema.firewall.instruction.v1": { "n": 1395, "accuracy": 0.8523297491039427, "nll": 0.3699874161727043, "brier": 0.21063317768835207, "ece": 0.02584742985959546, "mean_confidence": 0.8404050774151279 }, "schema.firewall.noul_inversion.v1": { "n": 292, "accuracy": 0.2808219178082192, "nll": 1.247713702195005, "brier": 0.8843806009206997, "ece": 0.47954579167170064, "mean_confidence": 0.7603677094799198 }, "schema.firewall.score_scale.v1": { "n": 134, "accuracy": 0.6044776119402985, "nll": 0.9575665259539191, "brier": 0.5227032665496887, "ece": 0.07600586627846333, "mean_confidence": 0.5778111495188812 } }, "by_option_bucket": { "2": { "n": 594, "accuracy": 0.6043771043771043, "nll": 0.7199648861602912, "brier": 0.4966259750298848, "ece": 0.22862747000524092, "mean_confidence": 0.8330045743823453 }, "3-5": { "n": 1375, "accuracy": 0.8218181818181818, "nll": 0.44010872633975223, "brier": 0.24742196818368095, "ece": 0.032660349141467676, "mean_confidence": 0.8119605057998137 }, "6-10": { "n": 31, "accuracy": 0.8709677419354839, "nll": 0.263998425413569, "brier": 0.16705004292117379, "ece": 0.09422032871553974, "mean_confidence": 0.905658035509048 } }, "by_truncation": { "not_truncated": { "n": 1998, "accuracy": 0.7587587587587588, "nll": 0.5200072703881736, "brier": 0.3198700578515258, "ece": 0.0704587096193651, "mean_confidence": 0.8200187474653289 }, "truncated": { "n": 2, "accuracy": 0.0, "nll": 1.0090330243110657, "brier": 0.6396055817604065, "ece": 0.4641771614551544, "mean_confidence": 0.4641771614551544 } }, "by_dataset_truncation": { "schema-firewall/PolyAI/banking77|not_truncated": { "n": 46, "accuracy": 0.8043478260869565, "nll": 0.35612511238073435, "brier": 0.22549919875508087, "ece": 0.10526247128196384, "mean_confidence": 0.8808949822964875 }, "schema-firewall/Rowan/hellaswag|not_truncated": { "n": 130, "accuracy": 0.9, "nll": 0.20414897868758125, "brier": 0.11925340885983747, "ece": 0.05373616493665252, "mean_confidence": 0.9159407689021184 }, "schema-firewall/alisawuffles/WANLI|not_truncated": { "n": 209, "accuracy": 0.7655502392344498, "nll": 0.5446630402214148, "brier": 0.32441898222513327, "ece": 0.059890149027536935, "mean_confidence": 0.7769627089135385 }, "schema-firewall/cais/mmlu|not_truncated": { "n": 134, "accuracy": 0.917910447761194, "nll": 0.22831852445149217, "brier": 0.11953794011554492, "ece": 0.049322252620511986, "mean_confidence": 0.909431281596867 }, "schema-firewall/google/boolq-negation|not_truncated": { "n": 60, "accuracy": 0.5166666666666667, "nll": 1.2234002001583577, "brier": 0.7434715281813321, "ece": 0.34772479931513467, "mean_confidence": 0.8540057341257731 }, "schema-firewall/google/boolq|not_truncated": { "n": 36, "accuracy": 0.8888888888888888, "nll": 0.44322766041700057, "brier": 0.22839449674166368, "ece": 0.1561499618821674, "mean_confidence": 0.8758690738015704 }, "schema-firewall/nvidia/HelpSteer2-threshold|not_truncated": { "n": 260, "accuracy": 0.5230769230769231, "nll": 0.8322646450712524, "brier": 0.6063095804116249, "ece": 0.2711401492357254, "mean_confidence": 0.7942170723126485 }, "schema-firewall/nvidia/HelpSteer2|not_truncated": { "n": 268, "accuracy": 0.6343283582089553, "nll": 0.8975105692646397, "brier": 0.49568653238742655, "ece": 0.07441585716694149, "mean_confidence": 0.5939769531919885 }, "schema-firewall/nvidia/HelpSteer2|truncated": { "n": 2, "accuracy": 0.0, "nll": 1.0090330243110657, "brier": 0.6396055817604065, "ece": 0.4641771614551544, "mean_confidence": 0.4641771614551544 }, "schema-firewall/nyu-mll/multi_nli|not_truncated": { "n": 200, "accuracy": 0.865, "nll": 0.41195048167137427, "brier": 0.21928038816042317, "ece": 0.061470215469598743, "mean_confidence": 0.8334396223723889 }, "schema-firewall/product-science/xlam-function-calling-60k-raw|not_truncated": { "n": 81, "accuracy": 1.0, "nll": 0.0050883564066314295, "brier": 0.001193973693776232, "ece": 0.004606815032017043, "mean_confidence": 0.9953931849679829 }, "schema-firewall/product-science/xlam-hard|not_truncated": { "n": 211, "accuracy": 0.6303317535545023, "nll": 0.5777726572084735, "brier": 0.400591680919437, "ece": 0.21581395203468354, "mean_confidence": 0.846145705589186 }, "schema-firewall/stanfordnlp/snli|not_truncated": { "n": 208, "accuracy": 0.8846153846153846, "nll": 0.26919846923891994, "brier": 0.15986069018861843, "ece": 0.04227677503457438, "mean_confidence": 0.8858483952398484 }, "schema-firewall/synthetic/nli-haystack|not_truncated": { "n": 155, "accuracy": 0.896774193548387, "nll": 0.28797892007766474, "brier": 0.15715603314421323, "ece": 0.06688037726186936, "mean_confidence": 0.8773749347656004 } }, "by_kind_truncation": { "choice|not_truncated": { "n": 1163, "accuracy": 0.8718830610490111, "nll": 0.3188156900246091, "brier": 0.1816513806989808, "ece": 0.026625588907830587, "mean_confidence": 0.8696533281128671 }, "noul|not_truncated": { "n": 567, "accuracy": 0.5855379188712522, "nll": 0.754248758684351, "brier": 0.5202748309605, "ece": 0.23951466205469096, "mean_confidence": 0.8250525809259431 }, "score|not_truncated": { "n": 268, "accuracy": 0.6343283582089553, "nll": 0.8975105692646397, "brier": 0.49568653238742655, "ece": 0.07441585716694149, "mean_confidence": 0.5939769531919885 }, "score|truncated": { "n": 2, "accuracy": 0.0, "nll": 1.0090330243110657, "brier": 0.6396055817604065, "ece": 0.4641771614551544, "mean_confidence": 0.4641771614551544 } }, "nli_confusion": { "n": 772, "labels": [ "entailment", "neutral", "contradiction" ], "matrix": { "entailment": { "entailment": 208, "neutral": 26, "contradiction": 3 }, "neutral": { "entailment": 40, "neutral": 259, "contradiction": 19 }, "contradiction": { "entailment": 4, "neutral": 24, "contradiction": 189 } } } }, "firewall": { "n": 2000, "accuracy": 0.8725, "nll": 0.34271046637929975, "brier": 0.2024325927257114, "ece": 0.021594500750303322, "mean_confidence": 0.8701762627661228, "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.097443699836731, "groups": { "choice:2": { "temperature": 0.7726892229265028, "raw_temperature": 0.4564978778362274, "n": 80, "shrinkage_weight": 0.4 }, "choice:3-5": { "temperature": 1.0735645605398847, "raw_temperature": 1.0707285404205322, "n": 998, "shrinkage_weight": 0.8926654740608229 }, "choice:6-10": { "temperature": 1.0213380776161125, "raw_temperature": 0.9631407260894775, "n": 147, "shrinkage_weight": 0.550561797752809 }, "noul:*": { "temperature": 1.18076753369684, "raw_temperature": 1.1976659297943115, "n": 618, "shrinkage_weight": 0.8373983739837398 }, "score:*": { "temperature": 1.1118863993359898, "raw_temperature": 1.1230534315109253, "n": 157, "shrinkage_weight": 0.5667870036101083 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "by_dataset": { "firewall/tommccoy1-hans": { "n": 2000, "accuracy": 0.8725, "nll": 0.34271046637929975, "brier": 0.2024325927257114, "ece": 0.021594500750303322, "mean_confidence": 0.8701762627661228 } }, "by_kind": { "noul": { "n": 2000, "accuracy": 0.8725, "nll": 0.34271046637929975, "brier": 0.2024325927257114, "ece": 0.021594500750303322, "mean_confidence": 0.8701762627661228 } }, "by_transform": { "nli.firewall.hans.v1": { "n": 2000, "accuracy": 0.8725, "nll": 0.34271046637929975, "brier": 0.2024325927257114, "ece": 0.021594500750303322, "mean_confidence": 0.8701762627661228 } }, "by_option_bucket": { "2": { "n": 2000, "accuracy": 0.8725, "nll": 0.34271046637929975, "brier": 0.2024325927257114, "ece": 0.021594500750303322, "mean_confidence": 0.8701762627661228 } }, "by_truncation": { "not_truncated": { "n": 2000, "accuracy": 0.8725, "nll": 0.34271046637929975, "brier": 0.2024325927257114, "ece": 0.021594500750303322, "mean_confidence": 0.8701762627661228 } }, "by_dataset_truncation": { "firewall/tommccoy1-hans|not_truncated": { "n": 2000, "accuracy": 0.8725, "nll": 0.34271046637929975, "brier": 0.2024325927257114, "ece": 0.021594500750303322, "mean_confidence": 0.8701762627661228 } }, "by_kind_truncation": { "noul|not_truncated": { "n": 2000, "accuracy": 0.8725, "nll": 0.34271046637929975, "brier": 0.2024325927257114, "ece": 0.021594500750303322, "mean_confidence": 0.8701762627661228 } } }, "confusability": { "n": 1400, "accuracy": 1.0, "nll": 0.00341335700372578, "brier": 7.767048636123954e-05, "ece": 0.003391873708793125, "mean_confidence": 0.9966081262912069, "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.097443699836731, "groups": { "choice:2": { "temperature": 0.7726892229265028, "raw_temperature": 0.4564978778362274, "n": 80, "shrinkage_weight": 0.4 }, "choice:3-5": { "temperature": 1.0735645605398847, "raw_temperature": 1.0707285404205322, "n": 998, "shrinkage_weight": 0.8926654740608229 }, "choice:6-10": { "temperature": 1.0213380776161125, "raw_temperature": 0.9631407260894775, "n": 147, "shrinkage_weight": 0.550561797752809 }, "noul:*": { "temperature": 1.18076753369684, "raw_temperature": 1.1976659297943115, "n": 618, "shrinkage_weight": 0.8373983739837398 }, "score:*": { "temperature": 1.1118863993359898, "raw_temperature": 1.1230534315109253, "n": 157, "shrinkage_weight": 0.5667870036101083 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "by_dataset": { "v142-confusability": { "n": 1400, "accuracy": 1.0, "nll": 0.00341335700372578, "brier": 7.767048636123954e-05, "ece": 0.003391873708793125, "mean_confidence": 0.9966081262912069 } }, "by_kind": { "choice": { "n": 1000, "accuracy": 1.0, "nll": 0.0030758743519090784, "brier": 8.714231419299279e-05, "ece": 0.0030530961751937813, "mean_confidence": 0.9969469038248062 }, "noul": { "n": 200, "accuracy": 1.0, "nll": 0.0011829274808405899, "brier": 3.475759723983174e-06, "ece": 0.0011820575594901905, "mean_confidence": 0.9988179424405098 }, "score": { "n": 200, "accuracy": 1.0, "nll": 0.00733119978569448, "brier": 0.00010450607383972965, "ece": 0.007295577526092556, "mean_confidence": 0.9927044224739074 } }, "by_transform": { "coreference": { "n": 200, "accuracy": 1.0, "nll": 0.005743390580028063, "brier": 0.00021617284226891798, "ece": 0.0056866601109504256, "mean_confidence": 0.9943133398890496 }, "entity_binding": { "n": 200, "accuracy": 1.0, "nll": 0.00016332407969457563, "brier": 4.9414514606826996e-08, "ece": 0.00016330808401110453, "mean_confidence": 0.9998366919159889 }, "negation": { "n": 200, "accuracy": 1.0, "nll": 0.0011829274808405899, "brier": 3.475759723983174e-06, "ece": 0.0011820575594901905, "mean_confidence": 0.9988179424405098 }, "numeric_boundary": { "n": 200, "accuracy": 1.0, "nll": 0.00014066748185541654, "brier": 6.167312968208272e-08, "ece": 0.00014065206050872803, "mean_confidence": 0.9998593479394913 }, "ordinal_boundary": { "n": 200, "accuracy": 1.0, "nll": 0.00733119978569448, "brier": 0.00010450607383972965, "ece": 0.007295577526092556, "mean_confidence": 0.9927044224739074 }, "temporal_state": { "n": 200, "accuracy": 1.0, "nll": 0.007766418813844211, "brier": 0.00015806605823172503, "ece": 0.007725143730640371, "mean_confidence": 0.9922748562693596 }, "tool_argument_validity": { "n": 200, "accuracy": 1.0, "nll": 0.0015655708041231264, "brier": 6.1361582820032e-05, "ece": 0.001549716889858277, "mean_confidence": 0.9984502831101417 } }, "by_option_bucket": { "2": { "n": 400, "accuracy": 1.0, "nll": 0.0006617974813480032, "brier": 1.7687164268326283e-06, "ece": 0.0006613548099995148, "mean_confidence": 0.9993386451900005 }, "3-5": { "n": 1000, "accuracy": 1.0, "nll": 0.004513980812676891, "brier": 0.0001080311943350023, "ece": 0.004484081268310569, "mean_confidence": 0.9955159187316894 } }, "by_truncation": { "not_truncated": { "n": 1400, "accuracy": 1.0, "nll": 0.00341335700372578, "brier": 7.767048636123954e-05, "ece": 0.003391873708793125, "mean_confidence": 0.9966081262912069 } }, "by_dataset_truncation": { "v142-confusability|not_truncated": { "n": 1400, "accuracy": 1.0, "nll": 0.00341335700372578, "brier": 7.767048636123954e-05, "ece": 0.003391873708793125, "mean_confidence": 0.9966081262912069 } }, "by_kind_truncation": { "choice|not_truncated": { "n": 1000, "accuracy": 1.0, "nll": 0.0030758743519090784, "brier": 8.714231419299279e-05, "ece": 0.0030530961751937813, "mean_confidence": 0.9969469038248062 }, "noul|not_truncated": { "n": 200, "accuracy": 1.0, "nll": 0.0011829274808405899, "brier": 3.475759723983174e-06, "ece": 0.0011820575594901905, "mean_confidence": 0.9988179424405098 }, "score|not_truncated": { "n": 200, "accuracy": 1.0, "nll": 0.00733119978569448, "brier": 0.00010450607383972965, "ece": 0.007295577526092556, "mean_confidence": 0.9927044224739074 } } } }, "latency": { "validation": { "eval_batch_size": 8, "batch_latency_ms_mean": 3647.106461352203, "effective_ms_per_example": 455.8883076690254, "skipped_unusable": 0 }, "schema": { "eval_batch_size": 8, "batch_latency_ms_mean": 4148.647082126699, "effective_ms_per_example": 518.5808852658374, "skipped_unusable": 0 }, "firewall": { "eval_batch_size": 8, "batch_latency_ms_mean": 335.46002532273997, "effective_ms_per_example": 41.932503165342496, "skipped_unusable": 0 }, "confusability": { "eval_batch_size": 8, "batch_latency_ms_mean": 658.9870701618826, "effective_ms_per_example": 82.37338377023532, "skipped_unusable": 0 } } }, "head_rlcd": { "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.0621812343597412, "groups": { "choice:2": { "temperature": 0.7285364649335917, "raw_temperature": 0.3852079510688782, "n": 71, "shrinkage_weight": 0.3717277486910995 }, "choice:3-5": { "temperature": 1.0312828950285642, "raw_temperature": 1.0276613235473633, "n": 1007, "shrinkage_weight": 0.8935226264418811 }, "choice:6-10": { "temperature": 1.018117407640952, "raw_temperature": 0.9839622974395752, "n": 149, "shrinkage_weight": 0.5539033457249071 }, "noul:*": { "temperature": 1.12278263121421, "raw_temperature": 1.1349645853042603, "n": 617, "shrinkage_weight": 0.8371777476255088 }, "score:*": { "temperature": 1.1063119547040072, "raw_temperature": 1.1415023803710938, "n": 156, "shrinkage_weight": 0.5652173913043478 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "sets": { "validation": { "n": 2000, "accuracy": 0.8705, "nll": 0.3296415164664375, "brier": 0.1840096441006748, "ece": 0.015826731696724952, "mean_confidence": 0.8667946958690882, "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.0621812343597412, "groups": { "choice:2": { "temperature": 0.7285364649335917, "raw_temperature": 0.3852079510688782, "n": 71, "shrinkage_weight": 0.3717277486910995 }, "choice:3-5": { "temperature": 1.0312828950285642, "raw_temperature": 1.0276613235473633, "n": 1007, "shrinkage_weight": 0.8935226264418811 }, "choice:6-10": { "temperature": 1.018117407640952, "raw_temperature": 0.9839622974395752, "n": 149, "shrinkage_weight": 0.5539033457249071 }, "noul:*": { "temperature": 1.12278263121421, "raw_temperature": 1.1349645853042603, "n": 617, "shrinkage_weight": 0.8371777476255088 }, "score:*": { "temperature": 1.1063119547040072, "raw_temperature": 1.1415023803710938, "n": 156, "shrinkage_weight": 0.5652173913043478 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "by_dataset": { "PolyAI/banking77": { "n": 136, "accuracy": 0.9338235294117647, "nll": 0.18870303818857792, "brier": 0.10519016720641593, "ece": 0.05590041091336926, "mean_confidence": 0.9336026931071982 }, "Rowan/hellaswag": { "n": 153, "accuracy": 0.9215686274509803, "nll": 0.2399784766869879, "brier": 0.11934398814653138, "ece": 0.040469242272034245, "mean_confidence": 0.9290768474924798 }, "alisawuffles/WANLI": { "n": 160, "accuracy": 0.80625, "nll": 0.5381927386275492, "brier": 0.30977863866601185, "ece": 0.06804261095821856, "mean_confidence": 0.7853333316743374 }, "cais/mmlu": { "n": 154, "accuracy": 0.8831168831168831, "nll": 0.3485408564913748, "brier": 0.17917580716248588, "ece": 0.06689213796869503, "mean_confidence": 0.8730696820980542 }, "google/boolq": { "n": 153, "accuracy": 0.869281045751634, "nll": 0.33914438616278964, "brier": 0.19378073975049515, "ece": 0.04670275738036708, "mean_confidence": 0.8901991384481293 }, "google/boolq-negation": { "n": 153, "accuracy": 0.8627450980392157, "nll": 0.29727962399145064, "brier": 0.17111000223924158, "ece": 0.06959474671120737, "mean_confidence": 0.8627709676237667 }, "nvidia/HelpSteer2": { "n": 156, "accuracy": 0.5448717948717948, "nll": 1.00863150984813, "brier": 0.549988131922407, "ece": 0.07374643553525977, "mean_confidence": 0.5746784233129941 }, "nvidia/HelpSteer2-threshold": { "n": 157, "accuracy": 0.8853503184713376, "nll": 0.31764734924213284, "brier": 0.1801272323022331, "ece": 0.05705484463151096, "mean_confidence": 0.8431880648728389 }, "nyu-mll/multi_nli": { "n": 157, "accuracy": 0.8853503184713376, "nll": 0.2977778198298328, "brier": 0.17204729962497461, "ece": 0.0343798731163049, "mean_confidence": 0.8596658488367773 }, "product-science/xlam-function-calling-60k-raw": { "n": 153, "accuracy": 1.0, "nll": 0.0033377126531411883, "brier": 0.001834733447133328, "ece": 0.002718002188439416, "mean_confidence": 0.9972819978115606 }, "product-science/xlam-hard": { "n": 154, "accuracy": 0.9935064935064936, "nll": 0.0357667315122369, "brier": 0.018634809771057987, "ece": 0.020999968051910404, "mean_confidence": 0.9805593196447794 }, "stanfordnlp/snli": { "n": 159, "accuracy": 0.8805031446540881, "nll": 0.3138472866193749, "brier": 0.1782226289040226, "ece": 0.04512691535289933, "mean_confidence": 0.8852375029767834 }, "synthetic/nli-haystack": { "n": 155, "accuracy": 0.864516129032258, "nll": 0.32164501180992494, "brier": 0.19284149839020814, "ece": 0.02968011844542717, "mean_confidence": 0.8698015168789894 } }, "by_kind": { "choice": { "n": 1227, "accuracy": 0.895680521597392, "nll": 0.28458429776615213, "brier": 0.15912234712374151, "ece": 0.016193954776532293, "mean_confidence": 0.8902595518027852 }, "noul": { "n": 617, "accuracy": 0.9027552674230146, "nll": 0.24757144900729028, "brier": 0.14096923776438128, "ece": 0.031755420241981885, "mean_confidence": 0.8939887968222557 }, "score": { "n": 156, "accuracy": 0.5448717948717948, "nll": 1.00863150984813, "brier": 0.549988131922407, "ece": 0.07374643553525977, "mean_confidence": 0.5746784233129941 } }, "by_transform": { "agent.invocation.correct.v1": { "n": 50, "accuracy": 1.0, "nll": 0.04205997154116631, "brier": 0.01677228283144359, "ece": 0.03617538213729857, "mean_confidence": 0.9638246178627015 }, "agent.invocation.corrupt_argument.v1": { "n": 61, "accuracy": 0.9836065573770492, "nll": 0.041105691575422335, "brier": 0.025920638532433198, "ece": 0.02006097699775066, "mean_confidence": 0.9925722235538921 }, "agent.invocation.wrong_tool.v1": { "n": 43, "accuracy": 1.0, "nll": 0.020875136970358272, "brier": 0.010464828202147104, "ece": 0.017023286154103805, "mean_confidence": 0.9829767138458961 }, "commonsense.choice.v2": { "n": 153, "accuracy": 0.9215686274509803, "nll": 0.2399784766869879, "brier": 0.11934398814653138, "ece": 0.040469242272034245, "mean_confidence": 0.9290768474924798 }, "intent.choice.v3": { "n": 136, "accuracy": 0.9338235294117647, "nll": 0.18870303818857792, "brier": 0.10519016720641593, "ece": 0.05590041091336926, "mean_confidence": 0.9336026931071982 }, "knowledge.choice.v1": { "n": 154, "accuracy": 0.8831168831168831, "nll": 0.3485408564913748, "brier": 0.17917580716248588, "ece": 0.06689213796869503, "mean_confidence": 0.8730696820980542 }, "nli.choice.v4": { "n": 157, "accuracy": 0.8853503184713376, "nll": 0.2977778198298328, "brier": 0.17204729962497461, "ece": 0.0343798731163049, "mean_confidence": 0.8596658488367773 }, "nli.hard.v4": { "n": 160, "accuracy": 0.80625, "nll": 0.5381927386275492, "brier": 0.30977863866601185, "ece": 0.06804261095821856, "mean_confidence": 0.7853333316743374 }, "nli.haystack.absent.v1": { "n": 13, "accuracy": 0.9230769230769231, "nll": 0.293056732795846, "brier": 0.1554580526787434, "ece": 0.09880676177831796, "mean_confidence": 0.9119220834511977 }, "nli.haystack.present.v1": { "n": 142, "accuracy": 0.8591549295774648, "nll": 0.324262248621073, "brier": 0.19626392651872251, "ece": 0.028329969921582165, "mean_confidence": 0.8659454086716746 }, "nli.snli.v1": { "n": 159, "accuracy": 0.8805031446540881, "nll": 0.3138472866193749, "brier": 0.1782226289040226, "ece": 0.04512691535289933, "mean_confidence": 0.8852375029767834 }, "noul.negated_answer.v1": { "n": 153, "accuracy": 0.8627450980392157, "nll": 0.29727962399145064, "brier": 0.17111000223924158, "ece": 0.06959474671120737, "mean_confidence": 0.8627709676237667 }, "noul.v3": { "n": 153, "accuracy": 0.869281045751634, "nll": 0.33914438616278964, "brier": 0.19378073975049515, "ece": 0.04670275738036708, "mean_confidence": 0.8901991384481293 }, "score.coherence.threshold.v1": { "n": 29, "accuracy": 0.9655172413793104, "nll": 0.19833977429742186, "brier": 0.10644655702463783, "ece": 0.1178100828466744, "mean_confidence": 0.8477071585326359 }, "score.coherence.v3": { "n": 27, "accuracy": 0.8148148148148148, "nll": 0.4967265730654752, "brier": 0.2842161387757019, "ece": 0.099791513548957, "mean_confidence": 0.7278836170832316 }, "score.complexity.threshold.v1": { "n": 36, "accuracy": 0.9166666666666666, "nll": 0.32385558721661156, "brier": 0.17315642751181157, "ece": 0.1148082067569097, "mean_confidence": 0.8971044040388532 }, "score.complexity.v3": { "n": 40, "accuracy": 0.6, "nll": 0.9197511970996857, "brier": 0.5391126235947012, "ece": 0.10441315844655036, "mean_confidence": 0.6230462364852428 }, "score.correctness.threshold.v1": { "n": 29, "accuracy": 0.7931034482758621, "nll": 0.342922273381003, "brier": 0.21508417756232465, "ece": 0.14295518809351426, "mean_confidence": 0.7817615517254534 }, "score.correctness.v3": { "n": 37, "accuracy": 0.35135135135135137, "nll": 1.3171428655450408, "brier": 0.7058205044752842, "ece": 0.20661347017094894, "mean_confidence": 0.45046308274204666 }, "score.helpfulness.threshold.v1": { "n": 25, "accuracy": 0.8, "nll": 0.44108421333134173, "brier": 0.28551679542288183, "ece": 0.13267860651016236, "mean_confidence": 0.7273315024375916 }, "score.helpfulness.v3": { "n": 28, "accuracy": 0.42857142857142855, "nll": 1.2584992029837199, "brier": 0.6579760227884565, "ece": 0.2361379388187613, "mean_confidence": 0.43264825482453617 }, "score.verbosity.threshold.v1": { "n": 38, "accuracy": 0.9210526315789473, "nll": 0.30231915724365727, "brier": 0.14694823401135326, "ece": 0.07115527046354195, "mean_confidence": 0.9117601968740162 }, "score.verbosity.v3": { "n": 24, "accuracy": 0.5833333333333334, "nll": 0.9655244362850984, "brier": 0.5008800243958831, "ece": 0.10565355420112608, "mean_confidence": 0.6789100716511408 }, "tool.choice.single.v3": { "n": 153, "accuracy": 1.0, "nll": 0.0033377126531411883, "brier": 0.001834733447133328, "ece": 0.002718002188439416, "mean_confidence": 0.9972819978115606 } }, "by_option_bucket": { "2": { "n": 688, "accuracy": 0.9127906976744186, "nll": 0.22202356312992652, "brier": 0.1264215409403835, "ece": 0.02847742297968196, "mean_confidence": 0.9049279947093752 }, "3-5": { "n": 1163, "accuracy": 0.8366294067067928, "nll": 0.4134522701136551, "brier": 0.22935059510468528, "ece": 0.013319709643350023, "mean_confidence": 0.8349535806058095 }, "6-10": { "n": 149, "accuracy": 0.9395973154362416, "nll": 0.17238812991479716, "brier": 0.09601695253434102, "ece": 0.05117117458541918, "mean_confidence": 0.9392477659971122 } }, "by_truncation": { "not_truncated": { "n": 1999, "accuracy": 0.8704352176088044, "nll": 0.3295081802774739, "brier": 0.18389991385929264, "ece": 0.015609992585103528, "mean_confidence": 0.8669527163351697 }, "truncated": { "n": 1, "accuracy": 1.0, "nll": 0.5961805582046509, "brier": 0.40336039662361145, "ece": 0.4490882158279419, "mean_confidence": 0.5509117841720581 } }, "by_dataset_truncation": { "PolyAI/banking77|not_truncated": { "n": 136, "accuracy": 0.9338235294117647, "nll": 0.18870303818857792, "brier": 0.10519016720641593, "ece": 0.05590041091336926, "mean_confidence": 0.9336026931071982 }, "Rowan/hellaswag|not_truncated": { "n": 153, "accuracy": 0.9215686274509803, "nll": 0.2399784766869879, "brier": 0.11934398814653138, "ece": 0.040469242272034245, "mean_confidence": 0.9290768474924798 }, "alisawuffles/WANLI|not_truncated": { "n": 160, "accuracy": 0.80625, "nll": 0.5381927386275492, "brier": 0.30977863866601185, "ece": 0.06804261095821856, "mean_confidence": 0.7853333316743374 }, "cais/mmlu|not_truncated": { "n": 154, "accuracy": 0.8831168831168831, "nll": 0.3485408564913748, "brier": 0.17917580716248588, "ece": 0.06689213796869503, "mean_confidence": 0.8730696820980542 }, "google/boolq-negation|not_truncated": { "n": 153, "accuracy": 0.8627450980392157, "nll": 0.29727962399145064, "brier": 0.17111000223924158, "ece": 0.06959474671120737, "mean_confidence": 0.8627709676237667 }, "google/boolq|not_truncated": { "n": 153, "accuracy": 0.869281045751634, "nll": 0.33914438616278964, "brier": 0.19378073975049515, "ece": 0.04670275738036708, "mean_confidence": 0.8901991384481293 }, "nvidia/HelpSteer2-threshold|not_truncated": { "n": 156, "accuracy": 0.8846153846153846, "nll": 0.3158618799539116, "brier": 0.17869625047966017, "ece": 0.0545418102007646, "mean_confidence": 0.8450616307747669 }, "nvidia/HelpSteer2-threshold|truncated": { "n": 1, "accuracy": 1.0, "nll": 0.5961805582046509, "brier": 0.40336039662361145, "ece": 0.4490882158279419, "mean_confidence": 0.5509117841720581 }, "nvidia/HelpSteer2|not_truncated": { "n": 156, "accuracy": 0.5448717948717948, "nll": 1.00863150984813, "brier": 0.549988131922407, "ece": 0.07374643553525977, "mean_confidence": 0.5746784233129941 }, "nyu-mll/multi_nli|not_truncated": { "n": 157, "accuracy": 0.8853503184713376, "nll": 0.2977778198298328, "brier": 0.17204729962497461, "ece": 0.0343798731163049, "mean_confidence": 0.8596658488367773 }, "product-science/xlam-function-calling-60k-raw|not_truncated": { "n": 153, "accuracy": 1.0, "nll": 0.0033377126531411883, "brier": 0.001834733447133328, "ece": 0.002718002188439416, "mean_confidence": 0.9972819978115606 }, "product-science/xlam-hard|not_truncated": { "n": 154, "accuracy": 0.9935064935064936, "nll": 0.0357667315122369, "brier": 0.018634809771057987, "ece": 0.020999968051910404, "mean_confidence": 0.9805593196447794 }, "stanfordnlp/snli|not_truncated": { "n": 159, "accuracy": 0.8805031446540881, "nll": 0.3138472866193749, "brier": 0.1782226289040226, "ece": 0.04512691535289933, "mean_confidence": 0.8852375029767834 }, "synthetic/nli-haystack|not_truncated": { "n": 155, "accuracy": 0.864516129032258, "nll": 0.32164501180992494, "brier": 0.19284149839020814, "ece": 0.02968011844542717, "mean_confidence": 0.8698015168789894 } }, "by_kind_truncation": { "choice|not_truncated": { "n": 1227, "accuracy": 0.895680521597392, "nll": 0.28458429776615213, "brier": 0.15912234712374151, "ece": 0.016193954776532293, "mean_confidence": 0.8902595518027852 }, "noul|not_truncated": { "n": 616, "accuracy": 0.9025974025974026, "nll": 0.24700552512872315, "brier": 0.1405432780909085, "ece": 0.031077931937459235, "mean_confidence": 0.89454574002461 }, "noul|truncated": { "n": 1, "accuracy": 1.0, "nll": 0.5961805582046509, "brier": 0.40336039662361145, "ece": 0.4490882158279419, "mean_confidence": 0.5509117841720581 }, "score|not_truncated": { "n": 156, "accuracy": 0.5448717948717948, "nll": 1.00863150984813, "brier": 0.549988131922407, "ece": 0.07374643553525977, "mean_confidence": 0.5746784233129941 } }, "nli_confusion": { "n": 631, "labels": [ "entailment", "neutral", "contradiction" ], "matrix": { "entailment": { "entailment": 187, "neutral": 24, "contradiction": 2 }, "neutral": { "entailment": 21, "neutral": 214, "contradiction": 20 }, "contradiction": { "entailment": 1, "neutral": 21, "contradiction": 141 } } } }, "schema": { "n": 2000, "accuracy": 0.7645, "nll": 0.5092308543568372, "brier": 0.31048622908528783, "ece": 0.06814040819555524, "mean_confidence": 0.8320965484306216, "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.0621812343597412, "groups": { "choice:2": { "temperature": 0.7285364649335917, "raw_temperature": 0.3852079510688782, "n": 71, "shrinkage_weight": 0.3717277486910995 }, "choice:3-5": { "temperature": 1.0312828950285642, "raw_temperature": 1.0276613235473633, "n": 1007, "shrinkage_weight": 0.8935226264418811 }, "choice:6-10": { "temperature": 1.018117407640952, "raw_temperature": 0.9839622974395752, "n": 149, "shrinkage_weight": 0.5539033457249071 }, "noul:*": { "temperature": 1.12278263121421, "raw_temperature": 1.1349645853042603, "n": 617, "shrinkage_weight": 0.8371777476255088 }, "score:*": { "temperature": 1.1063119547040072, "raw_temperature": 1.1415023803710938, "n": 156, "shrinkage_weight": 0.5652173913043478 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "by_dataset": { "schema-firewall/PolyAI/banking77": { "n": 46, "accuracy": 0.8043478260869565, "nll": 0.35900251572955155, "brier": 0.2278477099184452, "ece": 0.11447685244290724, "mean_confidence": 0.8849650835213454 }, "schema-firewall/Rowan/hellaswag": { "n": 130, "accuracy": 0.9076923076923077, "nll": 0.20391861355223787, "brier": 0.12032575284317323, "ece": 0.05504847719119148, "mean_confidence": 0.9189063984614152 }, "schema-firewall/alisawuffles/WANLI": { "n": 214, "accuracy": 0.7570093457943925, "nll": 0.5224599649424656, "brier": 0.3108746913783881, "ece": 0.08684171750166704, "mean_confidence": 0.7917580997275415 }, "schema-firewall/cais/mmlu": { "n": 134, "accuracy": 0.917910447761194, "nll": 0.2270736379013853, "brier": 0.11838710439602669, "ece": 0.05173284882929787, "mean_confidence": 0.9138704791887483 }, "schema-firewall/google/boolq": { "n": 36, "accuracy": 0.8888888888888888, "nll": 0.45656878421868896, "brier": 0.2262111558433004, "ece": 0.13308524257606935, "mean_confidence": 0.8821887258026335 }, "schema-firewall/google/boolq-negation": { "n": 60, "accuracy": 0.5166666666666667, "nll": 1.2670575649982008, "brier": 0.7512862990629401, "ece": 0.35960156718889874, "mean_confidence": 0.860178780555725 }, "schema-firewall/nvidia/HelpSteer2": { "n": 257, "accuracy": 0.6108949416342413, "nll": 0.8855918734569957, "brier": 0.49536697754948056, "ece": 0.09132106381863471, "mean_confidence": 0.5927287479898808 }, "schema-firewall/nvidia/HelpSteer2-threshold": { "n": 254, "accuracy": 0.5433070866141733, "nll": 0.8280847921860258, "brier": 0.595998754226859, "ece": 0.26783039860838037, "mean_confidence": 0.8042075009796563 }, "schema-firewall/nyu-mll/multi_nli": { "n": 216, "accuracy": 0.8703703703703703, "nll": 0.35441261314122974, "brier": 0.1908318395181865, "ece": 0.05234303628956832, "mean_confidence": 0.8628430830107795 }, "schema-firewall/product-science/xlam-function-calling-60k-raw": { "n": 81, "accuracy": 1.0, "nll": 0.004432377281443713, "brier": 0.0009832953881987287, "ece": 0.004043651215824124, "mean_confidence": 0.9959563487841759 }, "schema-firewall/product-science/xlam-hard": { "n": 207, "accuracy": 0.6521739130434783, "nll": 0.5901507024798354, "brier": 0.3917848867662268, "ece": 0.21952186143340696, "mean_confidence": 0.8716957744768852 }, "schema-firewall/stanfordnlp/snli": { "n": 212, "accuracy": 0.8915094339622641, "nll": 0.28470200158541425, "brier": 0.16288797511131478, "ece": 0.05838072018803289, "mean_confidence": 0.8921123490018664 }, "schema-firewall/synthetic/nli-haystack": { "n": 153, "accuracy": 0.9019607843137255, "nll": 0.28354985410883243, "brier": 0.15433488102038004, "ece": 0.06596269872453478, "mean_confidence": 0.8859285342147927 } }, "by_kind": { "choice": { "n": 1186, "accuracy": 0.8735244519392917, "nll": 0.3085243966904598, "brier": 0.17534503393928574, "ece": 0.025424960818499803, "mean_confidence": 0.8800864731029153 }, "noul": { "n": 557, "accuracy": 0.6032315978456014, "nll": 0.762934762585891, "brier": 0.5129334554548768, "ece": 0.23712612482546905, "mean_confidence": 0.8403577226710705 }, "score": { "n": 257, "accuracy": 0.6108949416342413, "nll": 0.8855918734569957, "brier": 0.49536697754948056, "ece": 0.09132106381863471, "mean_confidence": 0.5927287479898808 } }, "by_transform": { "schema.firewall.distractor_drop.v1": { "n": 179, "accuracy": 0.9162011173184358, "nll": 0.18095475863384314, "brier": 0.10300702612023682, "ece": 0.04983094084862223, "mean_confidence": 0.9392977393539258 }, "schema.firewall.instruction.v1": { "n": 1419, "accuracy": 0.8611698379140239, "nll": 0.34634081345778583, "brier": 0.19550318506250666, "ece": 0.02325636315421375, "mean_confidence": 0.853838113273761 }, "schema.firewall.noul_inversion.v1": { "n": 276, "accuracy": 0.2572463768115942, "nll": 1.357916534608365, "brier": 0.9378885064367711, "ece": 0.5137078725341437, "mean_confidence": 0.770954249345738 }, "schema.firewall.score_scale.v1": { "n": 126, "accuracy": 0.5714285714285714, "nll": 0.9510176910294427, "brier": 0.5258567707524413, "ece": 0.10069926556140658, "mean_confidence": 0.568882110809523 } }, "by_option_bucket": { "2": { "n": 584, "accuracy": 0.6215753424657534, "nll": 0.7276621770525656, "brier": 0.48921906625380607, "ece": 0.2261630186479385, "mean_confidence": 0.8477383611136919 }, "3-5": { "n": 1385, "accuracy": 0.8223826714801444, "nll": 0.42260318807209163, "brier": 0.23831324855760574, "ece": 0.029846963028184778, "mean_confidence": 0.8238253159320742 }, "6-10": { "n": 31, "accuracy": 0.8709677419354839, "nll": 0.2645671559719103, "brier": 0.16789271696997024, "ece": 0.09428581979966932, "mean_confidence": 0.9069623033846578 } }, "by_truncation": { "not_truncated": { "n": 2000, "accuracy": 0.7645, "nll": 0.5092308543568372, "brier": 0.31048622908528783, "ece": 0.06814040819555524, "mean_confidence": 0.8320965484306216 } }, "by_dataset_truncation": { "schema-firewall/PolyAI/banking77|not_truncated": { "n": 46, "accuracy": 0.8043478260869565, "nll": 0.35900251572955155, "brier": 0.2278477099184452, "ece": 0.11447685244290724, "mean_confidence": 0.8849650835213454 }, "schema-firewall/Rowan/hellaswag|not_truncated": { "n": 130, "accuracy": 0.9076923076923077, "nll": 0.20391861355223787, "brier": 0.12032575284317323, "ece": 0.05504847719119148, "mean_confidence": 0.9189063984614152 }, "schema-firewall/alisawuffles/WANLI|not_truncated": { "n": 214, "accuracy": 0.7570093457943925, "nll": 0.5224599649424656, "brier": 0.3108746913783881, "ece": 0.08684171750166704, "mean_confidence": 0.7917580997275415 }, "schema-firewall/cais/mmlu|not_truncated": { "n": 134, "accuracy": 0.917910447761194, "nll": 0.2270736379013853, "brier": 0.11838710439602669, "ece": 0.05173284882929787, "mean_confidence": 0.9138704791887483 }, "schema-firewall/google/boolq-negation|not_truncated": { "n": 60, "accuracy": 0.5166666666666667, "nll": 1.2670575649982008, "brier": 0.7512862990629401, "ece": 0.35960156718889874, "mean_confidence": 0.860178780555725 }, "schema-firewall/google/boolq|not_truncated": { "n": 36, "accuracy": 0.8888888888888888, "nll": 0.45656878421868896, "brier": 0.2262111558433004, "ece": 0.13308524257606935, "mean_confidence": 0.8821887258026335 }, "schema-firewall/nvidia/HelpSteer2-threshold|not_truncated": { "n": 254, "accuracy": 0.5433070866141733, "nll": 0.8280847921860258, "brier": 0.595998754226859, "ece": 0.26783039860838037, "mean_confidence": 0.8042075009796563 }, "schema-firewall/nvidia/HelpSteer2|not_truncated": { "n": 257, "accuracy": 0.6108949416342413, "nll": 0.8855918734569957, "brier": 0.49536697754948056, "ece": 0.09132106381863471, "mean_confidence": 0.5927287479898808 }, "schema-firewall/nyu-mll/multi_nli|not_truncated": { "n": 216, "accuracy": 0.8703703703703703, "nll": 0.35441261314122974, "brier": 0.1908318395181865, "ece": 0.05234303628956832, "mean_confidence": 0.8628430830107795 }, "schema-firewall/product-science/xlam-function-calling-60k-raw|not_truncated": { "n": 81, "accuracy": 1.0, "nll": 0.004432377281443713, "brier": 0.0009832953881987287, "ece": 0.004043651215824124, "mean_confidence": 0.9959563487841759 }, "schema-firewall/product-science/xlam-hard|not_truncated": { "n": 207, "accuracy": 0.6521739130434783, "nll": 0.5901507024798354, "brier": 0.3917848867662268, "ece": 0.21952186143340696, "mean_confidence": 0.8716957744768852 }, "schema-firewall/stanfordnlp/snli|not_truncated": { "n": 212, "accuracy": 0.8915094339622641, "nll": 0.28470200158541425, "brier": 0.16288797511131478, "ece": 0.05838072018803289, "mean_confidence": 0.8921123490018664 }, "schema-firewall/synthetic/nli-haystack|not_truncated": { "n": 153, "accuracy": 0.9019607843137255, "nll": 0.28354985410883243, "brier": 0.15433488102038004, "ece": 0.06596269872453478, "mean_confidence": 0.8859285342147927 } }, "by_kind_truncation": { "choice|not_truncated": { "n": 1186, "accuracy": 0.8735244519392917, "nll": 0.3085243966904598, "brier": 0.17534503393928574, "ece": 0.025424960818499803, "mean_confidence": 0.8800864731029153 }, "noul|not_truncated": { "n": 557, "accuracy": 0.6032315978456014, "nll": 0.762934762585891, "brier": 0.5129334554548768, "ece": 0.23712612482546905, "mean_confidence": 0.8403577226710705 }, "score|not_truncated": { "n": 257, "accuracy": 0.6108949416342413, "nll": 0.8855918734569957, "brier": 0.49536697754948056, "ece": 0.09132106381863471, "mean_confidence": 0.5927287479898808 } }, "nli_confusion": { "n": 795, "labels": [ "entailment", "neutral", "contradiction" ], "matrix": { "entailment": { "entailment": 222, "neutral": 30, "contradiction": 4 }, "neutral": { "entailment": 40, "neutral": 255, "contradiction": 18 }, "contradiction": { "entailment": 1, "neutral": 25, "contradiction": 200 } } } }, "firewall": { "n": 2000, "accuracy": 0.8785, "nll": 0.32404716382944027, "brier": 0.19121665542999927, "ece": 0.019892057061195338, "mean_confidence": 0.8801644538640976, "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.0621812343597412, "groups": { "choice:2": { "temperature": 0.7285364649335917, "raw_temperature": 0.3852079510688782, "n": 71, "shrinkage_weight": 0.3717277486910995 }, "choice:3-5": { "temperature": 1.0312828950285642, "raw_temperature": 1.0276613235473633, "n": 1007, "shrinkage_weight": 0.8935226264418811 }, "choice:6-10": { "temperature": 1.018117407640952, "raw_temperature": 0.9839622974395752, "n": 149, "shrinkage_weight": 0.5539033457249071 }, "noul:*": { "temperature": 1.12278263121421, "raw_temperature": 1.1349645853042603, "n": 617, "shrinkage_weight": 0.8371777476255088 }, "score:*": { "temperature": 1.1063119547040072, "raw_temperature": 1.1415023803710938, "n": 156, "shrinkage_weight": 0.5652173913043478 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "by_dataset": { "firewall/tommccoy1-hans": { "n": 2000, "accuracy": 0.8785, "nll": 0.32404716382944027, "brier": 0.19121665542999927, "ece": 0.019892057061195338, "mean_confidence": 0.8801644538640976 } }, "by_kind": { "noul": { "n": 2000, "accuracy": 0.8785, "nll": 0.32404716382944027, "brier": 0.19121665542999927, "ece": 0.019892057061195338, "mean_confidence": 0.8801644538640976 } }, "by_transform": { "nli.firewall.hans.v1": { "n": 2000, "accuracy": 0.8785, "nll": 0.32404716382944027, "brier": 0.19121665542999927, "ece": 0.019892057061195338, "mean_confidence": 0.8801644538640976 } }, "by_option_bucket": { "2": { "n": 2000, "accuracy": 0.8785, "nll": 0.32404716382944027, "brier": 0.19121665542999927, "ece": 0.019892057061195338, "mean_confidence": 0.8801644538640976 } }, "by_truncation": { "not_truncated": { "n": 2000, "accuracy": 0.8785, "nll": 0.32404716382944027, "brier": 0.19121665542999927, "ece": 0.019892057061195338, "mean_confidence": 0.8801644538640976 } }, "by_dataset_truncation": { "firewall/tommccoy1-hans|not_truncated": { "n": 2000, "accuracy": 0.8785, "nll": 0.32404716382944027, "brier": 0.19121665542999927, "ece": 0.019892057061195338, "mean_confidence": 0.8801644538640976 } }, "by_kind_truncation": { "noul|not_truncated": { "n": 2000, "accuracy": 0.8785, "nll": 0.32404716382944027, "brier": 0.19121665542999927, "ece": 0.019892057061195338, "mean_confidence": 0.8801644538640976 } } }, "confusability": { "n": 1400, "accuracy": 1.0, "nll": 0.0022301711396669685, "brier": 3.975738844519126e-05, "ece": 0.002218638658523564, "mean_confidence": 0.9977813613414764, "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.0621812343597412, "groups": { "choice:2": { "temperature": 0.7285364649335917, "raw_temperature": 0.3852079510688782, "n": 71, "shrinkage_weight": 0.3717277486910995 }, "choice:3-5": { "temperature": 1.0312828950285642, "raw_temperature": 1.0276613235473633, "n": 1007, "shrinkage_weight": 0.8935226264418811 }, "choice:6-10": { "temperature": 1.018117407640952, "raw_temperature": 0.9839622974395752, "n": 149, "shrinkage_weight": 0.5539033457249071 }, "noul:*": { "temperature": 1.12278263121421, "raw_temperature": 1.1349645853042603, "n": 617, "shrinkage_weight": 0.8371777476255088 }, "score:*": { "temperature": 1.1063119547040072, "raw_temperature": 1.1415023803710938, "n": 156, "shrinkage_weight": 0.5652173913043478 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "by_dataset": { "v142-confusability": { "n": 1400, "accuracy": 1.0, "nll": 0.0022301711396669685, "brier": 3.975738844519126e-05, "ece": 0.002218638658523564, "mean_confidence": 0.9977813613414764 } }, "by_kind": { "choice": { "n": 1000, "accuracy": 1.0, "nll": 0.001581937825678324, "brier": 3.273811125628399e-05, "ece": 0.0015734966397285, "mean_confidence": 0.9984265033602715 }, "noul": { "n": 200, "accuracy": 1.0, "nll": 0.0006024025060469285, "brier": 8.392574431326238e-07, "ece": 0.000602192580699934, "mean_confidence": 0.9993978074193001 }, "score": { "n": 200, "accuracy": 1.0, "nll": 0.007099106343230232, "brier": 0.00011377190539178627, "ece": 0.007060794830322292, "mean_confidence": 0.9929392051696777 } }, "by_transform": { "coreference": { "n": 200, "accuracy": 1.0, "nll": 0.0031666626091100625, "brier": 8.986054856441328e-05, "ece": 0.00314338117837909, "mean_confidence": 0.9968566188216209 }, "entity_binding": { "n": 200, "accuracy": 1.0, "nll": 5.6076028463394324e-05, "brier": 6.457422713664063e-09, "ece": 5.607396364215589e-05, "mean_confidence": 0.9999439260363578 }, "negation": { "n": 200, "accuracy": 1.0, "nll": 0.0006024025060469285, "brier": 8.392574431326238e-07, "ece": 0.000602192580699934, "mean_confidence": 0.9993978074193001 }, "numeric_boundary": { "n": 200, "accuracy": 1.0, "nll": 8.565005251966796e-05, "brier": 2.4895733549690215e-08, "ece": 8.564382791520497e-05, "mean_confidence": 0.9999143561720848 }, "ordinal_boundary": { "n": 200, "accuracy": 1.0, "nll": 0.007099106343230232, "brier": 0.00011377190539178627, "ece": 0.007060794830322292, "mean_confidence": 0.9929392051696777 }, "temporal_state": { "n": 200, "accuracy": 1.0, "nll": 0.0034405127598438412, "brier": 3.449978205878779e-05, "ece": 0.0034316426515579312, "mean_confidence": 0.9965683573484421 }, "tool_argument_validity": { "n": 200, "accuracy": 1.0, "nll": 0.0011607876784546533, "brier": 3.929887250195552e-05, "ece": 0.0011507415771484508, "mean_confidence": 0.9988492584228515 } }, "by_option_bucket": { "2": { "n": 400, "accuracy": 1.0, "nll": 0.0003440262792832982, "brier": 4.32076588341157e-07, "ece": 0.0003439182043075695, "mean_confidence": 0.9996560817956924 }, "3-5": { "n": 1000, "accuracy": 1.0, "nll": 0.0029846290838204366, "brier": 5.5487513187931304e-05, "ece": 0.002968526840209984, "mean_confidence": 0.99703147315979 } }, "by_truncation": { "not_truncated": { "n": 1400, "accuracy": 1.0, "nll": 0.0022301711396669685, "brier": 3.975738844519126e-05, "ece": 0.002218638658523564, "mean_confidence": 0.9977813613414764 } }, "by_dataset_truncation": { "v142-confusability|not_truncated": { "n": 1400, "accuracy": 1.0, "nll": 0.0022301711396669685, "brier": 3.975738844519126e-05, "ece": 0.002218638658523564, "mean_confidence": 0.9977813613414764 } }, "by_kind_truncation": { "choice|not_truncated": { "n": 1000, "accuracy": 1.0, "nll": 0.001581937825678324, "brier": 3.273811125628399e-05, "ece": 0.0015734966397285, "mean_confidence": 0.9984265033602715 }, "noul|not_truncated": { "n": 200, "accuracy": 1.0, "nll": 0.0006024025060469285, "brier": 8.392574431326238e-07, "ece": 0.000602192580699934, "mean_confidence": 0.9993978074193001 }, "score|not_truncated": { "n": 200, "accuracy": 1.0, "nll": 0.007099106343230232, "brier": 0.00011377190539178627, "ece": 0.007060794830322292, "mean_confidence": 0.9929392051696777 } } } }, "latency": { "validation": { "eval_batch_size": 8, "batch_latency_ms_mean": 3746.8793283732375, "effective_ms_per_example": 468.3599160466547, "skipped_unusable": 0 }, "schema": { "eval_batch_size": 8, "batch_latency_ms_mean": 3996.121177363326, "effective_ms_per_example": 499.51514717041573, "skipped_unusable": 0 }, "firewall": { "eval_batch_size": 8, "batch_latency_ms_mean": 335.970186118735, "effective_ms_per_example": 41.996273264841875, "skipped_unusable": 0 }, "confusability": { "eval_batch_size": 8, "batch_latency_ms_mean": 658.859060741961, "effective_ms_per_example": 82.35738259274513, "skipped_unusable": 0 } } }, "head_lora_rlcd": { "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.2130241394042969, "groups": { "choice:2": { "temperature": 1.131178691354451, "raw_temperature": 0.9999842643737793, "n": 68, "shrinkage_weight": 0.3617021276595745 }, "choice:3-5": { "temperature": 1.229406620643328, "raw_temperature": 1.2313735485076904, "n": 1007, "shrinkage_weight": 0.8935226264418811 }, "choice:6-10": { "temperature": 1.1287873284307708, "raw_temperature": 1.0643781423568726, "n": 147, "shrinkage_weight": 0.550561797752809 }, "noul:*": { "temperature": 1.2522414435081368, "raw_temperature": 1.259989619255066, "n": 619, "shrinkage_weight": 0.8376184032476319 }, "score:*": { "temperature": 1.1708337953943981, "raw_temperature": 1.1399664878845215, "n": 159, "shrinkage_weight": 0.5698924731182796 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "sets": { "validation": { "n": 2000, "accuracy": 0.874, "nll": 0.3413633666240774, "brier": 0.18889581276680206, "ece": 0.022634778514504437, "mean_confidence": 0.8636561000198126, "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.2130241394042969, "groups": { "choice:2": { "temperature": 1.131178691354451, "raw_temperature": 0.9999842643737793, "n": 68, "shrinkage_weight": 0.3617021276595745 }, "choice:3-5": { "temperature": 1.229406620643328, "raw_temperature": 1.2313735485076904, "n": 1007, "shrinkage_weight": 0.8935226264418811 }, "choice:6-10": { "temperature": 1.1287873284307708, "raw_temperature": 1.0643781423568726, "n": 147, "shrinkage_weight": 0.550561797752809 }, "noul:*": { "temperature": 1.2522414435081368, "raw_temperature": 1.259989619255066, "n": 619, "shrinkage_weight": 0.8376184032476319 }, "score:*": { "temperature": 1.1708337953943981, "raw_temperature": 1.1399664878845215, "n": 159, "shrinkage_weight": 0.5698924731182796 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "by_dataset": { "PolyAI/banking77": { "n": 136, "accuracy": 0.9338235294117647, "nll": 0.18907353777312358, "brier": 0.10504818881286704, "ece": 0.06510368646944269, "mean_confidence": 0.9324855239075773 }, "Rowan/hellaswag": { "n": 154, "accuracy": 0.922077922077922, "nll": 0.21983189224844757, "brier": 0.11673031982369635, "ece": 0.040538395186523364, "mean_confidence": 0.9012584014759435 }, "alisawuffles/WANLI": { "n": 156, "accuracy": 0.7564102564102564, "nll": 0.5891167478373226, "brier": 0.35234179292866685, "ece": 0.06832877985942057, "mean_confidence": 0.7601621380219092 }, "cais/mmlu": { "n": 153, "accuracy": 0.869281045751634, "nll": 0.40682872855269997, "brier": 0.20491553394237635, "ece": 0.04900335096845434, "mean_confidence": 0.8429947429233127 }, "google/boolq": { "n": 153, "accuracy": 0.8758169934640523, "nll": 0.3271605430153393, "brier": 0.18819155338973212, "ece": 0.05587063856374201, "mean_confidence": 0.8991910511372136 }, "google/boolq-negation": { "n": 153, "accuracy": 0.869281045751634, "nll": 0.3173225029292547, "brier": 0.18408045429418515, "ece": 0.0648853922981063, "mean_confidence": 0.8740317903312982 }, "nvidia/HelpSteer2": { "n": 159, "accuracy": 0.660377358490566, "nll": 0.8923262642614497, "brier": 0.4767587937402931, "ece": 0.0820022630241682, "mean_confidence": 0.6223008767613825 }, "nvidia/HelpSteer2-threshold": { "n": 158, "accuracy": 0.8860759493670886, "nll": 0.24275886275381134, "brier": 0.14648212109415587, "ece": 0.04942092601257034, "mean_confidence": 0.8698332471938073 }, "nyu-mll/multi_nli": { "n": 157, "accuracy": 0.910828025477707, "nll": 0.3416944928808363, "brier": 0.17827376386410335, "ece": 0.08725254399001975, "mean_confidence": 0.844191101705952 }, "product-science/xlam-function-calling-60k-raw": { "n": 154, "accuracy": 1.0, "nll": 0.0060333324190392805, "brier": 0.002306750856381662, "ece": 0.005282950865757921, "mean_confidence": 0.9947170491342421 }, "product-science/xlam-hard": { "n": 155, "accuracy": 0.967741935483871, "nll": 0.06760209277854301, "brier": 0.040693268881268306, "ece": 0.02164438040025776, "mean_confidence": 0.9642918621340105 }, "stanfordnlp/snli": { "n": 158, "accuracy": 0.8607594936708861, "nll": 0.405171424059103, "brier": 0.22694251017077752, "ece": 0.06340998675249801, "mean_confidence": 0.8710687118995039 }, "synthetic/nli-haystack": { "n": 154, "accuracy": 0.8636363636363636, "nll": 0.39674904061384597, "brier": 0.21296215491428802, "ece": 0.03454946155671952, "mean_confidence": 0.8677647721457791 } }, "by_kind": { "choice": { "n": 1222, "accuracy": 0.8887070376432079, "nll": 0.32193676473427585, "brier": 0.17641385298182077, "ece": 0.02652165657077016, "mean_confidence": 0.8757482603948362 }, "noul": { "n": 619, "accuracy": 0.8998384491114702, "nll": 0.23819084123634773, "brier": 0.13959490950745163, "ece": 0.017604535538853695, "mean_confidence": 0.9017802688724967 }, "score": { "n": 159, "accuracy": 0.660377358490566, "nll": 0.8923262642614497, "brier": 0.4767587937402931, "ece": 0.0820022630241682, "mean_confidence": 0.6223008767613825 } }, "by_transform": { "agent.invocation.correct.v1": { "n": 52, "accuracy": 0.9230769230769231, "nll": 0.1550768981419074, "brier": 0.09798712961044377, "ece": 0.06195802413500273, "mean_confidence": 0.9290768962640029 }, "agent.invocation.corrupt_argument.v1": { "n": 58, "accuracy": 1.0, "nll": 0.015422354992740016, "brier": 0.007456194708894294, "ece": 0.01274871106805474, "mean_confidence": 0.9872512889319452 }, "agent.invocation.wrong_tool.v1": { "n": 45, "accuracy": 0.9777777777777777, "nll": 0.03377397972702359, "brier": 0.017325925416392034, "ece": 0.02705154154035782, "mean_confidence": 0.9753925614886814 }, "commonsense.choice.v2": { "n": 154, "accuracy": 0.922077922077922, "nll": 0.21983189224844757, "brier": 0.11673031982369635, "ece": 0.040538395186523364, "mean_confidence": 0.9012584014759435 }, "intent.choice.v3": { "n": 136, "accuracy": 0.9338235294117647, "nll": 0.18907353777312358, "brier": 0.10504818881286704, "ece": 0.06510368646944269, "mean_confidence": 0.9324855239075773 }, "knowledge.choice.v1": { "n": 153, "accuracy": 0.869281045751634, "nll": 0.40682872855269997, "brier": 0.20491553394237635, "ece": 0.04900335096845434, "mean_confidence": 0.8429947429233127 }, "nli.choice.v4": { "n": 157, "accuracy": 0.910828025477707, "nll": 0.3416944928808363, "brier": 0.17827376386410335, "ece": 0.08725254399001975, "mean_confidence": 0.844191101705952 }, "nli.hard.v4": { "n": 156, "accuracy": 0.7564102564102564, "nll": 0.5891167478373226, "brier": 0.35234179292866685, "ece": 0.06832877985942057, "mean_confidence": 0.7601621380219092 }, "nli.haystack.absent.v1": { "n": 6, "accuracy": 1.0, "nll": 0.12007932023455699, "brier": 0.042778262247641884, "ece": 0.10573691129684447, "mean_confidence": 0.8942630887031555 }, "nli.haystack.present.v1": { "n": 148, "accuracy": 0.8581081081081081, "nll": 0.40796538062922255, "brier": 0.2198615019142872, "ece": 0.038296295178903136, "mean_confidence": 0.8666905160691287 }, "nli.snli.v1": { "n": 158, "accuracy": 0.8607594936708861, "nll": 0.405171424059103, "brier": 0.22694251017077752, "ece": 0.06340998675249801, "mean_confidence": 0.8710687118995039 }, "noul.negated_answer.v1": { "n": 153, "accuracy": 0.869281045751634, "nll": 0.3173225029292547, "brier": 0.18408045429418515, "ece": 0.0648853922981063, "mean_confidence": 0.8740317903312982 }, "noul.v3": { "n": 153, "accuracy": 0.8758169934640523, "nll": 0.3271605430153393, "brier": 0.18819155338973212, "ece": 0.05587063856374201, "mean_confidence": 0.8991910511372136 }, "score.coherence.threshold.v1": { "n": 36, "accuracy": 0.9444444444444444, "nll": 0.15781310862965053, "brier": 0.08520178628396025, "ece": 0.08752633300092487, "mean_confidence": 0.8862901313437356 }, "score.coherence.v3": { "n": 36, "accuracy": 0.8611111111111112, "nll": 0.49905938903490704, "brier": 0.27527502324018216, "ece": 0.14887905451986522, "mean_confidence": 0.7211348878012763 }, "score.complexity.threshold.v1": { "n": 28, "accuracy": 0.9285714285714286, "nll": 0.19121888622508518, "brier": 0.10774182653299899, "ece": 0.1085933659757887, "mean_confidence": 0.9150230714253017 }, "score.complexity.v3": { "n": 30, "accuracy": 0.8, "nll": 0.6180950621763865, "brier": 0.36170755798618, "ece": 0.1601276993751526, "mean_confidence": 0.6420709570248921 }, "score.correctness.threshold.v1": { "n": 30, "accuracy": 0.8, "nll": 0.3484654218579332, "brier": 0.23496840521499204, "ece": 0.13440750638643903, "mean_confidence": 0.8265082140763601 }, "score.correctness.v3": { "n": 22, "accuracy": 0.5, "nll": 1.0662444193254819, "brier": 0.5692707571116361, "ece": 0.18789786439050324, "mean_confidence": 0.49144711751829495 }, "score.helpfulness.threshold.v1": { "n": 30, "accuracy": 0.8, "nll": 0.3833232717588544, "brier": 0.2426174452722383, "ece": 0.1406406263510386, "mean_confidence": 0.7806143343448639 }, "score.helpfulness.v3": { "n": 32, "accuracy": 0.40625, "nll": 1.2641449673101306, "brier": 0.6717381589114666, "ece": 0.10654595633968711, "mean_confidence": 0.4571774401701987 }, "score.verbosity.threshold.v1": { "n": 34, "accuracy": 0.9411764705882353, "nll": 0.15784819945966935, "brier": 0.08036953438568162, "ece": 0.047484531122095476, "mean_confidence": 0.9321437022265266 }, "score.verbosity.v3": { "n": 39, "accuracy": 0.6666666666666666, "nll": 1.0630992299471147, "brier": 0.5390754586372238, "ece": 0.19610086618325653, "mean_confidence": 0.7251625657081604 }, "tool.choice.single.v3": { "n": 154, "accuracy": 1.0, "nll": 0.0060333324190392805, "brier": 0.002306750856381662, "ece": 0.005282950865757921, "mean_confidence": 0.9947170491342421 } }, "by_option_bucket": { "2": { "n": 687, "accuracy": 0.9097525473071325, "nll": 0.21461692523512385, "brier": 0.1257776561957685, "ece": 0.015864489380449164, "mean_confidence": 0.9114996939246831 }, "3-5": { "n": 1166, "accuracy": 0.8447684391080618, "nll": 0.4370019674408599, "brier": 0.2376457747592827, "ece": 0.03868178026447169, "mean_confidence": 0.8261751901888724 }, "6-10": { "n": 147, "accuracy": 0.9387755102040817, "nll": 0.1751062011944359, "brier": 0.09719321331828265, "ece": 0.0604115242049808, "mean_confidence": 0.9373580853955276 } }, "by_truncation": { "not_truncated": { "n": 1999, "accuracy": 0.8744372186093047, "nll": 0.34017474692719035, "brier": 0.18811760026757368, "ece": 0.022178890631758277, "mean_confidence": 0.8636209331583297 }, "truncated": { "n": 1, "accuracy": 0.0, "nll": 2.717414140701294, "brier": 1.7445425987243652, "ece": 0.933954656124115, "mean_confidence": 0.933954656124115 } }, "by_dataset_truncation": { "PolyAI/banking77|not_truncated": { "n": 136, "accuracy": 0.9338235294117647, "nll": 0.18907353777312358, "brier": 0.10504818881286704, "ece": 0.06510368646944269, "mean_confidence": 0.9324855239075773 }, "Rowan/hellaswag|not_truncated": { "n": 154, "accuracy": 0.922077922077922, "nll": 0.21983189224844757, "brier": 0.11673031982369635, "ece": 0.040538395186523364, "mean_confidence": 0.9012584014759435 }, "alisawuffles/WANLI|not_truncated": { "n": 156, "accuracy": 0.7564102564102564, "nll": 0.5891167478373226, "brier": 0.35234179292866685, "ece": 0.06832877985942057, "mean_confidence": 0.7601621380219092 }, "cais/mmlu|not_truncated": { "n": 153, "accuracy": 0.869281045751634, "nll": 0.40682872855269997, "brier": 0.20491553394237635, "ece": 0.04900335096845434, "mean_confidence": 0.8429947429233127 }, "google/boolq-negation|not_truncated": { "n": 153, "accuracy": 0.869281045751634, "nll": 0.3173225029292547, "brier": 0.18408045429418515, "ece": 0.0648853922981063, "mean_confidence": 0.8740317903312982 }, "google/boolq|not_truncated": { "n": 153, "accuracy": 0.8758169934640523, "nll": 0.3271605430153393, "brier": 0.18819155338973212, "ece": 0.05587063856374201, "mean_confidence": 0.8991910511372136 }, "nvidia/HelpSteer2-threshold|not_truncated": { "n": 157, "accuracy": 0.89171974522293, "nll": 0.22699672722548342, "brier": 0.13630339193727556, "ece": 0.055684464752294466, "mean_confidence": 0.8694248305764168 }, "nvidia/HelpSteer2-threshold|truncated": { "n": 1, "accuracy": 0.0, "nll": 2.717414140701294, "brier": 1.7445425987243652, "ece": 0.933954656124115, "mean_confidence": 0.933954656124115 }, "nvidia/HelpSteer2|not_truncated": { "n": 159, "accuracy": 0.660377358490566, "nll": 0.8923262642614497, "brier": 0.4767587937402931, "ece": 0.0820022630241682, "mean_confidence": 0.6223008767613825 }, "nyu-mll/multi_nli|not_truncated": { "n": 157, "accuracy": 0.910828025477707, "nll": 0.3416944928808363, "brier": 0.17827376386410335, "ece": 0.08725254399001975, "mean_confidence": 0.844191101705952 }, "product-science/xlam-function-calling-60k-raw|not_truncated": { "n": 154, "accuracy": 1.0, "nll": 0.0060333324190392805, "brier": 0.002306750856381662, "ece": 0.005282950865757921, "mean_confidence": 0.9947170491342421 }, "product-science/xlam-hard|not_truncated": { "n": 155, "accuracy": 0.967741935483871, "nll": 0.06760209277854301, "brier": 0.040693268881268306, "ece": 0.02164438040025776, "mean_confidence": 0.9642918621340105 }, "stanfordnlp/snli|not_truncated": { "n": 158, "accuracy": 0.8607594936708861, "nll": 0.405171424059103, "brier": 0.22694251017077752, "ece": 0.06340998675249801, "mean_confidence": 0.8710687118995039 }, "synthetic/nli-haystack|not_truncated": { "n": 154, "accuracy": 0.8636363636363636, "nll": 0.39674904061384597, "brier": 0.21296215491428802, "ece": 0.03454946155671952, "mean_confidence": 0.8677647721457791 } }, "by_kind_truncation": { "choice|not_truncated": { "n": 1222, "accuracy": 0.8887070376432079, "nll": 0.32193676473427585, "brier": 0.17641385298182077, "ece": 0.02652165657077016, "mean_confidence": 0.8757482603948362 }, "noul|not_truncated": { "n": 618, "accuracy": 0.901294498381877, "nll": 0.23417915304951123, "brier": 0.13699790677409093, "ece": 0.019144275331188614, "mean_confidence": 0.9017282067572029 }, "noul|truncated": { "n": 1, "accuracy": 0.0, "nll": 2.717414140701294, "brier": 1.7445425987243652, "ece": 0.933954656124115, "mean_confidence": 0.933954656124115 }, "score|not_truncated": { "n": 159, "accuracy": 0.660377358490566, "nll": 0.8923262642614497, "brier": 0.4767587937402931, "ece": 0.0820022630241682, "mean_confidence": 0.6223008767613825 } }, "nli_confusion": { "n": 625, "labels": [ "entailment", "neutral", "contradiction" ], "matrix": { "entailment": { "entailment": 179, "neutral": 28, "contradiction": 6 }, "neutral": { "entailment": 17, "neutral": 178, "contradiction": 17 }, "contradiction": { "entailment": 0, "neutral": 27, "contradiction": 173 } } } }, "schema": { "n": 2000, "accuracy": 0.777, "nll": 0.48716114530798577, "brier": 0.2934196735615337, "ece": 0.046331627137959026, "mean_confidence": 0.8208303308263422, "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.2130241394042969, "groups": { "choice:2": { "temperature": 1.131178691354451, "raw_temperature": 0.9999842643737793, "n": 68, "shrinkage_weight": 0.3617021276595745 }, "choice:3-5": { "temperature": 1.229406620643328, "raw_temperature": 1.2313735485076904, "n": 1007, "shrinkage_weight": 0.8935226264418811 }, "choice:6-10": { "temperature": 1.1287873284307708, "raw_temperature": 1.0643781423568726, "n": 147, "shrinkage_weight": 0.550561797752809 }, "noul:*": { "temperature": 1.2522414435081368, "raw_temperature": 1.259989619255066, "n": 619, "shrinkage_weight": 0.8376184032476319 }, "score:*": { "temperature": 1.1708337953943981, "raw_temperature": 1.1399664878845215, "n": 159, "shrinkage_weight": 0.5698924731182796 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "by_dataset": { "schema-firewall/PolyAI/banking77": { "n": 46, "accuracy": 0.8043478260869565, "nll": 0.35682467644105403, "brier": 0.2239034981356199, "ece": 0.12990026823852374, "mean_confidence": 0.8811038130003473 }, "schema-firewall/Rowan/hellaswag": { "n": 130, "accuracy": 0.9076923076923077, "nll": 0.20287465754243106, "brier": 0.11677193438268157, "ece": 0.047463341171924855, "mean_confidence": 0.9111302653184303 }, "schema-firewall/alisawuffles/WANLI": { "n": 205, "accuracy": 0.7219512195121951, "nll": 0.6024273006109203, "brier": 0.36376239454186315, "ece": 0.06989647847850149, "mean_confidence": 0.7581731124622065 }, "schema-firewall/cais/mmlu": { "n": 134, "accuracy": 0.917910447761194, "nll": 0.2275545184470434, "brier": 0.11969006923772431, "ece": 0.05594150722026825, "mean_confidence": 0.9056024651474027 }, "schema-firewall/google/boolq": { "n": 36, "accuracy": 0.8888888888888888, "nll": 0.42648325838106554, "brier": 0.2071607477258826, "ece": 0.13994142909844717, "mean_confidence": 0.8997285084591972 }, "schema-firewall/google/boolq-negation": { "n": 60, "accuracy": 0.5166666666666667, "nll": 1.2815287677299543, "brier": 0.7706017639355196, "ece": 0.36829513410727177, "mean_confidence": 0.8849618007739385 }, "schema-firewall/nvidia/HelpSteer2": { "n": 273, "accuracy": 0.608058608058608, "nll": 0.9225735204778748, "brier": 0.502879177783053, "ece": 0.05189086431330377, "mean_confidence": 0.6023068452601904 }, "schema-firewall/nvidia/HelpSteer2-threshold": { "n": 263, "accuracy": 0.6121673003802282, "nll": 0.7050396920068494, "brier": 0.5057733078411408, "ece": 0.20569920426539143, "mean_confidence": 0.8013089970037511 }, "schema-firewall/nyu-mll/multi_nli": { "n": 214, "accuracy": 0.8691588785046729, "nll": 0.3600308844127701, "brier": 0.19274686407703126, "ece": 0.06172461933064685, "mean_confidence": 0.8438963853867254 }, "schema-firewall/product-science/xlam-function-calling-60k-raw": { "n": 81, "accuracy": 1.0, "nll": 0.0060919241069499275, "brier": 0.00158877624618902, "ece": 0.005427672539228313, "mean_confidence": 0.9945723274607717 }, "schema-firewall/product-science/xlam-hard": { "n": 202, "accuracy": 0.7673267326732673, "nll": 0.3468692268006905, "brier": 0.24214135508483428, "ece": 0.09207318767462629, "mean_confidence": 0.8529215121623313 }, "schema-firewall/stanfordnlp/snli": { "n": 202, "accuracy": 0.8811881188118812, "nll": 0.2818646928258609, "brier": 0.16245338230385975, "ece": 0.040539493360141704, "mean_confidence": 0.8788761552607659 }, "schema-firewall/synthetic/nli-haystack": { "n": 154, "accuracy": 0.8961038961038961, "nll": 0.28225524406940616, "brier": 0.15355019212951115, "ece": 0.0682181323116476, "mean_confidence": 0.8718665600210042 } }, "by_kind": { "choice": { "n": 1166, "accuracy": 0.8653516295025729, "nll": 0.3213735620484178, "brier": 0.1834721313419164, "ece": 0.029624137982700523, "mean_confidence": 0.8651016697057524 }, "noul": { "n": 561, "accuracy": 0.6755793226381461, "nll": 0.6198540930072308, "brier": 0.4200086032863093, "ece": 0.15957637436674665, "mean_confidence": 0.8351556970048928 }, "score": { "n": 273, "accuracy": 0.608058608058608, "nll": 0.9225735204778748, "brier": 0.502879177783053, "ece": 0.05189086431330377, "mean_confidence": 0.6023068452601904 } }, "by_transform": { "schema.firewall.distractor_drop.v1": { "n": 179, "accuracy": 0.9162011173184358, "nll": 0.18068733821303656, "brier": 0.1011414483847378, "ece": 0.032388108236163664, "mean_confidence": 0.9310932723836526 }, "schema.firewall.instruction.v1": { "n": 1416, "accuracy": 0.8509887005649718, "nll": 0.36921639851671667, "brier": 0.2073376011020019, "ece": 0.02585129172535939, "mean_confidence": 0.8416002465801388 }, "schema.firewall.noul_inversion.v1": { "n": 273, "accuracy": 0.3882783882783883, "nll": 1.0767006237783057, "brier": 0.7550764729788241, "ece": 0.35771637259822187, "mean_confidence": 0.7459947608766102 }, "schema.firewall.score_scale.v1": { "n": 132, "accuracy": 0.5984848484848485, "nll": 0.9487088370052251, "brier": 0.5227962695344379, "ece": 0.10626354102383959, "mean_confidence": 0.6032761137819651 } }, "by_option_bucket": { "2": { "n": 588, "accuracy": 0.6904761904761905, "nll": 0.5913940336594421, "brier": 0.4007224980486256, "ece": 0.15224626021725784, "mean_confidence": 0.8427224506934484 }, "3-5": { "n": 1381, "accuracy": 0.8117306299782766, "nll": 0.44767054287509334, "brier": 0.25053120708128784, "ece": 0.027608664106148723, "mean_confidence": 0.8095618236397246 }, "6-10": { "n": 31, "accuracy": 0.8709677419354839, "nll": 0.26934126173276335, "brier": 0.16873939649087583, "ece": 0.11212440363822443, "mean_confidence": 0.9075800709186062 } }, "by_truncation": { "not_truncated": { "n": 1999, "accuracy": 0.7773886943471736, "nll": 0.4869243776422231, "brier": 0.2932593345994733, "ece": 0.04612398467372813, "mean_confidence": 0.8210101314354921 }, "truncated": { "n": 1, "accuracy": 0.0, "nll": 0.9604597091674805, "brier": 0.613937258720398, "ece": 0.46140891313552856, "mean_confidence": 0.46140891313552856 } }, "by_dataset_truncation": { "schema-firewall/PolyAI/banking77|not_truncated": { "n": 46, "accuracy": 0.8043478260869565, "nll": 0.35682467644105403, "brier": 0.2239034981356199, "ece": 0.12990026823852374, "mean_confidence": 0.8811038130003473 }, "schema-firewall/Rowan/hellaswag|not_truncated": { "n": 130, "accuracy": 0.9076923076923077, "nll": 0.20287465754243106, "brier": 0.11677193438268157, "ece": 0.047463341171924855, "mean_confidence": 0.9111302653184303 }, "schema-firewall/alisawuffles/WANLI|not_truncated": { "n": 205, "accuracy": 0.7219512195121951, "nll": 0.6024273006109203, "brier": 0.36376239454186315, "ece": 0.06989647847850149, "mean_confidence": 0.7581731124622065 }, "schema-firewall/cais/mmlu|not_truncated": { "n": 134, "accuracy": 0.917910447761194, "nll": 0.2275545184470434, "brier": 0.11969006923772431, "ece": 0.05594150722026825, "mean_confidence": 0.9056024651474027 }, "schema-firewall/google/boolq-negation|not_truncated": { "n": 60, "accuracy": 0.5166666666666667, "nll": 1.2815287677299543, "brier": 0.7706017639355196, "ece": 0.36829513410727177, "mean_confidence": 0.8849618007739385 }, "schema-firewall/google/boolq|not_truncated": { "n": 36, "accuracy": 0.8888888888888888, "nll": 0.42648325838106554, "brier": 0.2071607477258826, "ece": 0.13994142909844717, "mean_confidence": 0.8997285084591972 }, "schema-firewall/nvidia/HelpSteer2-threshold|not_truncated": { "n": 263, "accuracy": 0.6121673003802282, "nll": 0.7050396920068494, "brier": 0.5057733078411408, "ece": 0.20569920426539143, "mean_confidence": 0.8013089970037511 }, "schema-firewall/nvidia/HelpSteer2|not_truncated": { "n": 272, "accuracy": 0.6102941176470589, "nll": 0.9224342330194572, "brier": 0.502470876014901, "ece": 0.050385283251457344, "mean_confidence": 0.60282485236359 }, "schema-firewall/nvidia/HelpSteer2|truncated": { "n": 1, "accuracy": 0.0, "nll": 0.9604597091674805, "brier": 0.613937258720398, "ece": 0.46140891313552856, "mean_confidence": 0.46140891313552856 }, "schema-firewall/nyu-mll/multi_nli|not_truncated": { "n": 214, "accuracy": 0.8691588785046729, "nll": 0.3600308844127701, "brier": 0.19274686407703126, "ece": 0.06172461933064685, "mean_confidence": 0.8438963853867254 }, "schema-firewall/product-science/xlam-function-calling-60k-raw|not_truncated": { "n": 81, "accuracy": 1.0, "nll": 0.0060919241069499275, "brier": 0.00158877624618902, "ece": 0.005427672539228313, "mean_confidence": 0.9945723274607717 }, "schema-firewall/product-science/xlam-hard|not_truncated": { "n": 202, "accuracy": 0.7673267326732673, "nll": 0.3468692268006905, "brier": 0.24214135508483428, "ece": 0.09207318767462629, "mean_confidence": 0.8529215121623313 }, "schema-firewall/stanfordnlp/snli|not_truncated": { "n": 202, "accuracy": 0.8811881188118812, "nll": 0.2818646928258609, "brier": 0.16245338230385975, "ece": 0.040539493360141704, "mean_confidence": 0.8788761552607659 }, "schema-firewall/synthetic/nli-haystack|not_truncated": { "n": 154, "accuracy": 0.8961038961038961, "nll": 0.28225524406940616, "brier": 0.15355019212951115, "ece": 0.0682181323116476, "mean_confidence": 0.8718665600210042 } }, "by_kind_truncation": { "choice|not_truncated": { "n": 1166, "accuracy": 0.8653516295025729, "nll": 0.3213735620484178, "brier": 0.1834721313419164, "ece": 0.029624137982700523, "mean_confidence": 0.8651016697057524 }, "noul|not_truncated": { "n": 561, "accuracy": 0.6755793226381461, "nll": 0.6198540930072308, "brier": 0.4200086032863093, "ece": 0.15957637436674665, "mean_confidence": 0.8351556970048928 }, "score|not_truncated": { "n": 272, "accuracy": 0.6102941176470589, "nll": 0.9224342330194572, "brier": 0.502470876014901, "ece": 0.050385283251457344, "mean_confidence": 0.60282485236359 }, "score|truncated": { "n": 1, "accuracy": 0.0, "nll": 0.9604597091674805, "brier": 0.613937258720398, "ece": 0.46140891313552856, "mean_confidence": 0.46140891313552856 } }, "nli_confusion": { "n": 775, "labels": [ "entailment", "neutral", "contradiction" ], "matrix": { "entailment": { "entailment": 201, "neutral": 38, "contradiction": 1 }, "neutral": { "entailment": 31, "neutral": 263, "contradiction": 22 }, "contradiction": { "entailment": 4, "neutral": 29, "contradiction": 186 } } } }, "firewall": { "n": 2000, "accuracy": 0.889, "nll": 0.3097366738639539, "brier": 0.1801932942780604, "ece": 0.02010454627871515, "mean_confidence": 0.8843429988920689, "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.2130241394042969, "groups": { "choice:2": { "temperature": 1.131178691354451, "raw_temperature": 0.9999842643737793, "n": 68, "shrinkage_weight": 0.3617021276595745 }, "choice:3-5": { "temperature": 1.229406620643328, "raw_temperature": 1.2313735485076904, "n": 1007, "shrinkage_weight": 0.8935226264418811 }, "choice:6-10": { "temperature": 1.1287873284307708, "raw_temperature": 1.0643781423568726, "n": 147, "shrinkage_weight": 0.550561797752809 }, "noul:*": { "temperature": 1.2522414435081368, "raw_temperature": 1.259989619255066, "n": 619, "shrinkage_weight": 0.8376184032476319 }, "score:*": { "temperature": 1.1708337953943981, "raw_temperature": 1.1399664878845215, "n": 159, "shrinkage_weight": 0.5698924731182796 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "by_dataset": { "firewall/tommccoy1-hans": { "n": 2000, "accuracy": 0.889, "nll": 0.3097366738639539, "brier": 0.1801932942780604, "ece": 0.02010454627871515, "mean_confidence": 0.8843429988920689 } }, "by_kind": { "noul": { "n": 2000, "accuracy": 0.889, "nll": 0.3097366738639539, "brier": 0.1801932942780604, "ece": 0.02010454627871515, "mean_confidence": 0.8843429988920689 } }, "by_transform": { "nli.firewall.hans.v1": { "n": 2000, "accuracy": 0.889, "nll": 0.3097366738639539, "brier": 0.1801932942780604, "ece": 0.02010454627871515, "mean_confidence": 0.8843429988920689 } }, "by_option_bucket": { "2": { "n": 2000, "accuracy": 0.889, "nll": 0.3097366738639539, "brier": 0.1801932942780604, "ece": 0.02010454627871515, "mean_confidence": 0.8843429988920689 } }, "by_truncation": { "not_truncated": { "n": 2000, "accuracy": 0.889, "nll": 0.3097366738639539, "brier": 0.1801932942780604, "ece": 0.02010454627871515, "mean_confidence": 0.8843429988920689 } }, "by_dataset_truncation": { "firewall/tommccoy1-hans|not_truncated": { "n": 2000, "accuracy": 0.889, "nll": 0.3097366738639539, "brier": 0.1801932942780604, "ece": 0.02010454627871515, "mean_confidence": 0.8843429988920689 } }, "by_kind_truncation": { "noul|not_truncated": { "n": 2000, "accuracy": 0.889, "nll": 0.3097366738639539, "brier": 0.1801932942780604, "ece": 0.02010454627871515, "mean_confidence": 0.8843429988920689 } } }, "confusability": { "n": 1400, "accuracy": 1.0, "nll": 0.0021607741066457364, "brier": 3.6698297335017013e-05, "ece": 0.0021507594840867515, "mean_confidence": 0.9978492405159133, "calibration": { "version": 2, "type": "primitive_option_temperature", "global_temperature": 1.2130241394042969, "groups": { "choice:2": { "temperature": 1.131178691354451, "raw_temperature": 0.9999842643737793, "n": 68, "shrinkage_weight": 0.3617021276595745 }, "choice:3-5": { "temperature": 1.229406620643328, "raw_temperature": 1.2313735485076904, "n": 1007, "shrinkage_weight": 0.8935226264418811 }, "choice:6-10": { "temperature": 1.1287873284307708, "raw_temperature": 1.0643781423568726, "n": 147, "shrinkage_weight": 0.550561797752809 }, "noul:*": { "temperature": 1.2522414435081368, "raw_temperature": 1.259989619255066, "n": 619, "shrinkage_weight": 0.8376184032476319 }, "score:*": { "temperature": 1.1708337953943981, "raw_temperature": 1.1399664878845215, "n": 159, "shrinkage_weight": 0.5698924731182796 } }, "min_group": 50, "prior_strength": 120.0, "n": 2000 }, "by_dataset": { "v142-confusability": { "n": 1400, "accuracy": 1.0, "nll": 0.0021607741066457364, "brier": 3.6698297335017013e-05, "ece": 0.0021507594840867515, "mean_confidence": 0.9978492405159133 } }, "by_kind": { "choice": { "n": 1000, "accuracy": 1.0, "nll": 0.001977618149085174, "brier": 4.1289017551536136e-05, "ece": 0.001966866195201821, "mean_confidence": 0.9980331338047981 }, "noul": { "n": 200, "accuracy": 1.0, "nll": 0.0006506792725122068, "brier": 9.928545260606826e-07, "ece": 0.0006504309177398637, "mean_confidence": 0.9993495690822601 }, "score": { "n": 200, "accuracy": 1.0, "nll": 0.004586648728582077, "brier": 4.945013906137774e-05, "ece": 0.00457055449485777, "mean_confidence": 0.9954294455051422 } }, "by_transform": { "coreference": { "n": 200, "accuracy": 1.0, "nll": 0.00249042555813503, "brier": 4.369134036505073e-05, "ece": 0.002478986680507611, "mean_confidence": 0.9975210133194924 }, "entity_binding": { "n": 200, "accuracy": 1.0, "nll": 0.0001403224484056409, "brier": 3.782463963375449e-08, "ece": 0.0001403102278709767, "mean_confidence": 0.999859689772129 }, "negation": { "n": 200, "accuracy": 1.0, "nll": 0.0006506792725122068, "brier": 9.928545260606826e-07, "ece": 0.0006504309177398637, "mean_confidence": 0.9993495690822601 }, "numeric_boundary": { "n": 200, "accuracy": 1.0, "nll": 0.0012902074388694018, "brier": 4.61289219387595e-06, "ece": 0.0012890523672104415, "mean_confidence": 0.9987109476327896 }, "ordinal_boundary": { "n": 200, "accuracy": 1.0, "nll": 0.004586648728582077, "brier": 4.945013906137774e-05, "ece": 0.00457055449485777, "mean_confidence": 0.9954294455051422 }, "temporal_state": { "n": 200, "accuracy": 1.0, "nll": 0.003792114706011489, "brier": 3.8238258427440995e-05, "ece": 0.0037821048498153376, "mean_confidence": 0.9962178951501847 }, "tool_argument_validity": { "n": 200, "accuracy": 1.0, "nll": 0.002175020594004309, "brier": 0.00011986477213167924, "ece": 0.0021438768506050277, "mean_confidence": 0.997856123149395 } }, "by_option_bucket": { "2": { "n": 400, "accuracy": 1.0, "nll": 0.0009704433556908043, "brier": 2.8028733599683165e-06, "ece": 0.0009697416424750971, "mean_confidence": 0.9990302583575249 }, "3-5": { "n": 1000, "accuracy": 1.0, "nll": 0.002636906407027709, "brier": 5.025646692503649e-05, "ece": 0.0026231666207314017, "mean_confidence": 0.9973768333792686 } }, "by_truncation": { "not_truncated": { "n": 1400, "accuracy": 1.0, "nll": 0.0021607741066457364, "brier": 3.6698297335017013e-05, "ece": 0.0021507594840867515, "mean_confidence": 0.9978492405159133 } }, "by_dataset_truncation": { "v142-confusability|not_truncated": { "n": 1400, "accuracy": 1.0, "nll": 0.0021607741066457364, "brier": 3.6698297335017013e-05, "ece": 0.0021507594840867515, "mean_confidence": 0.9978492405159133 } }, "by_kind_truncation": { "choice|not_truncated": { "n": 1000, "accuracy": 1.0, "nll": 0.001977618149085174, "brier": 4.1289017551536136e-05, "ece": 0.001966866195201821, "mean_confidence": 0.9980331338047981 }, "noul|not_truncated": { "n": 200, "accuracy": 1.0, "nll": 0.0006506792725122068, "brier": 9.928545260606826e-07, "ece": 0.0006504309177398637, "mean_confidence": 0.9993495690822601 }, "score|not_truncated": { "n": 200, "accuracy": 1.0, "nll": 0.004586648728582077, "brier": 4.945013906137774e-05, "ece": 0.00457055449485777, "mean_confidence": 0.9954294455051422 } } } }, "latency": { "validation": { "eval_batch_size": 8, "batch_latency_ms_mean": 3659.3674976518378, "effective_ms_per_example": 457.4209372064797, "skipped_unusable": 0 }, "schema": { "eval_batch_size": 8, "batch_latency_ms_mean": 4174.043139163288, "effective_ms_per_example": 521.755392395411, "skipped_unusable": 0 }, "firewall": { "eval_batch_size": 8, "batch_latency_ms_mean": 336.11437101254705, "effective_ms_per_example": 42.01429637656838, "skipped_unusable": 0 }, "confusability": { "eval_batch_size": 8, "batch_latency_ms_mean": 658.7701784420226, "effective_ms_per_example": 82.34627230525282, "skipped_unusable": 0 } } } }, "lora_mode": "auto" }