Adding GLM, Qwen3 and some old models

static/leaderboard.csv

@@ -1,23 +1,33 @@
 Model,Ordinal (Win rate),Cardinal (Score),RO Stability,Stress,CFI,SRMR,RMSEA
-reka-flash-3,0.36177248677248675,0.4899723723251134,0.2885614355228928,0.21908343015733012,0.6356444444444445,0.32448055555555566,0.3265472222222222
-Llama-4-Scout-17B-16E-Instruct,0.49007936507936495,0.6180917882104732,0.4967998772578945,0.1945790922366818,0.6979444444444444,0.2716777777777778,0.2741527777777778
-Llama-3.3-70B-Instruct,0.7824074074074073,0.7826258982618856,0.8101440811870466,0.15086913865310292,0.7070694444444444,0.2484805555555556,0.2872888888888889
-Llama-3.1-70B-Instruct,0.703042328042328,0.7716257197180342,0.7627363693137972,0.1590258306220288,0.7405305555555556,0.2069805555555555,0.2524638888888888
-Llama-3.1-Nemotron-70B-Instruct,0.7843915343915344,0.806561245024004,0.7986269216633632,0.15639272646142155,0.7771694444444445,0.1574805555555555,0.2053138888888889
-Llama-3.1-8B-Instruct,0.4728835978835979,0.6200396929619997,0.5544135256446929,0.177589336660551,0.6203138888888889,0.3474166666666667,0.3526444444444444
-Llama-3.2-3B-Instruct,0.3498677248677249,0.4920203960749596,0.3095420768898777,0.23477458340427637,0.6402583333333334,0.3671361111111112,0.34035277777777784
-Llama-3.2-1B-Instruct,0.14814814814814814,0.28494630130657267,0.017845599375419068,0.29245365371576204,0.4570666666666667,0.49829999999999997,0.458125
-Mistral-Large-Instruct-2411,0.6117724867724867,0.7334649247049873,0.7061158170869489,0.17181887070789792,0.7237666666666667,0.24669166666666675,0.262
-Mistral-Large-Instruct-2407,0.7096560846560847,0.7865403071348558,0.7786865940656633,0.18242669696158476,0.7757583333333334,0.19862499999999994,0.21713055555555572
-Mistral-Nemo-Instruct-2407,0.34589947089947093,0.5243749748919985,0.409719354830282,0.21289984240736382,0.565475,0.40046944444444443,0.39598333333333324
-Mistral-Small-3.1-24B-Instruct-2503,0.5582010582010581,0.7026337346865648,0.6848141294613206,0.17721164035276304,0.6713749999999999,0.29685555555555565,0.3154944444444444
-QwQ-32B,0.7367724867724869,0.7719313363889678,0.8091295835194909,0.1769481985217758,0.6934333333333333,0.27074999999999994,0.30680277777777776
-Qwen2.5-VL-72B-Instruct,0.8716931216931216,0.8360251140700989,0.8125460511443046,0.15932218090531514,0.8454611111111111,0.10019722222222227,0.14792499999999997
-Qwen2.5-VL-7B-Instruct,0.22817460317460314,0.42422904460473104,0.28582477835186304,0.25948620101404807,0.45279444444444444,0.4787000000000001,0.464075
-Qwen2.5-VL-3B-Instruct,0.10052910052910052,0.2655381207197522,0.059657867275330144,0.2877042811210807,0.3650055555555556,0.6013444444444445,0.5902833333333335
-Qwen2.5-14B-Instruct-1M,0.5654761904761904,0.7022023055217502,0.6551137143166985,0.1785253019816836,0.7130944444444444,0.26870277777777785,0.26870277777777773
-Dracarys2-72B-Instruct,0.7533068783068784,0.7740958932029343,0.789501612210195,0.15836985877285903,0.7307833333333333,0.2581972222222222,0.27945555555555546
-Nautilus-70B-v0.1,0.5958994708994709,0.724110072700376,0.7188870305946458,0.1688175407755762,0.6729305555555555,0.2789694444444444,0.30781111111111104
-Cydonia-22B-v1.2,0.40674603174603174,0.5660454327134821,0.4926092723062008,0.20778751640583537,0.5617833333333333,0.40446111111111116,0.39160833333333345
-Ministrations-8B-v1,0.3220899470899471,0.4918972616356232,0.34889128317469287,0.2302731507248964,0.5484527777777779,0.4000972222222221,0.3784694444444444
-dummy,0.09722222222222221,0.2291015386716794,-0.009004148398032956,0.2928877637010999,0.3755222222222222,0.622275,0.5915305555555557
+GLM-4-32B-0414,0.45519713261648753,0.5924102038593776,0.5231398045089577,0.1885053649385876,0.5887694444444445,0.37460555555555564,0.37893611111111114
+Qwen3-235B-A22B-FP8,0.6115591397849462,0.7184051653356879,0.679310553271359,0.18145366817771102,0.7192055555555557,0.24603055555555564,0.2617222222222222
+Qwen3-32B-A3B,0.6666666666666666,0.748951174933817,0.695394000772024,0.17751938139533974,0.7652638888888889,0.18118611111111127,0.19652499999999995
+Qwen3-32B,0.6796594982078853,0.7364185588714854,0.7356392645359847,0.17307803161650182,0.6883833333333335,0.27474999999999994,0.29176388888888893
+Qwen3-8B,0.6115591397849462,0.7184453974322014,0.7185440961120562,0.19916042721283456,0.6796805555555555,0.29908055555555557,0.30805277777777773
+Qwen3-4B,0.5613799283154122,0.6970544522219364,0.659760116763837,0.2009520715020784,0.6930888888888889,0.27008333333333345,0.28465833333333335
+reka-flash-3,0.3853046594982079,0.4899723723251134,0.2885614355228928,0.21908343015733012,0.6356444444444445,0.32448055555555566,0.3265472222222222
+Llama-4-Scout-17B-16E-Instruct,0.5094086021505376,0.6180917882104732,0.4967998772578945,0.1945790922366818,0.6979444444444444,0.2716777777777778,0.2741527777777778
+Llama-3.3-70B-Instruct,0.8127240143369175,0.7826258982618856,0.8101440811870466,0.15086913865310292,0.7070694444444444,0.2484805555555556,0.2872888888888889
+Llama-3.1-70B-Instruct,0.739247311827957,0.7716257197180342,0.7627363693137972,0.1590258306220288,0.7405305555555556,0.2069805555555555,0.2524638888888888
+Llama-3.1-Nemotron-70B-Instruct,0.8234767025089607,0.806561245024004,0.7986269216633632,0.15639272646142155,0.7771694444444445,0.1574805555555555,0.2053138888888889
+Llama-3.1-8B-Instruct,0.5022401433691756,0.6200396929619997,0.5544135256446929,0.177589336660551,0.6203138888888889,0.3474166666666667,0.3526444444444444
+Llama-3.2-3B-Instruct,0.3781362007168459,0.4920203960749596,0.3095420768898777,0.23477458340427637,0.6402583333333334,0.3671361111111112,0.34035277777777784
+Llama-3.2-1B-Instruct,0.17338709677419353,0.28494630130657267,0.017845599375419068,0.29245365371576204,0.4570666666666667,0.49829999999999997,0.458125
+Mistral-Large-Instruct-2411,0.6563620071684587,0.7334649247049873,0.7061158170869489,0.17181887070789792,0.7237666666666667,0.24669166666666675,0.262
+Mistral-Large-Instruct-2407,0.7576164874551972,0.7865403071348558,0.7786865940656633,0.18242669696158476,0.7757583333333334,0.19862499999999994,0.21713055555555572
+Mistral-Nemo-Instruct-2407,0.3736559139784947,0.5243749748919985,0.409719354830282,0.21289984240736382,0.565475,0.40046944444444443,0.39598333333333324
+Mistral-Small-3.1-24B-Instruct-2503,0.6048387096774194,0.7026337346865648,0.6848141294613206,0.17721164035276304,0.6713749999999999,0.29685555555555565,0.3154944444444444
+Mistral-7B-Instruct-v0.2,0.12096774193548386,0.23355028783364112,0.026105642941629623,0.29118693576405585,0.3286527777777778,0.6504444444444444,0.6230416666666667
+Mixtral-8x7B-Instruct-v0.1,0.28763440860215056,0.41570329172979925,0.2810231146470137,0.24656890252743824,0.4673222222222222,0.5223027777777778,0.4969166666666667
+QwQ-32B,0.7665770609318997,0.7719313363889678,0.8091295835194909,0.1769481985217758,0.6934333333333333,0.27074999999999994,0.30680277777777776
+Qwen2.5-VL-72B-Instruct,0.8987455197132618,0.8360251140700989,0.8125460511443046,0.15932218090531514,0.8454611111111111,0.10019722222222227,0.14792499999999997
+Qwen2.5-VL-7B-Instruct,0.2594086021505376,0.42422904460473104,0.28582477835186304,0.25948620101404807,0.45279444444444444,0.4787000000000001,0.464075
+Qwen2.5-VL-3B-Instruct,0.132168458781362,0.2655381207197522,0.059657867275330144,0.2877042811210807,0.3650055555555556,0.6013444444444445,0.5902833333333335
+Qwen2.5-14B-Instruct-1M,0.5882616487455197,0.7022023055217502,0.6551137143166985,0.1785253019816836,0.7130944444444444,0.26870277777777785,0.26870277777777773
+phi-4,0.2181899641577061,0.3158767913757621,0.09850602595828868,0.25104588393816885,0.4269,0.5274305555555556,0.5154333333333333
+phi-3-medium-128k-instruct,0.1160394265232975,0.22454028939367154,-0.009343401612861558,0.2918818561769593,0.3458111111111111,0.6230527777777779,0.5971805555555556
+Dracarys2-72B-Instruct,0.7912186379928315,0.7740958932029343,0.789501612210195,0.15836985877285903,0.7307833333333333,0.2581972222222222,0.27945555555555546
+Nautilus-70B-v0.1,0.6424731182795698,0.724110072700376,0.7188870305946458,0.1688175407755762,0.6729305555555555,0.2789694444444444,0.30781111111111104
+Cydonia-22B-v1.2,0.418010752688172,0.5660454327134821,0.4926092723062008,0.20778751640583537,0.5617833333333333,0.40446111111111116,0.39160833333333345
+Ministrations-8B-v1,0.3387096774193548,0.4918972616356232,0.34889128317469287,0.2302731507248964,0.5484527777777779,0.4000972222222221,0.3784694444444444
+dummy,0.11379928315412184,0.2291015386716794,-0.009004148398032956,0.2928877637010999,0.3755222222222222,0.622275,0.5915305555555557

static/models_data/GLM-4-32B-0414/cfa_metrics.csv

@@ -0,0 +1,10 @@
+Context chunk,CFI,TLI,SRMR,RMSEA
+chunk_0,0.714025,0.7311749999999999,0.30745,0.269025
+chunk_1,0.6368,0.596375,0.325625,0.333775
+chunk_2,0.5689,0.5095999999999999,0.3386,0.340075
+chunk_3,0.447325,0.42979999999999996,0.54565,0.5361750000000001
+chunk_4,0.2274,0.221575,0.7733,0.7666
+chunk_chess_0,0.663625,0.63765,0.31845,0.31532499999999997
+chunk_grammar_1,0.587375,0.5273249999999999,0.328675,0.34140000000000004
+chunk_no_conv,0.8316749999999999,0.773575,0.09542500000000001,0.15935
+chunk_svs_no_conv,0.6218,0.5860749999999999,0.33827500000000005,0.3487

static/models_data/Mistral-7B-Instruct-v0.2/cfa_metrics.csv

@@ -1,10 +1,10 @@
 Context chunk,CFI,TLI,SRMR,RMSEA
-chunk_0,0.2286,0.215425,0.769275,0.767
-chunk_1,0.489025,0.574625,0.33595,0.28885
-chunk_2,0.25,-0.61065,0.77235,0.75
-chunk_3,0.230025,0.1477,0.555075,0.5286249999999999
-chunk_4,0.8181,0.8782500000000001,0.1053,0.034625
-chunk_chess_0,0.46909999999999996,0.4561,0.5388499999999999,0.5301
-chunk_grammar_1,0.0,0.0,1.0,1.0
-chunk_no_conv,0.48325,0.480575,0.535775,0.5183
-chunk_svs_no_conv,0.45435000000000003,0.563225,0.542,0.51775
+chunk_0,0.0,0.0,1.0,1.0
+chunk_1,0.218575,0.210475,0.7756000000000001,0.755375
+chunk_2,0.676575,-1.228625,0.323525,0.2662
+chunk_3,0.4391,-0.10727500000000001,0.54785,0.5106999999999999
+chunk_4,0.227925,0.22225,0.7778499999999999,0.75515
+chunk_chess_0,0.39675,0.362575,0.55215,0.516675
+chunk_grammar_1,0.3744,-0.36275,0.5526249999999999,0.5240750000000001
+chunk_no_conv,0.4171,-0.008050000000000002,0.546,0.5115
+chunk_svs_no_conv,0.20745,0.1873,0.7784,0.7677

static/models_data/Mixtral-8x7B-Instruct-v0.1/cfa_metrics.csv

@@ -1,10 +1,10 @@
 Context chunk,CFI,TLI,SRMR,RMSEA
-chunk_0,0.55385,0.483175,0.32705,0.309875
-chunk_1,0.3198,0.23665,0.55555,0.55375
-chunk_2,0.465175,0.54145,0.542825,0.514175
-chunk_3,0.186075,0.169575,0.777075,0.7779499999999999
-chunk_4,0.434525,0.414225,0.5503,0.5361
-chunk_chess_0,0.408375,0.37695,0.559725,0.553825
-chunk_grammar_1,0.5,0.5588,0.54405,0.5
-chunk_no_conv,0.66535,0.6313500000000001,0.3213,0.33345
-chunk_svs_no_conv,0.5416,0.5277999999999999,0.353325,0.336
+chunk_0,0.673325,0.650575,0.31995,0.2783
+chunk_1,0.8479,0.818725,0.10035,0.04554999999999999
+chunk_2,0.4071,0.3673,0.55065,0.5336249999999999
+chunk_3,0.4052,0.7982750000000001,0.5474500000000001,0.528125
+chunk_4,0.0,0.0,1.0,1.0
+chunk_chess_0,0.218425,0.199,0.7734,0.787925
+chunk_grammar_1,0.667975,0.663175,0.32755,0.274975
+chunk_no_conv,0.485975,0.4795,0.539875,0.5237499999999999
+chunk_svs_no_conv,0.5,0.659675,0.5415,0.5

static/models_data/Qwen3-235B-A22B-FP8/cfa_metrics.csv

@@ -0,0 +1,10 @@
+Context chunk,CFI,TLI,SRMR,RMSEA
+chunk_0,0.702875,0.697425,0.31420000000000003,0.28565
+chunk_1,0.7716249999999999,0.673475,0.10987500000000001,0.15315
+chunk_2,0.6148,0.5755,0.326025,0.32575
+chunk_3,0.83,0.7633,0.1052,0.133675
+chunk_4,0.8997000000000002,0.863325,0.08554999999999999,0.10615
+chunk_chess_0,0.664925,0.624475,0.31895,0.35740000000000005
+chunk_grammar_1,0.632425,0.5948,0.327375,0.33545
+chunk_no_conv,0.69345,0.6711,0.306,0.33375
+chunk_svs_no_conv,0.66305,0.6449750000000001,0.3211,0.324525

static/models_data/Qwen3-32B-A3B/cfa_metrics.csv

@@ -0,0 +1,10 @@
+Context chunk,CFI,TLI,SRMR,RMSEA
+chunk_0,0.8257000000000001,0.7571249999999999,0.11432500000000001,0.124475
+chunk_1,0.687125,0.66195,0.31177499999999997,0.32182499999999997
+chunk_2,0.8473250000000001,0.792375,0.10077499999999999,0.104725
+chunk_3,0.810775,0.737275,0.11862500000000001,0.140375
+chunk_4,0.81445,0.7428250000000001,0.10757499999999999,0.153675
+chunk_chess_0,0.80805,0.7309749999999999,0.1145,0.155075
+chunk_grammar_1,0.5756,0.531875,0.33992500000000003,0.334675
+chunk_no_conv,0.8491249999999999,0.799575,0.104275,0.12614999999999998
+chunk_svs_no_conv,0.669225,0.660925,0.3189,0.30775

static/models_data/Qwen3-32B/cfa_metrics.csv

@@ -0,0 +1,10 @@
+Context chunk,CFI,TLI,SRMR,RMSEA
+chunk_0,0.6745,0.6556,0.315025,0.323575
+chunk_1,0.386425,0.3537,0.5589999999999999,0.57215
+chunk_2,0.633,0.595975,0.323925,0.338825
+chunk_3,0.8520749999999999,0.8076999999999999,0.101925,0.108975
+chunk_4,0.607675,0.5660999999999999,0.3341,0.357575
+chunk_chess_0,0.6383500000000001,0.598075,0.33045,0.363225
+chunk_grammar_1,0.6745000000000001,0.643825,0.31825000000000003,0.34875
+chunk_no_conv,0.863025,0.8203250000000001,0.10187500000000001,0.12295
+chunk_svs_no_conv,0.8659,0.88045,0.0882,0.08985

static/models_data/Qwen3-4B/cfa_metrics.csv

@@ -0,0 +1,10 @@
+Context chunk,CFI,TLI,SRMR,RMSEA
+chunk_0,0.722525,0.7160249999999999,0.30675,0.28235
+chunk_1,0.840125,0.7831250000000001,0.09225,0.113375
+chunk_2,0.6275,0.5741499999999999,0.32894999999999996,0.35135000000000005
+chunk_3,0.65655,0.6145,0.31522500000000003,0.346825
+chunk_4,0.890425,0.8535999999999999,0.090775,0.091775
+chunk_chess_0,0.60715,0.558975,0.32799999999999996,0.36005
+chunk_grammar_1,0.643,0.61375,0.31472500000000003,0.33559999999999995
+chunk_no_conv,0.8724,0.83055,0.095525,0.11525000000000002
+chunk_svs_no_conv,0.378125,0.33952499999999997,0.55855,0.56535

static/models_data/Qwen3-8B/cfa_metrics.csv

@@ -0,0 +1,10 @@
+Context chunk,CFI,TLI,SRMR,RMSEA
+chunk_0,0.64905,0.597325,0.330575,0.33655
+chunk_1,0.683925,0.649275,0.30955,0.3251
+chunk_2,0.4252,0.39415,0.54635,0.5782499999999999
+chunk_3,0.70095,0.682475,0.31115,0.310875
+chunk_4,0.6596,0.62405,0.317875,0.3398
+chunk_chess_0,0.6003499999999999,0.5556,0.35495,0.3633
+chunk_grammar_1,0.66255,0.63755,0.32585,0.31055
+chunk_no_conv,0.902775,0.8695,0.09592500000000001,0.11175
+chunk_svs_no_conv,0.8327249999999999,0.828125,0.09949999999999999,0.0963

static/models_data/phi-3-medium-128k-instruct/cfa_metrics.csv

@@ -1,10 +1,10 @@
 Context chunk,CFI,TLI,SRMR,RMSEA
-chunk_0,0.663925,0.659125,0.31810000000000005,0.296275
-chunk_1,0.610375,0.562825,0.32152500000000006,0.31379999999999997
-chunk_2,0.684725,0.7767000000000001,0.312925,0.29450000000000004
-chunk_3,0.23005,0.2236,0.77475,0.76375
-chunk_4,0.0,0.0,1.0,1.0
-chunk_chess_0,0.42235,-4.0112749999999995,0.558075,0.525575
-chunk_grammar_1,0.4527,0.438125,0.552025,0.51625
-chunk_no_conv,0.60585,0.563825,0.33645,0.3627
-chunk_svs_no_conv,0.202325,0.17975,0.7791,0.7707
+chunk_0,0.5,0.674025,0.543775,0.5
+chunk_1,0.5,0.557375,0.541375,0.5
+chunk_2,0.14955,0.123625,0.7763249999999999,0.7654000000000001
+chunk_3,0.33135,0.28185000000000004,0.5547500000000001,0.53345
+chunk_4,0.25,-0.099775,0.7655,0.75
+chunk_chess_0,0.198875,0.11005000000000001,0.5578000000000001,0.533675
+chunk_grammar_1,0.75,1.19605,0.317375,0.25
+chunk_no_conv,0.0,0.0,1.0,1.0
+chunk_svs_no_conv,0.432525,0.4154,0.550575,0.5421

static/models_data/phi-4/cfa_metrics.csv

@@ -1,10 +1,10 @@
 Context chunk,CFI,TLI,SRMR,RMSEA
-chunk_0,0.4268,0.278175,0.335675,0.30652500000000005
-chunk_1,0.256475,0.188575,0.5588500000000001,0.538825
-chunk_2,0.25,-0.950775,0.77345,0.75
-chunk_3,0.5,5.5174,0.547825,0.5
-chunk_4,0.453,0.529825,0.54595,0.512925
-chunk_chess_0,0.221375,0.212125,0.77405,0.767375
-chunk_grammar_1,0.45230000000000004,0.57365,0.5435749999999999,0.509375
-chunk_no_conv,0.856425,0.806825,0.0991,0.13470000000000001
-chunk_svs_no_conv,0.29569999999999996,0.249925,0.59325,0.59805
+chunk_0,0.14515,0.111225,0.77815,0.773025
+chunk_1,0.0,0.0,1.0,1.0
+chunk_2,0.1902,0.174775,0.7784249999999999,0.7609250000000001
+chunk_3,0.61505,0.6323,0.32152500000000006,0.27362499999999995
+chunk_4,0.75,-0.507775,0.3112,0.25
+chunk_chess_0,0.6193500000000001,0.58185,0.32105,0.33265
+chunk_grammar_1,0.137225,0.10075,0.7762,0.76455
+chunk_no_conv,0.8282,0.78405,0.10767499999999999,0.1227
+chunk_svs_no_conv,0.556925,0.503875,0.35265,0.361425