2024-03-06 10:10:20 Publish script update

data/time_of_day_comparison.csv

@@ -1,177 +1,177 @@
-model,time_of_day,execution_time,characters_count,words_count
-01-ai Yi Chat (34B),early morning,9.13007491738049,397.0,40.625
-01-ai Yi Chat (34B),morning,9.243995500647504,410.5,40.625
-01-ai Yi Chat (34B),afternoon,12.255323665086614,512.3645833333334,51.59375
-01-ai Yi Chat (34B),late afternoon,9.417656523840767,384.9583333333333,40.125
-01-ai Yi Chat (34B),evening,8.16779062861488,360.8333333333333,39.375
-01-ai Yi Chat (34B),late evening,9.680033167203268,424.5,40.875
-01-ai Yi Chat (34B),midnight,9.311747227774726,397.0625,40.416666666666664
-01-ai Yi Chat (34B),night,9.625520552260966,421.6125,41.45
-Chronos Hermes (13B),early morning,8.063542557544396,507.8181818181818,54.63636363636363
-Chronos Hermes (13B),morning,8.221057146787643,595.375,64.5
-Chronos Hermes (13B),afternoon,6.651120054020601,539.5,76.10416666666667
-Chronos Hermes (13B),late afternoon,7.630303209478205,544.625,62.833333333333336
-Chronos Hermes (13B),evening,6.468730966250102,543.05,72.15
-Chronos Hermes (13B),late evening,7.710561646355523,423.1363636363636,46.31818181818182
-Chronos Hermes (13B),midnight,7.710010987520218,486.5217391304348,54.93478260869565
-Chronos Hermes (13B),night,16.544231017430622,536.2317073170732,62.76829268292683
-Falcon Instruct (7B),early morning,1.2471453740046574,46.291666666666664,8.569444444444445
-Falcon Instruct (7B),morning,1.4012710821060907,47.291666666666664,8.75
-Falcon Instruct (7B),afternoon,1.5774729509611387,36.96875,7.020833333333333
-Falcon Instruct (7B),late afternoon,1.0884015900748116,43.833333333333336,8.166666666666666
-Falcon Instruct (7B),evening,0.7874892950057983,39.2,7.4
-Falcon Instruct (7B),late evening,1.2339241071180864,46.791666666666664,8.708333333333334
-Falcon Instruct (7B),midnight,0.9382020387894068,38.541666666666664,7.270833333333333
-Falcon Instruct (7B),night,1.2997726936862892,45.90243902439025,8.524390243902438
-LLaMA-2 Chat (7B),early morning,2.8209988124230327,444.7361111111111,60.84722222222222
-LLaMA-2 Chat (7B),morning,2.678236266841059,455.0416666666667,60.791666666666664
-LLaMA-2 Chat (7B),afternoon,3.5475496424569024,477.9479166666667,53.135416666666664
-LLaMA-2 Chat (7B),late afternoon,2.7755608558654785,430.45,62.15
-LLaMA-2 Chat (7B),evening,2.9608939344232734,432.2916666666667,62.333333333333336
-LLaMA-2 Chat (7B),late evening,2.869330001913983,452.7083333333333,63.083333333333336
-LLaMA-2 Chat (7B),midnight,2.8870244533457656,460.6666666666667,63.166666666666664
-LLaMA-2 Chat (7B),night,4.6455250724417265,419.5,57.666666666666664
-Mistral (7B) Instruct v0.2 (Together AI),early morning,3.627201789248306,515.1416666666667,64.99166666666666
-Mistral (7B) Instruct v0.2 (Together AI),morning,3.6725051470205816,500.82894736842104,63.75
-Mistral (7B) Instruct v0.2 (Together AI),afternoon,3.255509059895914,481.76237623762376,60.7970297029703
-Mistral (7B) Instruct v0.2 (Together AI),late afternoon,3.542298033617545,543.5138888888889,65.65277777777777
-Mistral (7B) Instruct v0.2 (Together AI),evening,5.496347131400273,507.48387096774195,63.774193548387096
-Mistral (7B) Instruct v0.2 (Together AI),late evening,3.3933188574654713,515.2794117647059,66.45588235294117
-Mistral (7B) Instruct v0.2 (Together AI),midnight,4.510189438914205,517.34375,65.16666666666667
-Mistral (7B) Instruct v0.2 (Together AI),night,3.5451678467578573,500.3,62.823076923076925
-Mistral-7B-Instruct-v0.2,early morning,4.625752210617065,214.0,29.75
-Mistral-7B-Instruct-v0.2,morning,3.734026002883911,362.9,58.35
-Mistral-7B-Instruct-v0.2,afternoon,3.228973722457886,222.0625,32.25
-Mistral-7B-Instruct-v0.2,late afternoon,3.2048643112182615,219.8625,31.8
-Mistral-7B-Instruct-v0.2,evening,3.397640073299408,261.18333333333334,40.1
-Mistral-7B-Instruct-v0.2,late evening,3.389284573495388,175.79375,25.68125
-Mistral-7B-Instruct-v0.2,midnight,2.6972494465964183,88.14285714285714,10.607142857142858
-Mistral-7B-Instruct-v0.2,night,4.609350740909576,214.0,29.75
-Mixtral-8x7B-Instruct-v0.1,early morning,4.526968242530536,285.045,41.86
-Mixtral-8x7B-Instruct-v0.1,morning,3.9661054956285575,304.82,47.28
-Mixtral-8x7B-Instruct-v0.1,afternoon,5.362903979589355,369.3192307692308,54.353846153846156
-Mixtral-8x7B-Instruct-v0.1,late afternoon,5.80184749175942,347.9681818181818,47.27272727272727
-Mixtral-8x7B-Instruct-v0.1,evening,3.6435119574237023,326.69,48.545
-Mixtral-8x7B-Instruct-v0.1,late evening,5.62397656769588,395.15714285714284,49.02857142857143
-Mixtral-8x7B-Instruct-v0.1,midnight,4.639010797279158,323.0394736842105,42.69210526315789
-Mixtral-8x7B-Instruct-v0.1,night,4.009439338194697,301.24545454545455,42.21818181818182
-OpenHermes-2.5-Mistral (7B),early morning,2.8495449788040585,281.0416666666667,32.125
-OpenHermes-2.5-Mistral (7B),morning,2.46380607287089,277.875,32.125
-OpenHermes-2.5-Mistral (7B),afternoon,3.382803752858152,357.8125,32.791666666666664
-OpenHermes-2.5-Mistral (7B),late afternoon,2.52796063820521,281.125,32.125
-OpenHermes-2.5-Mistral (7B),evening,2.7269538966092197,254.08333333333334,26.916666666666668
-OpenHermes-2.5-Mistral (7B),late evening,2.720560759305954,281.125,32.125
-OpenHermes-2.5-Mistral (7B),midnight,2.8086547496470997,273.4583333333333,30.833333333333332
-OpenHermes-2.5-Mistral (7B),night,2.7470715910196306,274.4146341463415,31.317073170731707
-Qwen 1.5 Chat (7B),early morning,2.167507814036475,269.0416666666667,30.125
-Qwen 1.5 Chat (7B),morning,1.8359179911406145,253.66666666666666,27.541666666666668
-Qwen 1.5 Chat (7B),afternoon,2.539412996504042,344.93617021276594,37.37234042553192
-Qwen 1.5 Chat (7B),late afternoon,2.0715979735056558,269.0,30.125
-Qwen 1.5 Chat (7B),evening,2.325377941131592,244.375,27.25
-Qwen 1.5 Chat (7B),late evening,1.9339114997697913,253.66666666666666,27.541666666666668
-Qwen 1.5 Chat (7B),midnight,2.2824907505765872,262.1458333333333,29.020833333333332
-Qwen 1.5 Chat (7B),night,1.98169333812518,262.5375,29.425
-RedPajama-INCITE Chat (7B),early morning,2.149312309920788,55.15714285714286,4.571428571428571
-RedPajama-INCITE Chat (7B),morning,1.7558168431986934,57.375,4.791666666666667
-RedPajama-INCITE Chat (7B),afternoon,1.8110081959854474,57.71875,5.302083333333333
-RedPajama-INCITE Chat (7B),late afternoon,1.7436921298503876,59.625,5.0
-RedPajama-INCITE Chat (7B),evening,1.964497913013805,55.125,4.583333333333333
-RedPajama-INCITE Chat (7B),late evening,2.258105857031686,52.875,4.375
-RedPajama-INCITE Chat (7B),midnight,2.072928147845798,56.25,4.6875
-RedPajama-INCITE Chat (7B),night,2.0220184448437815,57.03658536585366,4.817073170731708
-Snorkel Mistral PairRM DPO (7B),early morning,20.892576156343733,4257.208333333333,100.45833333333333
-Snorkel Mistral PairRM DPO (7B),morning,22.59587260087331,4714.0,104.375
-Snorkel Mistral PairRM DPO (7B),afternoon,34.586263124148054,5710.697916666667,541.0833333333334
-Snorkel Mistral PairRM DPO (7B),late afternoon,22.353231539328892,4714.0,104.375
-Snorkel Mistral PairRM DPO (7B),evening,24.748520737602597,4596.416666666667,92.33333333333333
-Snorkel Mistral PairRM DPO (7B),late evening,24.120955445549704,4623.333333333333,89.625
-Snorkel Mistral PairRM DPO (7B),midnight,22.670283652366475,4691.333333333333,100.6875
-Snorkel Mistral PairRM DPO (7B),night,21.554972467422484,4255.55,95.1125
-TinyLlama/TinyLlama-1.1B-Chat-v1.0,early morning,2.0273348593711855,372.13,62.53
-TinyLlama/TinyLlama-1.1B-Chat-v1.0,morning,1.9041210174560548,372.05,62.6
-TinyLlama/TinyLlama-1.1B-Chat-v1.0,afternoon,1.8381905496120452,308.795,51.08
-TinyLlama/TinyLlama-1.1B-Chat-v1.0,late afternoon,1.7547113946505954,285.17857142857144,46.9
-TinyLlama/TinyLlama-1.1B-Chat-v1.0,evening,1.7984187936782836,313.99,51.96
-TinyLlama/TinyLlama-1.1B-Chat-v1.0,late evening,1.5875422928068372,192.78333333333333,31.261111111111113
-TinyLlama/TinyLlama-1.1B-Chat-v1.0,midnight,1.6341248273849487,210.2,34.2
-TinyLlama/TinyLlama-1.1B-Chat-v1.0,night,2.0128010153770446,372.05,62.6
-Vicuna v1.5 (7B),early morning,3.3891975168596233,274.36764705882354,31.602941176470587
-Vicuna v1.5 (7B),morning,5.204141834507817,308.375,32.833333333333336
-Vicuna v1.5 (7B),afternoon,3.670576659115878,353.1458333333333,43.791666666666664
-Vicuna v1.5 (7B),late afternoon,3.768600355495106,324.27272727272725,35.77272727272727
-Vicuna v1.5 (7B),evening,3.0816855430603027,296.1,38.55
-Vicuna v1.5 (7B),late evening,3.779375910758972,292.5416666666667,30.125
-Vicuna v1.5 (7B),midnight,3.8164960656847273,288.6666666666667,32.5625
-Vicuna v1.5 (7B),night,3.648655335108439,289.6375,31.0875
-WizardLM v1.2 (13B),early morning,5.290586443914884,296.2916666666667,33.708333333333336
-WizardLM v1.2 (13B),morning,4.6954833320949385,250.16666666666666,27.458333333333332
-WizardLM v1.2 (13B),afternoon,7.0199061699127885,320.51063829787233,28.43617021276596
-WizardLM v1.2 (13B),late afternoon,4.935352149217025,254.04166666666666,27.833333333333332
-WizardLM v1.2 (13B),evening,4.939989816058766,240.33333333333334,25.791666666666668
-WizardLM v1.2 (13B),late evening,5.11581133660816,226.25,23.0
-WizardLM v1.2 (13B),midnight,5.255108930847862,259.6041666666667,28.125
-WizardLM v1.2 (13B),night,5.411002513689873,303.9268292682927,35.048780487804876
-chat-bison (PaLM 2),early morning,2.4665334616388592,381.9214285714286,60.892857142857146
-chat-bison (PaLM 2),morning,2.488477897644043,381.73,60.88
-chat-bison (PaLM 2),afternoon,2.6460144804074215,376.28076923076924,55.965384615384615
-chat-bison (PaLM 2),late afternoon,3.0387172081253744,381.1363636363636,53.35454545454545
-chat-bison (PaLM 2),evening,2.688272579908371,367.07,55.89
-chat-bison (PaLM 2),late evening,2.7250528037548065,382.725,52.95
-chat-bison (PaLM 2),midnight,2.468383938074112,381.95,60.9
-chat-bison (PaLM 2),night,2.460119960308075,381.92,60.94
-chat-bison-32k (PaLM 2 32K),early morning,9.141417106560299,335.75,53.85
-chat-bison-32k (PaLM 2 32K),morning,7.7035503840446475,335.75,53.85
-chat-bison-32k (PaLM 2 32K),afternoon,5.014458654477046,339.3692307692308,49.323076923076925
-chat-bison-32k (PaLM 2 32K),late afternoon,4.5047362284226855,348.65909090909093,48.35454545454545
-chat-bison-32k (PaLM 2 32K),evening,7.7332194912433625,329.46,49.5
-chat-bison-32k (PaLM 2 32K),late evening,7.796841062307358,349.95,47.805
-chat-bison-32k (PaLM 2 32K),midnight,7.7498266498247785,335.75,53.85
-chat-bison-32k (PaLM 2 32K),night,6.491292915344238,335.75,53.85
-gemini-pro,early morning,2.7453590130460435,381.2214285714286,60.964285714285715
-gemini-pro,morning,2.497767536007628,371.93,59.39
-gemini-pro,afternoon,2.816922114008949,360.62307692307695,53.25769230769231
-gemini-pro,late afternoon,2.9268629640903114,364.57272727272726,50.1
-gemini-pro,evening,2.86901999375759,366.4,55.045
-gemini-pro,late evening,3.7189874940246117,390.9142857142857,51.35
-gemini-pro,midnight,3.338477972348531,369.825,55.9125
-gemini-pro,night,2.8375814715210272,374.8,60.04
-gpt-3.5-turbo,early morning,3.787998208734724,403.74444444444447,47.34444444444444
-gpt-3.5-turbo,morning,3.126271222697364,389.9888888888889,50.93888888888889
-gpt-3.5-turbo,afternoon,3.9458200880459375,381.4428571428571,43.76190476190476
-gpt-3.5-turbo,late afternoon,4.384064777692159,396.4357142857143,43.28333333333333
-gpt-3.5-turbo,evening,3.5153889304115657,385.18095238095236,46.01428571428571
-gpt-3.5-turbo,late evening,5.110168156187617,422.73510971786834,45.358934169278996
-gpt-3.5-turbo,midnight,3.822115447632102,426.71923076923076,49.05
-gpt-3.5-turbo,night,6.062970260473398,418.6192307692308,44.926923076923075
-gpt-4,early morning,14.348626694414351,323.5388888888889,40.544444444444444
-gpt-4,morning,12.759107512468733,338.18333333333334,46.85
-gpt-4,afternoon,16.002364798386893,318.3095238095238,38.77142857142857
-gpt-4,late afternoon,16.80607506932254,313.3595238095238,37.49285714285714
-gpt-4,evening,13.841120740345547,318.0809523809524,40.19285714285714
-gpt-4,late evening,14.298642643005493,314.336,37.012
-gpt-4,midnight,12.3578163115329,334.075,41.35
-gpt-4,night,12.813134506115546,316.93461538461537,37.93076923076923
-gpt-4-turbo,early morning,11.555620827939775,357.65555555555557,47.21666666666667
-gpt-4-turbo,morning,13.686854598257277,381.8888888888889,55.02777777777778
-gpt-4-turbo,afternoon,13.997754749229976,351.01190476190476,46.03333333333333
-gpt-4-turbo,late afternoon,22.320911452883767,381.65714285714284,47.35476190476191
-gpt-4-turbo,evening,16.550320884159632,382.31666666666666,48.45
-gpt-4-turbo,late evening,14.592236209392548,413.61,44.8
-gpt-4-turbo,midnight,13.770663784850727,382.7613636363636,47.61818181818182
-gpt-4-turbo,night,14.254795966698573,352.54615384615386,46.37692307692308
-llama-2-70b-chat,early morning,2.8660141522424265,289.6642857142857,44.614285714285714
-llama-2-70b-chat,morning,2.872361832027194,283.35,43.45
-llama-2-70b-chat,afternoon,4.234376892130426,375.44615384615383,55.238461538461536
-llama-2-70b-chat,late afternoon,3.6833307104881365,433.6636363636364,59.445454545454545
-llama-2-70b-chat,evening,2.9706250462084185,317.245,47.4
-llama-2-70b-chat,late evening,4.719581684340602,572.1689655172414,79.83103448275862
-llama-2-70b-chat,midnight,3.249819871626402,346.8875,52.06875
-llama-2-70b-chat,night,2.8264514451677147,313.71,48.42
-zephyr-7b-beta,early morning,3.937663261095683,273.03333333333336,43.45
-zephyr-7b-beta,morning,4.056525647640228,386.3,63.35
-zephyr-7b-beta,afternoon,3.5789777278900146,277.6,44.016666666666666
-zephyr-7b-beta,late afternoon,3.4592524923459447,248.59,38.9
-zephyr-7b-beta,evening,3.5673056403795878,273.3666666666667,43.5
-zephyr-7b-beta,late evening,3.45343524068594,176.23125,27.21875
-zephyr-7b-beta,midnight,3.7253047794103624,217.82083333333333,33.67916666666667
-zephyr-7b-beta,night,3.6954557319482166,216.55833333333334,33.46666666666667
+model,time_of_day,execution_time,execution_time_min,execution_time_max,characters_count,characters_count_min,characters_count_max,words_count,words_count_min,words_count_max
+01-ai Yi Chat (34B),early morning,9.13007491738049,2.8026041984558105,21.131053924560547,426.6268656716418,294.0,584.0,43.656716417910445,6.0,89.0
+01-ai Yi Chat (34B),morning,9.243995500647504,3.012052059173584,20.919682502746582,428.3478260869565,294.0,584.0,42.391304347826086,6.0,82.0
+01-ai Yi Chat (34B),afternoon,12.255323665086614,2.0155816078186035,54.017855167388916,571.9418604651163,124.0,1627.0,57.593023255813954,4.0,272.0
+01-ai Yi Chat (34B),late afternoon,9.417656523840767,3.0440945625305176,20.902742624282837,439.95238095238096,297.0,584.0,45.857142857142854,15.0,82.0
+01-ai Yi Chat (34B),evening,8.16779062861488,2.884411573410034,19.63676166534424,412.3809523809524,294.0,584.0,45.0,6.0,82.0
+01-ai Yi Chat (34B),late evening,9.680033167203268,2.7506120204925537,20.657597303390503,424.5,294.0,584.0,40.875,6.0,82.0
+01-ai Yi Chat (34B),midnight,9.311747227774726,2.620687961578369,21.379272937774658,423.53333333333336,294.0,584.0,43.111111111111114,6.0,89.0
+01-ai Yi Chat (34B),night,9.625520552260966,2.8000857830047607,23.83119535446167,426.9493670886076,294.0,584.0,41.9746835443038,6.0,89.0
+Chronos Hermes (13B),early morning,8.063542557544396,1.3158385753631592,31.60169529914856,549.4426229508197,155.0,1442.0,59.114754098360656,5.0,227.0
+Chronos Hermes (13B),morning,8.221057146787643,1.262566089630127,31.414015531539917,595.375,155.0,1442.0,64.5,5.0,227.0
+Chronos Hermes (13B),afternoon,6.651120054020601,1.5965147018432617,28.716880559921265,609.3176470588236,64.0,3031.0,85.95294117647059,2.0,539.0
+Chronos Hermes (13B),late afternoon,7.630303209478205,1.3507373332977295,30.631603956222534,594.1363636363636,155.0,1442.0,68.54545454545455,5.0,227.0
+Chronos Hermes (13B),evening,6.468730966250102,2.2841973304748535,28.56206250190735,603.3888888888889,322.0,1442.0,80.16666666666667,5.0,227.0
+Chronos Hermes (13B),late evening,7.710561646355523,2.432551383972168,30.597865104675293,517.1666666666666,155.0,1442.0,56.611111111111114,5.0,227.0
+Chronos Hermes (13B),midnight,7.710010987520218,1.3517813682556152,31.449679374694824,559.5,155.0,1442.0,63.175,5.0,227.0
+Chronos Hermes (13B),night,7.907361829784555,1.278723955154419,33.31306838989258,586.4929577464789,153.0,1442.0,67.12676056338029,5.0,227.0
+Falcon Instruct (7B),early morning,1.2471453740046574,0.418881893157959,4.323866844177246,51.276923076923076,3.0,101.0,9.492307692307692,1.0,17.0
+Falcon Instruct (7B),morning,1.4012710821060907,0.42804694175720215,3.9746317863464355,54.04761904761905,3.0,101.0,10.0,1.0,17.0
+Falcon Instruct (7B),afternoon,1.5774729509611387,0.38842058181762695,4.269047021865845,47.95945945945946,3.0,101.0,9.108108108108109,1.0,19.0
+Falcon Instruct (7B),late afternoon,1.0884015900748116,0.4407048225402832,3.778937816619873,50.095238095238095,3.0,101.0,9.333333333333334,1.0,17.0
+Falcon Instruct (7B),evening,0.7874892950057983,0.410703182220459,2.426168918609619,43.55555555555556,3.0,101.0,8.222222222222221,1.0,17.0
+Falcon Instruct (7B),late evening,1.2339241071180864,0.42011547088623047,3.741771697998047,51.04545454545455,3.0,101.0,9.5,1.0,17.0
+Falcon Instruct (7B),midnight,0.9382020387894068,0.417522668838501,3.5052387714385986,47.43589743589744,3.0,101.0,8.948717948717949,1.0,17.0
+Falcon Instruct (7B),night,1.2997726936862892,0.40680432319641113,4.247479677200317,51.56164383561644,3.0,101.0,9.575342465753424,1.0,17.0
+LLaMA-2 Chat (7B),early morning,2.8209988124230327,1.035428762435913,6.023603439331055,470.8970588235294,325.0,985.0,64.42647058823529,4.0,149.0
+LLaMA-2 Chat (7B),morning,2.678236266841059,1.0565335750579834,4.452203035354614,474.82608695652175,325.0,985.0,63.43478260869565,4.0,149.0
+LLaMA-2 Chat (7B),afternoon,3.5475496424569024,1.0705671310424805,14.094016075134277,509.81111111111113,152.0,1510.0,56.67777777777778,4.0,213.0
+LLaMA-2 Chat (7B),late afternoon,2.7755608558654785,1.3516404628753662,4.1744301319122314,506.4117647058824,325.0,985.0,73.11764705882354,4.0,149.0
+LLaMA-2 Chat (7B),evening,2.9608939344232734,2.064330577850342,4.3006792068481445,471.59090909090907,325.0,985.0,68.0,4.0,149.0
+LLaMA-2 Chat (7B),late evening,2.869330001913983,1.160935640335083,5.1047961711883545,472.39130434782606,325.0,985.0,65.82608695652173,4.0,149.0
+LLaMA-2 Chat (7B),midnight,2.8870244533457656,1.0700302124023438,5.987953424453735,470.468085106383,325.0,985.0,64.51063829787235,4.0,149.0
+LLaMA-2 Chat (7B),night,4.6455250724417265,1.1601483821868896,22.316121816635132,453.88524590163934,325.0,985.0,62.39344262295082,4.0,149.0
+Mistral (7B) Instruct v0.2 (Together AI),early morning,3.627201789248306,0.9703679084777832,11.814799070358276,547.0530973451328,171.0,1784.0,69.01769911504425,4.0,234.0
+Mistral (7B) Instruct v0.2 (Together AI),morning,3.6725051470205816,0.8738787174224854,11.855232238769531,536.0985915492957,181.0,1784.0,68.2394366197183,4.0,234.0
+Mistral (7B) Instruct v0.2 (Together AI),afternoon,3.255509059895914,0.9155564308166504,18.82912302017212,514.8994708994709,181.0,3014.0,64.97883597883597,2.0,285.0
+Mistral (7B) Instruct v0.2 (Together AI),late afternoon,3.542298033617545,0.9134914875030518,8.720620393753052,567.1449275362319,181.0,1784.0,68.5072463768116,2.0,234.0
+Mistral (7B) Instruct v0.2 (Together AI),evening,5.496347131400273,0.8985519409179688,54.58949375152588,542.4827586206897,181.0,1784.0,68.17241379310344,2.0,234.0
+Mistral (7B) Instruct v0.2 (Together AI),late evening,3.3933188574654713,0.8962244987487793,9.048192739486694,556.1746031746031,181.0,1784.0,71.73015873015873,4.0,234.0
+Mistral (7B) Instruct v0.2 (Together AI),midnight,4.510189438914205,0.8941171169281006,25.338401079177856,545.7692307692307,181.0,1784.0,68.74725274725274,4.0,234.0
+Mistral (7B) Instruct v0.2 (Together AI),night,3.5451678467578573,0.8892025947570801,12.310133695602417,533.1065573770492,171.0,1784.0,66.94262295081967,4.0,234.0
+Mistral-7B-Instruct-v0.2,early morning,4.625752210617065,3.949249505996704,5.039102077484131,214.0,95.0,377.0,29.75,3.0,61.0
+Mistral-7B-Instruct-v0.2,morning,3.734026002883911,2.8412067890167236,4.190511226654053,362.9,252.0,482.0,58.35,39.0,77.0
+Mistral-7B-Instruct-v0.2,afternoon,3.228973722457886,0.7488498687744141,4.515047550201416,222.0625,0.0,503.0,32.25,0.0,77.0
+Mistral-7B-Instruct-v0.2,late afternoon,3.2048643112182615,0.7534739971160889,4.48244833946228,219.8625,0.0,503.0,31.8,0.0,77.0
+Mistral-7B-Instruct-v0.2,evening,3.397640073299408,0.7322485446929932,7.530469655990601,261.18333333333334,0.0,482.0,40.1,0.0,77.0
+Mistral-7B-Instruct-v0.2,late evening,2.796920734381526,0.7476711273193359,5.995672702789307,174.32704402515722,0.0,503.0,25.440251572327043,0.0,77.0
+Mistral-7B-Instruct-v0.2,midnight,2.6972494465964183,0.7803714275360107,5.025636196136475,88.14285714285714,0.0,377.0,10.607142857142858,0.0,61.0
+Mistral-7B-Instruct-v0.2,night,4.609350740909576,3.91265869140625,4.975342035293579,214.0,95.0,377.0,29.75,3.0,61.0
+Mixtral-8x7B-Instruct-v0.1,early morning,3.791912331725612,1.3495268821716309,21.078988790512085,428.6666666666667,190.0,2033.0,62.93181818181818,2.0,310.0
+Mixtral-8x7B-Instruct-v0.1,morning,3.9661054956285575,1.3714869022369385,18.827024698257446,401.07894736842104,208.0,646.0,62.21052631578947,32.0,104.0
+Mixtral-8x7B-Instruct-v0.1,afternoon,5.362903979589355,1.2343645095825195,21.06575345993042,452.938679245283,190.0,2033.0,66.66037735849056,2.0,310.0
+Mixtral-8x7B-Instruct-v0.1,late afternoon,5.272096274973272,1.49995756149292,27.44243097305298,418.75824175824175,173.0,1271.0,56.82967032967033,1.0,216.0
+Mixtral-8x7B-Instruct-v0.1,evening,3.6435119574237023,1.3169512748718262,18.395429134368896,416.1656050955414,190.0,1269.0,61.84076433121019,2.0,214.0
+Mixtral-8x7B-Instruct-v0.1,late evening,4.792964628468389,1.255105972290039,23.370755672454834,478.80434782608694,173.0,2618.0,59.6695652173913,1.0,413.0
+Mixtral-8x7B-Instruct-v0.1,midnight,4.274431848884525,1.3257741928100586,36.458353757858276,459.8721804511278,190.0,1290.0,60.744360902255636,1.0,192.0
+Mixtral-8x7B-Instruct-v0.1,night,4.009439338194697,1.3533086776733398,13.848145723342896,436.0131578947368,190.0,1231.0,61.10526315789474,2.0,188.0
+OpenHermes-2.5-Mistral (7B),early morning,2.8495449788040585,1.1569948196411133,5.551050662994385,281.0416666666667,141.0,374.0,32.125,1.0,64.0
+OpenHermes-2.5-Mistral (7B),morning,2.46380607287089,1.0310070514678955,4.353039741516113,277.875,141.0,374.0,32.125,1.0,64.0
+OpenHermes-2.5-Mistral (7B),afternoon,3.382803752858152,0.9008805751800537,18.957056999206543,369.35483870967744,82.0,2991.0,33.8494623655914,1.0,99.0
+OpenHermes-2.5-Mistral (7B),late afternoon,2.52796063820521,0.9257056713104248,4.2641379833221436,281.125,141.0,374.0,32.125,1.0,64.0
+OpenHermes-2.5-Mistral (7B),evening,2.7269538966092197,1.0407700538635254,4.872406721115112,277.1818181818182,141.0,374.0,29.363636363636363,1.0,64.0
+OpenHermes-2.5-Mistral (7B),late evening,2.720560759305954,1.1110494136810303,4.519970893859863,281.125,141.0,374.0,32.125,1.0,64.0
+OpenHermes-2.5-Mistral (7B),midnight,2.8086547496470997,1.2991385459899902,4.7419211864471436,279.27659574468083,141.0,374.0,31.48936170212766,1.0,64.0
+OpenHermes-2.5-Mistral (7B),night,2.7470715910196306,1.1750683784484863,5.954946756362915,281.275,141.0,374.0,32.1,1.0,64.0
+Qwen 1.5 Chat (7B),early morning,2.167507814036475,0.9619903564453125,4.920398473739624,269.0416666666667,177.0,400.0,30.125,1.0,67.0
+Qwen 1.5 Chat (7B),morning,1.8359179911406145,0.8541216850280762,2.7175967693328857,264.69565217391306,177.0,400.0,28.73913043478261,1.0,67.0
+Qwen 1.5 Chat (7B),afternoon,2.539412996504042,0.8583283424377441,9.480445384979248,360.26666666666665,123.0,1083.0,39.03333333333333,1.0,182.0
+Qwen 1.5 Chat (7B),late afternoon,2.0715979735056558,1.233666181564331,4.4187798500061035,269.0,177.0,400.0,30.125,1.0,67.0
+Qwen 1.5 Chat (7B),evening,2.325377941131592,1.2237350940704346,4.710259199142456,266.59090909090907,177.0,400.0,29.727272727272727,1.0,67.0
+Qwen 1.5 Chat (7B),late evening,1.9339114997697913,1.0104713439941406,4.516375303268433,264.69565217391306,177.0,400.0,28.73913043478261,1.0,67.0
+Qwen 1.5 Chat (7B),midnight,2.2824907505765872,1.069124698638916,4.83498215675354,267.72340425531917,171.0,400.0,29.638297872340427,1.0,67.0
+Qwen 1.5 Chat (7B),night,1.98169333812518,0.8611364364624023,4.549734115600586,269.2692307692308,171.0,400.0,30.17948717948718,1.0,67.0
+RedPajama-INCITE Chat (7B),early morning,2.149312309920788,0.9338541030883789,4.298100709915161,60.328125,51.0,78.0,5.0,4.0,7.0
+RedPajama-INCITE Chat (7B),morning,1.7558168431986934,0.92230224609375,2.5155436992645264,59.869565217391305,51.0,78.0,5.0,4.0,7.0
+RedPajama-INCITE Chat (7B),afternoon,1.8110081959854474,0.9110820293426514,4.171107292175293,62.96590909090909,51.0,83.0,5.784090909090909,2.0,14.0
+RedPajama-INCITE Chat (7B),late afternoon,1.7436921298503876,0.9146637916564941,2.4647669792175293,59.625,51.0,78.0,5.0,4.0,7.0
+RedPajama-INCITE Chat (7B),evening,1.964497913013805,0.9337396621704102,4.010848760604858,60.13636363636363,51.0,78.0,5.0,4.0,7.0
+RedPajama-INCITE Chat (7B),late evening,2.258105857031686,0.9757111072540283,4.155341148376465,60.42857142857143,51.0,78.0,5.0,4.0,7.0
+RedPajama-INCITE Chat (7B),midnight,2.072928147845798,0.9375200271606445,4.163625478744507,60.0,51.0,78.0,5.0,4.0,7.0
+RedPajama-INCITE Chat (7B),night,2.0220184448437815,0.9105582237243652,4.773882150650024,59.96153846153846,51.0,78.0,5.064102564102564,4.0,10.0
+Snorkel Mistral PairRM DPO (7B),early morning,4.5189391105405745,1.3401362895965576,9.995759010314941,840.5,443.0,1251.0,103.11290322580645,70.0,177.0
+Snorkel Mistral PairRM DPO (7B),morning,4.7358971663883755,1.996814489364624,10.112697124481201,844.4285714285714,443.0,1251.0,104.28571428571429,70.0,177.0
+Snorkel Mistral PairRM DPO (7B),afternoon,5.831258978599157,1.2519428730010986,25.606034755706787,1203.423076923077,402.0,4847.0,147.0897435897436,15.0,750.0
+Snorkel Mistral PairRM DPO (7B),late afternoon,4.481673751558576,2.059854507446289,9.038954973220825,844.4285714285714,443.0,1251.0,104.28571428571429,70.0,177.0
+Snorkel Mistral PairRM DPO (7B),evening,4.241402758492364,1.5848162174224854,9.025120496749878,828.3888888888889,443.0,1251.0,105.61111111111111,70.0,177.0
+Snorkel Mistral PairRM DPO (7B),late evening,4.532450111288774,1.3126277923583984,10.296067953109741,818.7894736842105,443.0,1251.0,96.63157894736842,70.0,177.0
+Snorkel Mistral PairRM DPO (7B),midnight,4.327584865616589,1.3062994480133057,9.743696689605713,838.4878048780488,443.0,1251.0,102.51219512195122,70.0,177.0
+Snorkel Mistral PairRM DPO (7B),night,4.384899883559256,1.2272593975067139,10.19804573059082,821.7424242424242,391.0,1251.0,100.96969696969697,58.0,177.0
+TinyLlama/TinyLlama-1.1B-Chat-v1.0,early morning,2.0273348593711855,1.6510882377624512,2.396561861038208,372.13,280.0,448.0,62.53,38.0,74.0
+TinyLlama/TinyLlama-1.1B-Chat-v1.0,morning,1.9041210174560548,1.6700537204742432,2.252462148666382,372.05,280.0,448.0,62.6,38.0,74.0
+TinyLlama/TinyLlama-1.1B-Chat-v1.0,afternoon,1.8381905496120452,0.7359745502471924,3.315751552581787,308.795,0.0,448.0,51.08,0.0,75.0
+TinyLlama/TinyLlama-1.1B-Chat-v1.0,late afternoon,1.7547113946505954,0.7477977275848389,2.7170450687408447,285.17857142857144,0.0,448.0,46.9,0.0,75.0
+TinyLlama/TinyLlama-1.1B-Chat-v1.0,evening,1.7984187936782836,0.7911210060119629,2.242867946624756,313.99,0.0,448.0,51.96,0.0,75.0
+TinyLlama/TinyLlama-1.1B-Chat-v1.0,late evening,1.5875422928068372,0.7482233047485352,3.0341005325317383,192.78333333333333,0.0,448.0,31.261111111111113,0.0,75.0
+TinyLlama/TinyLlama-1.1B-Chat-v1.0,midnight,1.6341248273849487,0.7456190586090088,2.231257200241089,210.2,0.0,448.0,34.2,0.0,74.0
+TinyLlama/TinyLlama-1.1B-Chat-v1.0,night,2.0128010153770446,1.78045654296875,2.3337035179138184,372.05,280.0,448.0,62.6,38.0,74.0
+Vicuna v1.5 (7B),early morning,3.3891975168596233,1.4861681461334229,8.496459245681763,327.3157894736842,165.0,385.0,37.70175438596491,1.0,66.0
+Vicuna v1.5 (7B),morning,5.204141834507817,1.4868037700653076,22.59853982925415,321.7826086956522,165.0,385.0,34.26086956521739,1.0,66.0
+Vicuna v1.5 (7B),afternoon,3.670576659115878,0.9365975856781006,13.137462615966797,385.25,102.0,1137.0,47.77272727272727,1.0,198.0
+Vicuna v1.5 (7B),late afternoon,3.768600355495106,1.5326943397521973,8.839621782302856,324.27272727272725,165.0,385.0,35.77272727272727,1.0,66.0
+Vicuna v1.5 (7B),evening,3.0816855430603027,1.4715464115142822,6.439850091934204,329.0,165.0,385.0,42.833333333333336,1.0,66.0
+Vicuna v1.5 (7B),late evening,3.779375910758972,1.4631173610687256,8.507975101470947,319.1363636363636,165.0,385.0,32.86363636363637,1.0,66.0
+Vicuna v1.5 (7B),midnight,3.8164960656847273,1.5271682739257812,8.805369853973389,329.9047619047619,165.0,385.0,37.214285714285715,1.0,66.0
+Vicuna v1.5 (7B),night,3.648655335108439,1.4535038471221924,8.702306270599365,321.81944444444446,165.0,385.0,34.541666666666664,1.0,66.0
+WizardLM v1.2 (13B),early morning,5.290586443914884,1.9142282009124756,11.821624279022217,309.17391304347825,119.0,917.0,35.17391304347826,1.0,131.0
+WizardLM v1.2 (13B),morning,4.6954833320949385,2.2222862243652344,8.487712860107422,261.04347826086956,119.0,396.0,28.652173913043477,1.0,66.0
+WizardLM v1.2 (13B),afternoon,7.0199061699127885,0.5244314670562744,44.715752840042114,338.5168539325843,3.0,1662.0,30.03370786516854,1.0,124.0
+WizardLM v1.2 (13B),late afternoon,4.935352149217025,2.327338218688965,8.531365871429443,265.0869565217391,119.0,396.0,29.043478260869566,1.0,66.0
+WizardLM v1.2 (13B),evening,4.939989816058766,2.3061435222625732,8.714090824127197,262.1818181818182,119.0,396.0,28.136363636363637,1.0,66.0
+WizardLM v1.2 (13B),late evening,5.11581133660816,1.962632417678833,9.275582075119019,258.57142857142856,119.0,396.0,26.285714285714285,1.0,66.0
+WizardLM v1.2 (13B),midnight,5.255108930847862,1.6706929206848145,11.424241065979004,283.20454545454544,119.0,917.0,30.681818181818183,1.0,131.0
+WizardLM v1.2 (13B),night,5.411002513689873,1.687584400177002,13.038219928741455,319.5128205128205,119.0,917.0,36.84615384615385,1.0,131.0
+chat-bison (PaLM 2),early morning,2.4665334616388592,2.0992136001586914,3.2351021766662598,381.9214285714286,124.0,873.0,60.892857142857146,25.0,129.0
+chat-bison (PaLM 2),morning,2.488477897644043,1.9684116840362549,3.46305775642395,381.73,124.0,873.0,60.88,25.0,129.0
+chat-bison (PaLM 2),afternoon,2.6460144804074215,1.8480100631713867,6.029164552688599,376.28076923076924,124.0,1416.0,55.965384615384615,6.0,227.0
+chat-bison (PaLM 2),late afternoon,3.0387172081253744,2.1669023036956787,10.080488443374634,381.1363636363636,106.0,1416.0,53.35454545454545,4.0,227.0
+chat-bison (PaLM 2),evening,2.688272579908371,2.1424288749694824,8.430241346359253,367.07,124.0,873.0,55.89,6.0,129.0
+chat-bison (PaLM 2),late evening,2.7250528037548065,1.8091888427734375,5.203229904174805,382.725,110.0,1416.0,52.95,4.0,227.0
+chat-bison (PaLM 2),midnight,2.468383938074112,2.146925687789917,3.8768985271453857,381.95,124.0,873.0,60.9,25.0,129.0
+chat-bison (PaLM 2),night,2.460119960308075,2.13185977935791,4.069646596908569,381.92,124.0,873.0,60.94,25.0,129.0
+chat-bison-32k (PaLM 2 32K),early morning,9.141417106560299,2.3653366565704346,73.6822509765625,335.75,124.0,607.0,53.85,25.0,91.0
+chat-bison-32k (PaLM 2 32K),morning,7.7035503840446475,2.2696146965026855,58.11552977561951,335.75,124.0,607.0,53.85,25.0,91.0
+chat-bison-32k (PaLM 2 32K),afternoon,4.637674991688673,1.7499048709869385,56.01672887802124,339.73359073359075,0.0,1405.0,49.37065637065637,0.0,160.0
+chat-bison-32k (PaLM 2 32K),late afternoon,4.5047362284226855,2.5358686447143555,54.873910903930664,348.65909090909093,107.0,1405.0,48.35454545454545,5.0,160.0
+chat-bison-32k (PaLM 2 32K),evening,7.7332194912433625,1.9094345569610596,71.89037847518921,329.46,124.0,1405.0,49.5,6.0,91.0
+chat-bison-32k (PaLM 2 32K),late evening,7.431327997140549,2.328852415084839,74.75657105445862,350.40201005025125,107.0,1405.0,47.81909547738694,5.0,160.0
+chat-bison-32k (PaLM 2 32K),midnight,7.7498266498247785,2.1571295261383057,79.42942953109741,335.75,124.0,607.0,53.85,25.0,91.0
+chat-bison-32k (PaLM 2 32K),night,6.491292915344238,2.1961522102355957,64.5129017829895,335.75,124.0,607.0,53.85,25.0,91.0
+gemini-pro,early morning,2.7453590130460435,1.9099113941192627,4.615639686584473,386.7463768115942,158.0,849.0,61.84782608695652,21.0,125.0
+gemini-pro,morning,2.497767536007628,1.8429582118988037,3.7975666522979736,379.5204081632653,154.0,974.0,60.60204081632653,27.0,166.0
+gemini-pro,afternoon,2.816922114008949,1.8365321159362793,11.142733573913574,372.07142857142856,85.0,781.0,54.948412698412696,4.0,137.0
+gemini-pro,late afternoon,2.9268629640903114,1.6890861988067627,7.8114728927612305,378.3301886792453,117.0,868.0,51.990566037735846,1.0,137.0
+gemini-pro,evening,2.86901999375759,1.938537836074829,8.93311882019043,375.79487179487177,93.0,879.0,56.45641025641026,4.0,135.0
+gemini-pro,late evening,3.7189874940246117,1.8872959613800049,19.177364349365234,417.7709923664122,120.0,3077.0,54.87786259541985,1.0,494.0
+gemini-pro,midnight,3.338477972348531,1.8768093585968018,13.734686613082886,394.48,58.0,1044.0,59.64,4.0,159.0
+gemini-pro,night,2.8375814715210272,1.851557731628418,8.02761721611023,382.44897959183675,189.0,647.0,61.265306122448976,32.0,101.0
+gpt-3.5-turbo,early morning,3.884872453212738,1.044034481048584,25.384448051452637,397.935,45.0,2089.0,47.315,1.0,235.0
+gpt-3.5-turbo,morning,3.126271222697364,1.1247034072875977,27.072819709777832,389.9888888888889,72.0,2518.0,50.93888888888889,1.0,283.0
+gpt-3.5-turbo,afternoon,3.9458200880459375,0.9733002185821533,29.636699676513672,381.4428571428571,47.0,1516.0,43.76190476190476,1.0,189.0
+gpt-3.5-turbo,late afternoon,4.384064777692159,1.116605520248413,28.255154848098755,396.4357142857143,39.0,2234.0,43.28333333333333,1.0,306.0
+gpt-3.5-turbo,evening,3.5153889304115657,0.9087526798248291,29.323059797286987,385.18095238095236,39.0,2382.0,46.01428571428571,1.0,266.0
+gpt-3.5-turbo,late evening,4.166355226842148,0.8945813179016113,32.65510892868042,425.5339652448657,49.0,2333.0,45.62875197472354,1.0,284.0
+gpt-3.5-turbo,midnight,3.822115447632102,0.8915491104125977,29.661185264587402,430.02713178294573,7.0,2092.0,49.43023255813954,1.0,345.0
+gpt-3.5-turbo,night,3.7583211977969726,1.0493357181549072,28.81316113471985,418.4942084942085,67.0,1952.0,44.81467181467181,1.0,239.0
+gpt-4,early morning,14.044063893556595,3.3942410945892334,50.56782078742981,328.725,86.0,720.0,41.66,1.0,115.0
+gpt-4,morning,12.759107512468733,4.081010580062866,49.05775475502014,340.07262569832403,83.0,602.0,47.11173184357542,1.0,94.0
+gpt-4,afternoon,16.002364798386893,4.019714117050171,71.8665201663971,318.3095238095238,74.0,786.0,38.77142857142857,1.0,118.0
+gpt-4,late afternoon,16.80607506932254,3.969972610473633,75.70255947113037,314.85885167464113,76.0,694.0,37.67224880382775,1.0,110.0
+gpt-4,evening,13.841120740345547,4.07988715171814,41.534831285476685,318.0809523809524,84.0,667.0,40.19285714285714,1.0,110.0
+gpt-4,late evening,14.135588308695096,2.7110557556152344,56.553797006607056,315.61971830985914,85.0,719.0,37.23138832997988,1.0,120.0
+gpt-4,midnight,12.3578163115329,2.3194186687469482,46.18606233596802,336.36842105263156,93.0,1060.0,41.63386727688787,1.0,167.0
+gpt-4,night,12.813134506115546,2.3531341552734375,65.33208322525024,316.93461538461537,81.0,618.0,37.93076923076923,1.0,107.0
+gpt-4-turbo,early morning,9.229236473940839,1.2326269149780273,62.002402544021606,332.75757575757575,2.0,3064.0,44.28787878787879,0.0,491.0
+gpt-4-turbo,morning,8.381914901462467,1.2959718704223633,55.72310924530029,338.22159090909093,2.0,2407.0,51.28977272727273,0.0,382.0
+gpt-4-turbo,afternoon,11.318254935023296,1.331334114074707,77.09595465660095,343.74939759036147,1.0,3169.0,46.2144578313253,0.0,472.0
+gpt-4-turbo,late afternoon,11.024975039472627,1.437279462814331,65.85150051116943,342.29353233830847,2.0,3446.0,45.16417910447761,0.0,502.0
+gpt-4-turbo,evening,10.294425025763593,1.4023358821868896,79.08030676841736,359.74695863746956,3.0,4617.0,47.38929440389295,0.0,472.0
+gpt-4-turbo,late evening,9.514694256112435,1.1116414070129395,78.97449040412903,353.9368635437882,1.0,3581.0,45.54175152749491,0.0,554.0
+gpt-4-turbo,midnight,9.439968414174492,1.2613868713378906,78.63678312301636,363.84988452655887,2.0,3370.0,47.01154734411085,0.0,473.0
+gpt-4-turbo,night,7.816354261556633,1.3317677974700928,52.49111771583557,333.12252964426875,2.0,3221.0,46.92885375494071,0.0,473.0
+llama-2-70b-chat,early morning,2.8660141522424265,1.683544397354126,6.075578689575195,355.7280701754386,211.0,582.0,54.78947368421053,31.0,102.0
+llama-2-70b-chat,morning,2.872361832027194,1.5270755290985107,5.8158605098724365,358.67088607594934,211.0,582.0,55.0,31.0,102.0
+llama-2-70b-chat,afternoon,4.234376892130426,1.5692036151885986,16.153216123580933,415.3872340425532,65.0,1360.0,61.11489361702127,2.0,220.0
+llama-2-70b-chat,late afternoon,3.6833307104881365,1.5582563877105713,23.04676628112793,481.8484848484849,65.0,2621.0,66.05050505050505,1.0,435.0
+llama-2-70b-chat,evening,2.9706250462084185,1.680816650390625,8.144080400466919,350.54696132596683,65.0,759.0,52.3756906077348,2.0,124.0
+llama-2-70b-chat,late evening,4.3372269847176295,0.9715635776519775,23.42638874053955,597.4690909090909,0.0,7395.0,84.14909090909092,0.0,1278.0
+llama-2-70b-chat,midnight,3.249819871626402,1.1929240226745605,10.860423564910889,365.14473684210526,0.0,855.0,54.80921052631579,0.0,136.0
+llama-2-70b-chat,night,2.8264514451677147,1.512779712677002,5.105381965637207,356.4886363636364,211.0,582.0,55.02272727272727,31.0,102.0
+zephyr-7b-beta,early morning,3.937663261095683,0.7500736713409424,5.431238174438477,273.03333333333336,0.0,473.0,43.45,0.0,75.0
+zephyr-7b-beta,morning,4.056525647640228,3.4508540630340576,4.232520341873169,386.3,278.0,473.0,63.35,51.0,75.0
+zephyr-7b-beta,afternoon,3.5789777278900146,0.7476871013641357,4.351201772689819,277.6,0.0,490.0,44.016666666666666,0.0,75.0
+zephyr-7b-beta,late afternoon,3.4592524923459447,0.7312085628509521,4.547306537628174,251.1010101010101,0.0,490.0,39.292929292929294,0.0,75.0
+zephyr-7b-beta,evening,3.5673056403795878,0.7499518394470215,4.394542217254639,273.3666666666667,0.0,473.0,43.5,0.0,75.0
+zephyr-7b-beta,late evening,2.845338335577047,0.7462735176086426,4.675246477127075,174.874213836478,0.0,490.0,26.955974842767297,0.0,75.0
+zephyr-7b-beta,midnight,3.7253047794103624,0.7638051509857178,8.057972431182861,217.82083333333333,0.0,490.0,33.67916666666667,0.0,75.0
+zephyr-7b-beta,night,3.6954557319482166,0.76888108253479,5.414386987686157,216.55833333333334,0.0,473.0,33.46666666666667,0.0,75.0

pipeline/config.py

@@ -42,7 +42,7 @@ class MeasurementsConfig(Config):
 class QueriesDatasetConfig(Config):
     dataset_name: str = "GEM/xlsum"
     samples_per_measurement: int = 20
-    languages: List[str] = ["english", "japanese"]
+    languages: List[str] = ["english", "japanese", "ukrainian"]
     query_config: QueriesConfig = QueriesConfig()
 
 

pipeline/models.py

@@ -513,18 +513,16 @@ MODELS = [
     ),
 ]
 
-MISSING_MEASUREMENTS = ['Mistral-7B-Instruct-v0.2',
- 'Vicuna v1.5 (7B)',
- '01-ai Yi Chat (34B)',
- 'Chronos Hermes (13B)',
- 'LLaMA-2 Chat (7B)',
- 'Qwen 1.5 Chat (7B)',
- 'Snorkel Mistral PairRM DPO (7B)',
- 'OpenHermes-2.5-Mistral (7B)',
- 'Falcon Instruct (7B)',
- 'RedPajama-INCITE Chat (7B)',
- 'WizardLM v1.2 (13B)',
- 'Upstage SOLAR Instruct v1 (11B)']
+MISSING_MEASUREMENTS = ['gpt-3.5-turbo', 'zephyr-7b-beta', 'Mistral-7B-Instruct-v0.2',
+       'llama-2-70b-chat', 'Mixtral-8x7B-Instruct-v0.1',
+       'TinyLlama/TinyLlama-1.1B-Chat-v1.0', 'gemini-pro', 'gpt-4-turbo',
+       'gpt-4', 'chat-bison (PaLM 2)', 'chat-bison-32k (PaLM 2 32K)',
+       'Vicuna v1.5 (7B)', 'Mistral (7B) Instruct v0.2 (Together AI)',
+       '01-ai Yi Chat (34B)', 'Chronos Hermes (13B)', 'LLaMA-2 Chat (7B)',
+       'Qwen 1.5 Chat (7B)', 'Snorkel Mistral PairRM DPO (7B)',
+       'OpenHermes-2.5-Mistral (7B)', 'Falcon Instruct (7B)',
+       'RedPajama-INCITE Chat (7B)', 'WizardLM v1.2 (13B)',
+       'Upstage SOLAR Instruct v1 (11B)']
 
 def together_ai_price(billions_of_parameters):
     if billions_of_parameters <= 4: