piotr-szleg-bards-ai commited on
Commit
f156119
·
1 Parent(s): 80c27a4

2024-02-22 10:15:23 Publish script update

Browse files
data/time_of_day_comparison.csv CHANGED
@@ -1,5 +1,6 @@
1
  model,time_of_day,execution_time,characters_count,words_count
2
  01-ai Yi Chat (34B),early morning,8.896911223729452,394.1666666666667,40.125
 
3
  01-ai Yi Chat (34B),afternoon,12.255323665086614,512.3645833333334,51.59375
4
  01-ai Yi Chat (34B),late afternoon,9.417656523840767,384.9583333333333,40.125
5
  01-ai Yi Chat (34B),evening,8.16779062861488,360.8333333333333,39.375
@@ -7,6 +8,7 @@ model,time_of_day,execution_time,characters_count,words_count
7
  01-ai Yi Chat (34B),midnight,9.165155977816195,389.65,39.95
8
  01-ai Yi Chat (34B),night,9.463179730354472,413.125,40.5625
9
  Chronos Hermes (13B),early morning,7.70508497012289,461.4761904761905,49.595238095238095
 
10
  Chronos Hermes (13B),afternoon,6.651120054020601,539.5,76.10416666666667
11
  Chronos Hermes (13B),late afternoon,7.630303209478205,544.625,62.833333333333336
12
  Chronos Hermes (13B),evening,6.468730966250102,543.05,72.15
@@ -14,6 +16,7 @@ Chronos Hermes (13B),late evening,7.710561646355523,423.1363636363636,46.3181818
14
  Chronos Hermes (13B),midnight,7.482297485524958,501.55263157894734,58.89473684210526
15
  Chronos Hermes (13B),night,7.764214697091476,546.28,62.56
16
  Falcon Instruct (7B),early morning,1.1807806547297988,45.6875,8.4375
 
17
  Falcon Instruct (7B),afternoon,1.5774729509611387,36.96875,7.020833333333333
18
  Falcon Instruct (7B),late afternoon,1.0884015900748116,43.833333333333336,8.166666666666666
19
  Falcon Instruct (7B),evening,0.7874892950057983,39.2,7.4
@@ -21,6 +24,7 @@ Falcon Instruct (7B),late evening,1.2339241071180864,46.791666666666664,8.708333
21
  Falcon Instruct (7B),midnight,0.9522609934210777,37.45,7.075
22
  Falcon Instruct (7B),night,1.4049408435821533,50.6,9.32
23
  LLaMA-2 Chat (7B),early morning,2.8788051708884863,451.5833333333333,62.9375
 
24
  LLaMA-2 Chat (7B),afternoon,3.5475496424569024,477.9479166666667,53.135416666666664
25
  LLaMA-2 Chat (7B),late afternoon,2.7755608558654785,430.45,62.15
26
  LLaMA-2 Chat (7B),evening,2.9608939344232734,432.2916666666667,62.333333333333336
@@ -28,7 +32,7 @@ LLaMA-2 Chat (7B),late evening,2.869330001913983,452.7083333333333,63.0833333333
28
  LLaMA-2 Chat (7B),midnight,2.8979568664844217,459.075,63.15
29
  LLaMA-2 Chat (7B),night,6.160623960196972,424.70588235294116,57.029411764705884
30
  Mistral (7B) Instruct v0.2 (Together AI),early morning,3.611201712254728,509.75,64.39583333333333
31
- Mistral (7B) Instruct v0.2 (Together AI),morning,3.7677523457274145,501.96153846153845,64.25
32
  Mistral (7B) Instruct v0.2 (Together AI),afternoon,3.255509059895914,481.76237623762376,60.7970297029703
33
  Mistral (7B) Instruct v0.2 (Together AI),late afternoon,3.542298033617545,543.5138888888889,65.65277777777777
34
  Mistral (7B) Instruct v0.2 (Together AI),evening,5.496347131400273,507.48387096774195,63.774193548387096
@@ -50,6 +54,7 @@ Mixtral-8x7B-Instruct-v0.1,late evening,5.62397656769588,395.15714285714284,49.0
50
  Mixtral-8x7B-Instruct-v0.1,midnight,4.639010797279158,323.0394736842105,42.69210526315789
51
  Mixtral-8x7B-Instruct-v0.1,night,4.009439338194697,301.24545454545455,42.21818181818182
52
  OpenHermes-2.5-Mistral (7B),early morning,2.613155171275139,281.125,32.125
 
53
  OpenHermes-2.5-Mistral (7B),afternoon,3.382803752858152,357.8125,32.791666666666664
54
  OpenHermes-2.5-Mistral (7B),late afternoon,2.52796063820521,281.125,32.125
55
  OpenHermes-2.5-Mistral (7B),evening,2.7269538966092197,254.08333333333334,26.916666666666668
@@ -57,6 +62,7 @@ OpenHermes-2.5-Mistral (7B),late evening,2.720560759305954,281.125,32.125
57
  OpenHermes-2.5-Mistral (7B),midnight,2.6603748981769266,271.975,30.575
58
  OpenHermes-2.5-Mistral (7B),night,2.5896553993225098,277.06,31.96
59
  Qwen 1.5 Chat (7B),early morning,2.234119971593221,269.0625,30.125
 
60
  Qwen 1.5 Chat (7B),afternoon,2.539412996504042,344.93617021276594,37.37234042553192
61
  Qwen 1.5 Chat (7B),late afternoon,2.0715979735056558,269.0,30.125
62
  Qwen 1.5 Chat (7B),evening,2.325377941131592,244.375,27.25
@@ -64,6 +70,7 @@ Qwen 1.5 Chat (7B),late evening,1.9339114997697913,253.66666666666666,27.5416666
64
  Qwen 1.5 Chat (7B),midnight,2.3441089570522307,268.7,30.125
65
  Qwen 1.5 Chat (7B),night,1.8622158120075862,269.0,30.125
66
  RedPajama-INCITE Chat (7B),early morning,2.1123720943927764,52.82608695652174,4.3478260869565215
 
67
  RedPajama-INCITE Chat (7B),afternoon,1.8110081959854474,57.71875,5.302083333333333
68
  RedPajama-INCITE Chat (7B),late afternoon,1.7436921298503876,59.625,5.0
69
  RedPajama-INCITE Chat (7B),evening,1.964497913013805,55.125,4.583333333333333
@@ -71,6 +78,7 @@ RedPajama-INCITE Chat (7B),late evening,2.258105857031686,52.875,4.375
71
  RedPajama-INCITE Chat (7B),midnight,2.118554650126277,55.575,4.625
72
  RedPajama-INCITE Chat (7B),night,1.8566838451053784,55.38,4.7
73
  Snorkel Mistral PairRM DPO (7B),early morning,22.729273674335886,4691.333333333333,100.6875
 
74
  Snorkel Mistral PairRM DPO (7B),afternoon,34.586263124148054,5710.697916666667,541.0833333333334
75
  Snorkel Mistral PairRM DPO (7B),late afternoon,22.353231539328892,4714.0,104.375
76
  Snorkel Mistral PairRM DPO (7B),evening,24.748520737602597,4596.416666666667,92.33333333333333
@@ -86,6 +94,7 @@ TinyLlama/TinyLlama-1.1B-Chat-v1.0,late evening,1.5875422928068372,192.783333333
86
  TinyLlama/TinyLlama-1.1B-Chat-v1.0,midnight,1.6341248273849487,210.2,34.2
87
  TinyLlama/TinyLlama-1.1B-Chat-v1.0,night,2.0128010153770446,372.05,62.6
88
  Vicuna v1.5 (7B),early morning,3.347896182859266,263.45652173913044,29.847826086956523
 
89
  Vicuna v1.5 (7B),afternoon,3.670576659115878,353.1458333333333,43.791666666666664
90
  Vicuna v1.5 (7B),late afternoon,3.768600355495106,324.27272727272725,35.77272727272727
91
  Vicuna v1.5 (7B),evening,3.0816855430603027,296.1,38.55
@@ -93,6 +102,7 @@ Vicuna v1.5 (7B),late evening,3.779375910758972,292.5416666666667,30.125
93
  Vicuna v1.5 (7B),midnight,3.760787125996181,290.125,32.525
94
  Vicuna v1.5 (7B),night,3.626542616974224,296.0833333333333,31.458333333333332
95
  WizardLM v1.2 (13B),early morning,5.147532618564108,250.89583333333334,27.416666666666668
 
96
  WizardLM v1.2 (13B),afternoon,7.0199061699127885,320.51063829787233,28.43617021276596
97
  WizardLM v1.2 (13B),late afternoon,4.935352149217025,254.04166666666666,27.833333333333332
98
  WizardLM v1.2 (13B),evening,4.939989816058766,240.33333333333334,25.791666666666668
 
1
  model,time_of_day,execution_time,characters_count,words_count
2
  01-ai Yi Chat (34B),early morning,8.896911223729452,394.1666666666667,40.125
3
+ 01-ai Yi Chat (34B),morning,9.243995500647504,410.5,40.625
4
  01-ai Yi Chat (34B),afternoon,12.255323665086614,512.3645833333334,51.59375
5
  01-ai Yi Chat (34B),late afternoon,9.417656523840767,384.9583333333333,40.125
6
  01-ai Yi Chat (34B),evening,8.16779062861488,360.8333333333333,39.375
 
8
  01-ai Yi Chat (34B),midnight,9.165155977816195,389.65,39.95
9
  01-ai Yi Chat (34B),night,9.463179730354472,413.125,40.5625
10
  Chronos Hermes (13B),early morning,7.70508497012289,461.4761904761905,49.595238095238095
11
+ Chronos Hermes (13B),morning,8.221057146787643,595.375,64.5
12
  Chronos Hermes (13B),afternoon,6.651120054020601,539.5,76.10416666666667
13
  Chronos Hermes (13B),late afternoon,7.630303209478205,544.625,62.833333333333336
14
  Chronos Hermes (13B),evening,6.468730966250102,543.05,72.15
 
16
  Chronos Hermes (13B),midnight,7.482297485524958,501.55263157894734,58.89473684210526
17
  Chronos Hermes (13B),night,7.764214697091476,546.28,62.56
18
  Falcon Instruct (7B),early morning,1.1807806547297988,45.6875,8.4375
19
+ Falcon Instruct (7B),morning,1.4012710821060907,47.291666666666664,8.75
20
  Falcon Instruct (7B),afternoon,1.5774729509611387,36.96875,7.020833333333333
21
  Falcon Instruct (7B),late afternoon,1.0884015900748116,43.833333333333336,8.166666666666666
22
  Falcon Instruct (7B),evening,0.7874892950057983,39.2,7.4
 
24
  Falcon Instruct (7B),midnight,0.9522609934210777,37.45,7.075
25
  Falcon Instruct (7B),night,1.4049408435821533,50.6,9.32
26
  LLaMA-2 Chat (7B),early morning,2.8788051708884863,451.5833333333333,62.9375
27
+ LLaMA-2 Chat (7B),morning,2.678236266841059,455.0416666666667,60.791666666666664
28
  LLaMA-2 Chat (7B),afternoon,3.5475496424569024,477.9479166666667,53.135416666666664
29
  LLaMA-2 Chat (7B),late afternoon,2.7755608558654785,430.45,62.15
30
  LLaMA-2 Chat (7B),evening,2.9608939344232734,432.2916666666667,62.333333333333336
 
32
  LLaMA-2 Chat (7B),midnight,2.8979568664844217,459.075,63.15
33
  LLaMA-2 Chat (7B),night,6.160623960196972,424.70588235294116,57.029411764705884
34
  Mistral (7B) Instruct v0.2 (Together AI),early morning,3.611201712254728,509.75,64.39583333333333
35
+ Mistral (7B) Instruct v0.2 (Together AI),morning,3.6725051470205816,500.82894736842104,63.75
36
  Mistral (7B) Instruct v0.2 (Together AI),afternoon,3.255509059895914,481.76237623762376,60.7970297029703
37
  Mistral (7B) Instruct v0.2 (Together AI),late afternoon,3.542298033617545,543.5138888888889,65.65277777777777
38
  Mistral (7B) Instruct v0.2 (Together AI),evening,5.496347131400273,507.48387096774195,63.774193548387096
 
54
  Mixtral-8x7B-Instruct-v0.1,midnight,4.639010797279158,323.0394736842105,42.69210526315789
55
  Mixtral-8x7B-Instruct-v0.1,night,4.009439338194697,301.24545454545455,42.21818181818182
56
  OpenHermes-2.5-Mistral (7B),early morning,2.613155171275139,281.125,32.125
57
+ OpenHermes-2.5-Mistral (7B),morning,2.46380607287089,277.875,32.125
58
  OpenHermes-2.5-Mistral (7B),afternoon,3.382803752858152,357.8125,32.791666666666664
59
  OpenHermes-2.5-Mistral (7B),late afternoon,2.52796063820521,281.125,32.125
60
  OpenHermes-2.5-Mistral (7B),evening,2.7269538966092197,254.08333333333334,26.916666666666668
 
62
  OpenHermes-2.5-Mistral (7B),midnight,2.6603748981769266,271.975,30.575
63
  OpenHermes-2.5-Mistral (7B),night,2.5896553993225098,277.06,31.96
64
  Qwen 1.5 Chat (7B),early morning,2.234119971593221,269.0625,30.125
65
+ Qwen 1.5 Chat (7B),morning,1.8359179911406145,253.66666666666666,27.541666666666668
66
  Qwen 1.5 Chat (7B),afternoon,2.539412996504042,344.93617021276594,37.37234042553192
67
  Qwen 1.5 Chat (7B),late afternoon,2.0715979735056558,269.0,30.125
68
  Qwen 1.5 Chat (7B),evening,2.325377941131592,244.375,27.25
 
70
  Qwen 1.5 Chat (7B),midnight,2.3441089570522307,268.7,30.125
71
  Qwen 1.5 Chat (7B),night,1.8622158120075862,269.0,30.125
72
  RedPajama-INCITE Chat (7B),early morning,2.1123720943927764,52.82608695652174,4.3478260869565215
73
+ RedPajama-INCITE Chat (7B),morning,1.7558168431986934,57.375,4.791666666666667
74
  RedPajama-INCITE Chat (7B),afternoon,1.8110081959854474,57.71875,5.302083333333333
75
  RedPajama-INCITE Chat (7B),late afternoon,1.7436921298503876,59.625,5.0
76
  RedPajama-INCITE Chat (7B),evening,1.964497913013805,55.125,4.583333333333333
 
78
  RedPajama-INCITE Chat (7B),midnight,2.118554650126277,55.575,4.625
79
  RedPajama-INCITE Chat (7B),night,1.8566838451053784,55.38,4.7
80
  Snorkel Mistral PairRM DPO (7B),early morning,22.729273674335886,4691.333333333333,100.6875
81
+ Snorkel Mistral PairRM DPO (7B),morning,22.59587260087331,4714.0,104.375
82
  Snorkel Mistral PairRM DPO (7B),afternoon,34.586263124148054,5710.697916666667,541.0833333333334
83
  Snorkel Mistral PairRM DPO (7B),late afternoon,22.353231539328892,4714.0,104.375
84
  Snorkel Mistral PairRM DPO (7B),evening,24.748520737602597,4596.416666666667,92.33333333333333
 
94
  TinyLlama/TinyLlama-1.1B-Chat-v1.0,midnight,1.6341248273849487,210.2,34.2
95
  TinyLlama/TinyLlama-1.1B-Chat-v1.0,night,2.0128010153770446,372.05,62.6
96
  Vicuna v1.5 (7B),early morning,3.347896182859266,263.45652173913044,29.847826086956523
97
+ Vicuna v1.5 (7B),morning,5.204141834507817,308.375,32.833333333333336
98
  Vicuna v1.5 (7B),afternoon,3.670576659115878,353.1458333333333,43.791666666666664
99
  Vicuna v1.5 (7B),late afternoon,3.768600355495106,324.27272727272725,35.77272727272727
100
  Vicuna v1.5 (7B),evening,3.0816855430603027,296.1,38.55
 
102
  Vicuna v1.5 (7B),midnight,3.760787125996181,290.125,32.525
103
  Vicuna v1.5 (7B),night,3.626542616974224,296.0833333333333,31.458333333333332
104
  WizardLM v1.2 (13B),early morning,5.147532618564108,250.89583333333334,27.416666666666668
105
+ WizardLM v1.2 (13B),morning,4.6954833320949385,250.16666666666666,27.458333333333332
106
  WizardLM v1.2 (13B),afternoon,7.0199061699127885,320.51063829787233,28.43617021276596
107
  WizardLM v1.2 (13B),late afternoon,4.935352149217025,254.04166666666666,27.833333333333332
108
  WizardLM v1.2 (13B),evening,4.939989816058766,240.33333333333334,25.791666666666668
data/time_of_day_plots.csv CHANGED
The diff for this file is too large to render. See raw diff