grg commited on
Commit
cf75424
1 Parent(s): 7d902d9
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. static/leaderboard.csv +27 -25
  2. static/models_data/Mistral-7B-Instruct-v0.1/cfa_metrics.csv +4 -4
  3. static/models_data/Mistral-7B-Instruct-v0.1/matrix.svg +24 -24
  4. static/models_data/Mistral-7B-Instruct-v0.1/ranks.svg +0 -0
  5. static/models_data/Mistral-7B-Instruct-v0.1/structure.svg +0 -0
  6. static/models_data/Mistral-7B-Instruct-v0.2/cfa_metrics.csv +7 -7
  7. static/models_data/Mistral-7B-Instruct-v0.2/matrix.svg +24 -24
  8. static/models_data/Mistral-7B-Instruct-v0.2/ranks.svg +0 -0
  9. static/models_data/Mistral-7B-Instruct-v0.2/structure.svg +0 -0
  10. static/models_data/Mistral-7B-Instruct-v0.3/cfa_metrics.csv +4 -4
  11. static/models_data/Mistral-7B-Instruct-v0.3/matrix.svg +24 -24
  12. static/models_data/Mistral-7B-Instruct-v0.3/ranks.svg +0 -0
  13. static/models_data/Mistral-7B-Instruct-v0.3/structure.svg +0 -0
  14. static/models_data/Mistral-Large-Instruct-2407/cfa_metrics.csv +5 -5
  15. static/models_data/Mistral-Large-Instruct-2407/matrix.svg +24 -24
  16. static/models_data/Mistral-Large-Instruct-2407/ranks.svg +0 -0
  17. static/models_data/Mistral-Large-Instruct-2407/structure.svg +0 -0
  18. static/models_data/Mistral-Small-Instruct-2409/cfa_metrics.csv +5 -5
  19. static/models_data/Mistral-Small-Instruct-2409/matrix.svg +24 -24
  20. static/models_data/Mistral-Small-Instruct-2409/ranks.svg +0 -0
  21. static/models_data/Mistral-Small-Instruct-2409/structure.svg +0 -0
  22. static/models_data/Mixtral-8x22B-Instruct-v0.1/cfa_metrics.csv +3 -3
  23. static/models_data/Mixtral-8x22B-Instruct-v0.1/matrix.svg +24 -24
  24. static/models_data/Mixtral-8x22B-Instruct-v0.1/ranks.svg +0 -0
  25. static/models_data/Mixtral-8x22B-Instruct-v0.1/structure.svg +0 -0
  26. static/models_data/Mixtral-8x7B-Instruct-v0.1/cfa_metrics.csv +7 -7
  27. static/models_data/Mixtral-8x7B-Instruct-v0.1/matrix.svg +24 -24
  28. static/models_data/Mixtral-8x7B-Instruct-v0.1/ranks.svg +0 -0
  29. static/models_data/Mixtral-8x7B-Instruct-v0.1/structure.svg +0 -0
  30. static/models_data/Qwen2-72B-Instruct/cfa_metrics.csv +7 -7
  31. static/models_data/Qwen2-72B-Instruct/matrix.svg +24 -24
  32. static/models_data/Qwen2-72B-Instruct/ranks.svg +0 -0
  33. static/models_data/Qwen2-72B-Instruct/structure.svg +0 -0
  34. static/models_data/Qwen2-7B-Instruct/cfa_metrics.csv +9 -9
  35. static/models_data/Qwen2-7B-Instruct/matrix.svg +24 -24
  36. static/models_data/Qwen2-7B-Instruct/ranks.svg +0 -0
  37. static/models_data/Qwen2-7B-Instruct/structure.svg +0 -0
  38. static/models_data/Qwen2.5-0.5B-Instruct/cfa_metrics.csv +5 -5
  39. static/models_data/Qwen2.5-0.5B-Instruct/matrix.svg +24 -24
  40. static/models_data/Qwen2.5-0.5B-Instruct/ranks.svg +0 -0
  41. static/models_data/Qwen2.5-0.5B-Instruct/structure.svg +0 -0
  42. static/models_data/Qwen2.5-32B-Instruct/cfa_metrics.csv +5 -5
  43. static/models_data/Qwen2.5-32B-Instruct/matrix.svg +24 -24
  44. static/models_data/Qwen2.5-32B-Instruct/ranks.svg +0 -0
  45. static/models_data/Qwen2.5-32B-Instruct/structure.svg +0 -0
  46. static/models_data/Qwen2.5-72B-Instruct/cfa_metrics.csv +5 -5
  47. static/models_data/Qwen2.5-72B-Instruct/matrix.svg +24 -24
  48. static/models_data/Qwen2.5-72B-Instruct/ranks.svg +0 -0
  49. static/models_data/Qwen2.5-72B-Instruct/structure.svg +0 -0
  50. static/models_data/Qwen2.5-7B-Instruct/cfa_metrics.csv +6 -6
static/leaderboard.csv CHANGED
@@ -1,26 +1,28 @@
1
  Model,Ordinal (Win rate),Cardinal (Score),RO Stability,Stress,CFI,SRMR,RMSEA
2
- phi-3-mini-128k-instruct,0.34490740740740744,0.39387631706052895,0.039299993295009855,0.281800547806919,0.7509527777777777,0.25489166666666674,0.22045000000000003
3
- phi-3-medium-128k-instruct,0.34317129629629634,0.4026069526718651,0.09692037989916814,0.2651981204439735,0.6727694444444445,0.2984500000000001,0.2759472222222221
4
- Mistral-7B-Instruct-v0.1,0.1996527777777778,0.30674462188144647,0.027216280472015988,0.2829498135031582,0.500288888888889,0.45314444444444446,0.4191027777777777
5
- Mistral-7B-Instruct-v0.2,0.3755787037037038,0.4028886762146369,0.14417876497818388,0.265188983528973,0.5787944444444445,0.35010277777777776,0.3171083333333333
6
- Mistral-7B-Instruct-v0.3,0.2708333333333333,0.34429493368035685,0.07960539866974455,0.2742399030139009,0.5231444444444444,0.4214972222222223,0.3914694444444443
7
- Mixtral-8x7B-Instruct-v0.1,0.4496527777777778,0.47204265176392696,0.21473356319081474,0.2624402608740656,0.6766166666666665,0.25611666666666666,0.24065277777777772
8
- Mixtral-8x22B-Instruct-v0.1,0.26620370370370366,0.3496962191659786,0.1414001940345544,0.2548838005881672,0.45902777777777776,0.4849916666666666,0.4871833333333333
9
- command_r_plus,0.5815972222222222,0.5698450422762357,0.3429686514651868,0.23811982320641845,0.7772111111111112,0.17755277777777778,0.17465277777777777
10
- llama_3_8b_instruct,0.48900462962962954,0.5066363890459272,0.24527785038654715,0.245806400289881,0.7348277777777779,0.20952222222222228,0.20751944444444437
11
- llama_3_70b_instruct,0.7291666666666666,0.7270613281502669,0.607020698814379,0.18525883672204868,0.8298166666666668,0.10965277777777771,0.14649722222222217
12
- llama_3.1_8b_instruct,0.5434027777777778,0.5599895255443657,0.4295080949846363,0.22060228669473025,0.6379333333333334,0.3225500000000001,0.3328972222222223
13
- llama_3.1_70b_instruct,0.7847222222222222,0.7630277652278956,0.691365862744007,0.1709718847084183,0.8203805555555554,0.14023055555555552,0.17041944444444446
14
- llama_3.1_405b_instruct_4bit,0.6886574074074073,0.6993503239272297,0.7232098126552619,0.1702199925365422,0.6062611111111111,0.3538527777777777,0.38022500000000004
15
- Qwen2-7B-Instruct,0.43287037037037035,0.46812644016430927,0.25108519506513916,0.25776537005719313,0.6248583333333334,0.32358611111111113,0.3028361111111111
16
- Qwen2-72B-Instruct,0.5810185185185186,0.64867678910782,0.6465993243020925,0.20297742879025626,0.5559722222222221,0.3575638888888889,0.39241388888888884
17
- Qwen2.5-0.5B-Instruct,0.28877314814814814,0.3796838812739187,0.002970456550606876,0.2928913315666324,0.7497416666666666,0.24648888888888887,0.18477222222222223
18
- Qwen2.5-7B-Instruct,0.6186342592592592,0.5896473181421169,0.333554494486959,0.2505866550331236,0.8311222222222222,0.10302222222222213,0.09455277777777782
19
- Qwen2.5-32B-Instruct,0.7442129629629629,0.731635015756055,0.6724190751477237,0.1806656189868978,0.7584111111111111,0.19748055555555544,0.21686111111111106
20
- Qwen2.5-72B-Instruct,0.7991898148148148,0.754401345305127,0.6974116787371809,0.16176650806326276,0.7859583333333332,0.177875,0.2007527777777779
21
- gpt-3.5-turbo-0125,0.21643518518518517,0.328243163867074,0.08240359836763214,0.28728574920060357,0.4998916666666666,0.47583055555555553,0.4404444444444445
22
- gpt-4o-0513,0.7025462962962963,0.6713251724661671,0.5122163952167618,0.19201420113771173,0.7998694444444445,0.14606111111111109,0.1400583333333334
23
- gpt-4o-mini-2024-07-18,0.3628472222222222,0.40825697940501954,0.13575309046266867,0.2707065266105181,0.6141777777777777,0.32648055555555555,0.29394722222222214
24
- Mistral-Large-Instruct-2407,0.8217592592592592,0.7808285247091349,0.7644582301049158,0.16944638941325085,0.7604888888888888,0.18767499999999993,0.21457222222222228
25
- Mistral-Small-Instruct-2409,0.7083333333333335,0.7319149695591499,0.6416815833333804,0.1894343546381,0.7891722222222222,0.1387222222222222,0.17242222222222225
26
- dummy,0.14872685185185186,0.2784036220050126,-0.009004148398032956,0.2928877637010999,0.5076361111111111,0.4973388888888889,0.4541638888888889
 
 
 
1
  Model,Ordinal (Win rate),Cardinal (Score),RO Stability,Stress,CFI,SRMR,RMSEA
2
+ phi-3-mini-128k-instruct,0.33279914529914534,0.32984992817164005,0.039299993295009855,0.281800547806919,0.5861361111111111,0.42524166666666674,0.3974944444444444
3
+ phi-3-medium-128k-instruct,0.3333333333333333,0.30802986933853177,0.09692037989916814,0.2651981204439735,0.43025555555555556,0.5503277777777777,0.5381722222222222
4
+ phi-3.5-mini-instruct,0.2435897435897436,0.2680653144619754,0.0361229186530762,0.28422749224983457,0.40715555555555555,0.5721138888888888,0.5507833333333333
5
+ phi-3.5-MoE-instruct,0.40010683760683763,0.36128192067041315,0.10985291697837646,0.2739229692168671,0.5530944444444444,0.4248777777777778,0.40345
6
+ Mistral-7B-Instruct-v0.1,0.22168803418803418,0.26609566354811315,0.027216280472015988,0.2829498135031582,0.38917777777777773,0.5561138888888888,0.530213888888889
7
+ Mistral-7B-Instruct-v0.2,0.35683760683760685,0.32133832899241477,0.14417876497818388,0.265188983528973,0.3802722222222222,0.5727305555555555,0.5483611111111111
8
+ Mistral-7B-Instruct-v0.3,0.25961538461538464,0.26572479479146804,0.07960539866974455,0.2742399030139009,0.31385,0.6241,0.6081333333333333
9
+ Mixtral-8x7B-Instruct-v0.1,0.44925213675213665,0.3819009850972602,0.21473356319081474,0.2624402608740656,0.45275,0.5034666666666667,0.4905694444444444
10
+ Mixtral-8x22B-Instruct-v0.1,0.3477564102564103,0.31529864972153404,0.1414001940345544,0.2548838005881672,0.3772361111111111,0.5810888888888889,0.5844750000000001
11
+ command_r_plus,0.5726495726495726,0.4995356672762356,0.3429686514651868,0.23811982320641845,0.6033000000000001,0.3740166666666668,0.3667527777777777
12
+ llama_3_8b_instruct,0.4983974358974359,0.4295836112681494,0.24527785038654715,0.245806400289881,0.5498222222222222,0.42656388888888896,0.42189444444444446
13
+ llama_3_70b_instruct,0.7777777777777778,0.6839540364836003,0.607020698814379,0.18525883672204868,0.7210055555555557,0.2346083333333333,0.25758888888888887
14
+ llama_3.1_8b_instruct,0.5592948717948718,0.4786874422110324,0.4295080949846363,0.22060228669473025,0.4305722222222223,0.5455027777777777,0.553
15
+ llama_3.1_70b_instruct,0.8215811965811967,0.7172545013390067,0.691365862744007,0.1709718847084183,0.6979472222222223,0.2636777777777777,0.2907250000000001
16
+ llama_3.1_405b_instruct_4bit,0.7329059829059829,0.6490864350383405,0.7232098126552619,0.1702199925365422,0.4875722222222223,0.4963444444444445,0.5211555555555556
17
+ Qwen2-7B-Instruct,0.40651709401709396,0.36370005127542027,0.25108519506513916,0.25776537005719313,0.3560861111111111,0.6009722222222222,0.5920888888888889
18
+ Qwen2-72B-Instruct,0.5721153846153846,0.5461212335522644,0.6465993243020925,0.20297742879025626,0.3045,0.6543138888888889,0.6646361111111111
19
+ Qwen2.5-0.5B-Instruct,0.2954059829059829,0.3005554090516966,0.002970456550606876,0.2928913315666324,0.5371250000000001,0.44709722222222226,0.404575
20
+ Qwen2.5-7B-Instruct,0.6132478632478633,0.5163098181421168,0.333554494486959,0.2505866550331236,0.6473694444444444,0.30400277777777773,0.29651944444444434
21
+ Qwen2.5-32B-Instruct,0.7323717948717948,0.656917654644944,0.6724190751477237,0.1806656189868978,0.5603222222222223,0.40237500000000004,0.41161666666666663
22
+ Qwen2.5-72B-Instruct,0.8253205128205129,0.7104489147495714,0.6974116787371809,0.16176650806326276,0.6734583333333333,0.2993,0.3184472222222223
23
+ gpt-3.5-turbo-0125,0.24626068376068375,0.28218378886707396,0.08240359836763214,0.28728574920060357,0.3873055555555555,0.599925,0.572238888888889
24
+ gpt-4o-0513,0.6810897435897435,0.5989532974661671,0.5122163952167618,0.19201420113771173,0.6235416666666667,0.34458611111111115,0.3441805555555555
25
+ gpt-4o-mini-2024-07-18,0.3782051282051282,0.3418785071827972,0.13575309046266867,0.2707065266105181,0.44214722222222214,0.5004583333333332,0.47896666666666665
26
+ Mistral-Large-Instruct-2407,0.8472222222222222,0.7374229691535793,0.7644582301049158,0.16944638941325085,0.6510750000000001,0.31028611111111104,0.3297916666666667
27
+ Mistral-Small-Instruct-2409,0.7745726495726496,0.6890378862258165,0.6416815833333804,0.1894343546381,0.6840472222222221,0.2601583333333335,0.2888777777777778
28
+ dummy,0.18269230769230768,0.2291015386716794,-0.009004148398032956,0.2928877637010999,0.3755222222222222,0.622275,0.5915305555555557
static/models_data/Mistral-7B-Instruct-v0.1/cfa_metrics.csv CHANGED
@@ -1,10 +1,10 @@
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
  chunk_0,0.25,-0.33005,0.770075,0.75
3
- chunk_1,0.6990000000000001,0.10372499999999997,0.326125,0.2572
4
  chunk_2,0.4095,0.8663000000000001,0.556,0.527325
5
  chunk_3,0.25,0.412175,0.770225,0.75
6
- chunk_4,0.861675,15.901375,0.09445,0.01455
7
  chunk_chess_0,0.351325,0.278825,0.549475,0.5339999999999999
8
- chunk_grammar_1,0.25,-0.812475,0.773025,0.75
9
- chunk_no_conv,0.8223,0.7799999999999999,0.09634999999999999,0.061875
10
  chunk_svs_no_conv,0.6088,0.49317500000000003,0.142575,0.126975
 
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
  chunk_0,0.25,-0.33005,0.770075,0.75
3
+ chunk_1,0.449,0.11142499999999998,0.555375,0.5072
4
  chunk_2,0.4095,0.8663000000000001,0.556,0.527325
5
  chunk_3,0.25,0.412175,0.770225,0.75
6
+ chunk_4,0.611675,14.6198,0.328625,0.26455
7
  chunk_chess_0,0.351325,0.278825,0.549475,0.5339999999999999
8
+ chunk_grammar_1,0.0,0.0,1.0,1.0
9
+ chunk_no_conv,0.5723,0.5137499999999999,0.332675,0.311875
10
  chunk_svs_no_conv,0.6088,0.49317500000000003,0.142575,0.126975
static/models_data/Mistral-7B-Instruct-v0.1/matrix.svg CHANGED
static/models_data/Mistral-7B-Instruct-v0.1/ranks.svg CHANGED
static/models_data/Mistral-7B-Instruct-v0.1/structure.svg CHANGED
static/models_data/Mistral-7B-Instruct-v0.2/cfa_metrics.csv CHANGED
@@ -1,10 +1,10 @@
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
- chunk_0,0.4786,0.602475,0.542475,0.517
3
  chunk_1,0.489025,0.574625,0.33595,0.28885
4
- chunk_2,0.455675,-0.43225,0.547675,0.519625
5
- chunk_3,0.48002500000000003,0.664025,0.33125,0.278625
6
  chunk_4,0.8181,0.8782500000000001,0.1053,0.034625
7
- chunk_chess_0,0.8186,0.762425,0.10205,0.0945
8
- chunk_grammar_1,0.169775,0.149075,0.777725,0.770725
9
- chunk_no_conv,0.9339500000000001,0.916125,0.08465,0.059
10
- chunk_svs_no_conv,0.5654,0.64235,0.32384999999999997,0.291025
 
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
+ chunk_0,0.2286,0.215425,0.769275,0.767
3
  chunk_1,0.489025,0.574625,0.33595,0.28885
4
+ chunk_2,0.25,-0.61065,0.77235,0.75
5
+ chunk_3,0.230025,0.1477,0.555075,0.5286249999999999
6
  chunk_4,0.8181,0.8782500000000001,0.1053,0.034625
7
+ chunk_chess_0,0.46909999999999996,0.4561,0.5388499999999999,0.5301
8
+ chunk_grammar_1,0.0,0.0,1.0,1.0
9
+ chunk_no_conv,0.48325,0.480575,0.535775,0.5183
10
+ chunk_svs_no_conv,0.45435000000000003,0.563225,0.542,0.51775
static/models_data/Mistral-7B-Instruct-v0.2/matrix.svg CHANGED
static/models_data/Mistral-7B-Instruct-v0.2/ranks.svg CHANGED
static/models_data/Mistral-7B-Instruct-v0.2/structure.svg CHANGED
static/models_data/Mistral-7B-Instruct-v0.3/cfa_metrics.csv CHANGED
@@ -2,9 +2,9 @@ Context chunk,CFI,TLI,SRMR,RMSEA
2
  chunk_0,0.5,0.718825,0.5425,0.5
3
  chunk_1,0.535775,0.34664999999999996,0.1094,0.0532
4
  chunk_2,0.18365,0.03275,0.566275,0.58125
5
- chunk_3,0.4505,0.43665,0.548,0.519775
6
- chunk_4,0.497675,0.64855,0.5425500000000001,0.5026999999999999
7
- chunk_chess_0,0.964675,0.986,0.084375,0.018250000000000002
8
  chunk_grammar_1,0.25,-2.811525,0.77145,0.75
9
  chunk_no_conv,0.8635999999999999,0.82725,0.08715,0.082875
10
- chunk_svs_no_conv,0.462425,0.52325,0.541775,0.5151749999999999
 
2
  chunk_0,0.5,0.718825,0.5425,0.5
3
  chunk_1,0.535775,0.34664999999999996,0.1094,0.0532
4
  chunk_2,0.18365,0.03275,0.566275,0.58125
5
+ chunk_3,0.0,0.0,1.0,1.0
6
+ chunk_4,0.0,0.0,1.0,1.0
7
+ chunk_chess_0,0.241625,0.238925,0.774275,0.755875
8
  chunk_grammar_1,0.25,-2.811525,0.77145,0.75
9
  chunk_no_conv,0.8635999999999999,0.82725,0.08715,0.082875
10
+ chunk_svs_no_conv,0.25,0.32365,0.76585,0.75
static/models_data/Mistral-7B-Instruct-v0.3/matrix.svg CHANGED
static/models_data/Mistral-7B-Instruct-v0.3/ranks.svg CHANGED
static/models_data/Mistral-7B-Instruct-v0.3/structure.svg CHANGED
static/models_data/Mistral-Large-Instruct-2407/cfa_metrics.csv CHANGED
@@ -1,10 +1,10 @@
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
  chunk_0,0.6065,0.5587500000000001,0.34195000000000003,0.3913
3
- chunk_1,0.63895,0.6054,0.33255,0.36092500000000005
4
- chunk_2,0.7953749999999999,0.7326750000000001,0.126225,0.15475000000000003
5
- chunk_3,0.6329,0.5987,0.3178500000000001,0.337525
6
- chunk_4,0.82855,0.770825,0.13394999999999999,0.163675
7
- chunk_chess_0,0.85985,0.8174,0.10362500000000001,0.12497500000000002
8
  chunk_grammar_1,0.8351500000000001,0.7926250000000001,0.116725,0.13017499999999999
9
  chunk_no_conv,0.8900250000000001,0.85555,0.09050000000000001,0.115975
10
  chunk_svs_no_conv,0.7571,0.7157250000000001,0.1257,0.15184999999999998
 
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
  chunk_0,0.6065,0.5587500000000001,0.34195000000000003,0.3913
3
+ chunk_1,0.452225,0.435025,0.5444,0.5771999999999999
4
+ chunk_2,0.604625,0.5611,0.34990000000000004,0.341325
5
+ chunk_3,0.438675,0.42252500000000004,0.5419,0.535475
6
+ chunk_4,0.6441250000000001,0.6033499999999999,0.35117499999999996,0.37655
7
+ chunk_chess_0,0.63125,0.5975,0.330325,0.348275
8
  chunk_grammar_1,0.8351500000000001,0.7926250000000001,0.116725,0.13017499999999999
9
  chunk_no_conv,0.8900250000000001,0.85555,0.09050000000000001,0.115975
10
  chunk_svs_no_conv,0.7571,0.7157250000000001,0.1257,0.15184999999999998
static/models_data/Mistral-Large-Instruct-2407/matrix.svg CHANGED
static/models_data/Mistral-Large-Instruct-2407/ranks.svg CHANGED
static/models_data/Mistral-Large-Instruct-2407/structure.svg CHANGED
static/models_data/Mistral-Small-Instruct-2409/cfa_metrics.csv CHANGED
@@ -1,10 +1,10 @@
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
  chunk_0,0.789825,0.71305,0.12095,0.167425
3
  chunk_1,0.862975,0.8278750000000001,0.086675,0.110275
4
- chunk_2,0.851325,0.796775,0.108275,0.132775
5
- chunk_3,0.6166499999999999,0.572825,0.33632500000000004,0.355075
6
  chunk_4,0.830525,0.769525,0.11827499999999999,0.15587499999999999
7
- chunk_chess_0,0.823325,0.765625,0.10767499999999999,0.147825
8
- chunk_grammar_1,0.77625,0.696125,0.12664999999999998,0.1973
9
  chunk_no_conv,0.8255,0.7659750000000001,0.12215000000000001,0.163975
10
- chunk_svs_no_conv,0.726175,0.6696500000000001,0.121525,0.121275
 
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
  chunk_0,0.789825,0.71305,0.12095,0.167425
3
  chunk_1,0.862975,0.8278750000000001,0.086675,0.110275
4
+ chunk_2,0.64195,0.6005499999999999,0.335725,0.34099999999999997
5
+ chunk_3,0.42010000000000003,0.390075,0.5437500000000001,0.57215
6
  chunk_4,0.830525,0.769525,0.11827499999999999,0.15587499999999999
7
+ chunk_chess_0,0.648375,0.61005,0.32037499999999997,0.36169999999999997
8
+ chunk_grammar_1,0.556125,0.4944,0.355875,0.4111
9
  chunk_no_conv,0.8255,0.7659750000000001,0.12215000000000001,0.163975
10
+ chunk_svs_no_conv,0.5810500000000001,0.5474749999999999,0.33765,0.3164
static/models_data/Mistral-Small-Instruct-2409/matrix.svg CHANGED
static/models_data/Mistral-Small-Instruct-2409/ranks.svg CHANGED
static/models_data/Mistral-Small-Instruct-2409/structure.svg CHANGED
static/models_data/Mixtral-8x22B-Instruct-v0.1/cfa_metrics.csv CHANGED
@@ -1,10 +1,10 @@
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
  chunk_0,0.25,0.30345,0.7763249999999999,0.75
3
- chunk_1,0.5,0.75465,0.54055,0.5
4
  chunk_2,0.382325,0.33275,0.5517,0.53875
5
  chunk_3,0.20485,0.177075,0.77485,0.771825
6
  chunk_4,0.1417,0.07505,0.77405,0.777825
7
  chunk_chess_0,0.7984,0.7256,0.117975,0.14932499999999999
8
- chunk_grammar_1,0.5304249999999999,0.45715,0.333975,0.33069999999999994
9
  chunk_no_conv,0.81835,0.751525,0.11614999999999999,0.172175
10
- chunk_svs_no_conv,0.5052,0.42625,0.37935,0.39404999999999996
 
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
  chunk_0,0.25,0.30345,0.7763249999999999,0.75
3
+ chunk_1,0.25,0.461775,0.7741,0.75
4
  chunk_2,0.382325,0.33275,0.5517,0.53875
5
  chunk_3,0.20485,0.177075,0.77485,0.771825
6
  chunk_4,0.1417,0.07505,0.77405,0.777825
7
  chunk_chess_0,0.7984,0.7256,0.117975,0.14932499999999999
8
+ chunk_grammar_1,0.2035,0.1915,0.7756000000000001,0.7658
9
  chunk_no_conv,0.81835,0.751525,0.11614999999999999,0.172175
10
+ chunk_svs_no_conv,0.346,0.287925,0.5690500000000001,0.5845750000000001
static/models_data/Mixtral-8x22B-Instruct-v0.1/matrix.svg CHANGED
static/models_data/Mixtral-8x22B-Instruct-v0.1/ranks.svg CHANGED
static/models_data/Mixtral-8x22B-Instruct-v0.1/structure.svg CHANGED
static/models_data/Mixtral-8x7B-Instruct-v0.1/cfa_metrics.csv CHANGED
@@ -1,10 +1,10 @@
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
  chunk_0,0.55385,0.483175,0.32705,0.309875
3
- chunk_1,0.7623,0.67495,0.10830000000000001,0.074675
4
- chunk_2,0.62615,0.67945,0.318625,0.283675
5
- chunk_3,0.64055,0.6078,0.326725,0.31475
6
- chunk_4,0.585725,0.504625,0.32845,0.32605
7
- chunk_chess_0,0.779525,0.696675,0.12285000000000001,0.13640000000000002
8
- chunk_grammar_1,0.730425,0.777175,0.315375,0.2685
9
- chunk_no_conv,0.869425,0.82355,0.10435,0.11595
10
  chunk_svs_no_conv,0.5416,0.5277999999999999,0.353325,0.336
 
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
  chunk_0,0.55385,0.483175,0.32705,0.309875
3
+ chunk_1,0.3198,0.23665,0.55555,0.55375
4
+ chunk_2,0.465175,0.54145,0.542825,0.514175
5
+ chunk_3,0.186075,0.169575,0.777075,0.7779499999999999
6
+ chunk_4,0.434525,0.414225,0.5503,0.5361
7
+ chunk_chess_0,0.408375,0.37695,0.559725,0.553825
8
+ chunk_grammar_1,0.5,0.5588,0.54405,0.5
9
+ chunk_no_conv,0.66535,0.6313500000000001,0.3213,0.33345
10
  chunk_svs_no_conv,0.5416,0.5277999999999999,0.353325,0.336
static/models_data/Mixtral-8x7B-Instruct-v0.1/matrix.svg CHANGED
static/models_data/Mixtral-8x7B-Instruct-v0.1/ranks.svg CHANGED
static/models_data/Mixtral-8x7B-Instruct-v0.1/structure.svg CHANGED
static/models_data/Qwen2-72B-Instruct/cfa_metrics.csv CHANGED
@@ -1,10 +1,10 @@
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
- chunk_0,0.449025,0.44882500000000003,0.5454,0.52355
3
- chunk_1,0.37865,0.3358,0.5538750000000001,0.5716
4
- chunk_2,0.54545,0.48317499999999997,0.33475,0.38744999999999996
5
- chunk_3,0.55185,0.4898,0.32939999999999997,0.33795
6
- chunk_4,0.574425,0.5200250000000001,0.32570000000000005,0.38675
7
  chunk_chess_0,0.564125,0.498925,0.33515,0.41045
8
- chunk_grammar_1,0.52735,0.458825,0.3337,0.40790000000000004
9
- chunk_no_conv,0.819025,0.7562,0.12045,0.17712500000000003
10
  chunk_svs_no_conv,0.59385,0.5464,0.33965,0.32895
 
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
+ chunk_0,0.25,0.26295,0.76175,0.75
3
+ chunk_1,0.1609,0.1379,0.7806500000000001,0.784825
4
+ chunk_2,0.164125,0.141975,0.787175,0.7872
5
+ chunk_3,0.14295,0.108325,0.778875,0.78025
6
+ chunk_4,0.0,0.0,1.0,1.0
7
  chunk_chess_0,0.564125,0.498925,0.33515,0.41045
8
+ chunk_grammar_1,0.414225,0.39,0.54935,0.555875
9
+ chunk_no_conv,0.450325,0.424725,0.5562250000000001,0.584175
10
  chunk_svs_no_conv,0.59385,0.5464,0.33965,0.32895
static/models_data/Qwen2-72B-Instruct/matrix.svg CHANGED
static/models_data/Qwen2-72B-Instruct/ranks.svg CHANGED
static/models_data/Qwen2-72B-Instruct/structure.svg CHANGED
static/models_data/Qwen2-7B-Instruct/cfa_metrics.csv CHANGED
@@ -1,10 +1,10 @@
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
- chunk_0,0.416825,0.508725,0.551675,0.521725
3
- chunk_1,0.742175,0.85405,0.30195000000000005,0.260275
4
- chunk_2,0.5747,0.4995,0.322075,0.297
5
- chunk_3,0.7651,0.76035,0.10045000000000001,0.08145
6
- chunk_4,0.8802,0.8637750000000001,0.0938,0.0473
7
- chunk_chess_0,0.57085,0.508575,0.33772500000000005,0.343725
8
- chunk_grammar_1,0.68985,0.67535,0.31980000000000003,0.283225
9
- chunk_no_conv,0.808125,0.747325,0.10464999999999999,0.117475
10
- chunk_svs_no_conv,0.1759,0.1506,0.78015,0.77335
 
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
+ chunk_0,0.166825,0.145375,0.7784,0.771725
3
+ chunk_1,0.25,0.2553,0.767475,0.75
4
+ chunk_2,0.3247,0.243,0.554275,0.5469999999999999
5
+ chunk_3,0.5151,0.439625,0.335375,0.33145
6
+ chunk_4,0.6302,0.600725,0.32280000000000003,0.2973
7
+ chunk_chess_0,0.405825,0.37102500000000005,0.5605,0.5591999999999999
8
+ chunk_grammar_1,0.43984999999999996,0.40635,0.5452750000000001,0.5332250000000001
9
+ chunk_no_conv,0.472275,0.4575,0.5446500000000001,0.5388999999999999
10
+ chunk_svs_no_conv,0.0,0.0,1.0,1.0
static/models_data/Qwen2-7B-Instruct/matrix.svg CHANGED
static/models_data/Qwen2-7B-Instruct/ranks.svg CHANGED
static/models_data/Qwen2-7B-Instruct/structure.svg CHANGED
static/models_data/Qwen2.5-0.5B-Instruct/cfa_metrics.csv CHANGED
@@ -1,10 +1,10 @@
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
  chunk_0,0.7500249999999999,-2.21615,0.10572500000000001,0.0545
3
- chunk_1,0.75,1.25775,0.3161,0.25
4
  chunk_2,0.73175,0.9472249999999999,0.31827500000000003,0.2564
5
- chunk_3,0.75,1.0417,0.32295,0.25
6
- chunk_4,0.8639749999999999,0.9359999999999999,0.10375,0.0408
7
- chunk_chess_0,0.904775,0.28604999999999997,0.09230000000000001,0.020275
8
- chunk_grammar_1,0.75,5.354475,0.30865,0.25
9
  chunk_no_conv,0.600475,0.04259999999999997,0.322475,0.27175000000000005
10
  chunk_svs_no_conv,0.646675,0.6581250000000001,0.328175,0.269225
 
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
  chunk_0,0.7500249999999999,-2.21615,0.10572500000000001,0.0545
3
+ chunk_1,0.5,0.907225,0.542325,0.5
4
  chunk_2,0.73175,0.9472249999999999,0.31827500000000003,0.2564
5
+ chunk_3,0.25,0.46795,0.773175,0.75
6
+ chunk_4,0.200425,0.184375,0.77845,0.769025
7
+ chunk_chess_0,0.654775,-0.03560000000000002,0.31667500000000004,0.270275
8
+ chunk_grammar_1,0.5,0.300325,0.5386000000000001,0.5
9
  chunk_no_conv,0.600475,0.04259999999999997,0.322475,0.27175000000000005
10
  chunk_svs_no_conv,0.646675,0.6581250000000001,0.328175,0.269225
static/models_data/Qwen2.5-0.5B-Instruct/matrix.svg CHANGED
static/models_data/Qwen2.5-0.5B-Instruct/ranks.svg CHANGED
static/models_data/Qwen2.5-0.5B-Instruct/structure.svg CHANGED
static/models_data/Qwen2.5-32B-Instruct/cfa_metrics.csv CHANGED
@@ -1,10 +1,10 @@
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
- chunk_0,0.407525,0.379375,0.54825,0.5520999999999999
3
  chunk_1,0.662925,0.6309750000000001,0.324725,0.33095
4
- chunk_2,0.8544,0.804925,0.09639999999999999,0.1127
5
- chunk_3,0.910725,0.8782500000000001,0.08855,0.0821
6
- chunk_4,0.63385,0.5929,0.320975,0.34572499999999995
7
  chunk_chess_0,0.8669749999999999,0.8261499999999999,0.09075,0.1154
8
- chunk_grammar_1,0.8252,0.7672,0.090675,0.1505
9
  chunk_no_conv,0.8589749999999999,0.81545,0.110475,0.13344999999999999
10
  chunk_svs_no_conv,0.805125,0.782975,0.106525,0.128825
 
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
+ chunk_0,0.169525,0.14875,0.781075,0.778025
3
  chunk_1,0.662925,0.6309750000000001,0.324725,0.33095
4
+ chunk_2,0.39495,0.361725,0.5589999999999999,0.562725
5
+ chunk_3,0.689025,0.67395,0.319175,0.2949
6
+ chunk_4,0.225225,0.209975,0.770725,0.78855
7
  chunk_chess_0,0.8669749999999999,0.8261499999999999,0.09075,0.1154
8
+ chunk_grammar_1,0.37017500000000003,0.32975,0.558925,0.571725
9
  chunk_no_conv,0.8589749999999999,0.81545,0.110475,0.13344999999999999
10
  chunk_svs_no_conv,0.805125,0.782975,0.106525,0.128825
static/models_data/Qwen2.5-32B-Instruct/matrix.svg CHANGED
static/models_data/Qwen2.5-32B-Instruct/ranks.svg CHANGED
static/models_data/Qwen2.5-32B-Instruct/structure.svg CHANGED
static/models_data/Qwen2.5-72B-Instruct/cfa_metrics.csv CHANGED
@@ -1,10 +1,10 @@
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
- chunk_0,0.848975,0.804025,0.102925,0.09962499999999999
3
- chunk_1,0.87305,0.83225,0.10135,0.122
4
  chunk_2,0.82955,0.7724,0.106,0.12817499999999998
5
  chunk_3,0.86325,0.8176,0.11270000000000001,0.13125
6
- chunk_4,0.615525,0.568375,0.32222500000000004,0.388625
7
  chunk_chess_0,0.6388,0.59945,0.32495,0.3749
8
- chunk_grammar_1,0.66355,0.6322749999999999,0.33272500000000005,0.33187500000000003
9
- chunk_no_conv,0.889775,0.8521249999999999,0.09684999999999999,0.14735
10
  chunk_svs_no_conv,0.85115,0.8130499999999999,0.10115,0.082975
 
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
+ chunk_0,0.62435,0.589725,0.322575,0.32787499999999997
3
+ chunk_1,0.681925,0.656325,0.318375,0.336975
4
  chunk_2,0.82955,0.7724,0.106,0.12817499999999998
5
  chunk_3,0.86325,0.8176,0.11270000000000001,0.13125
6
+ chunk_4,0.42317499999999997,0.3947,0.54825,0.5777749999999999
7
  chunk_chess_0,0.6388,0.59945,0.32495,0.3749
8
+ chunk_grammar_1,0.452025,0.430675,0.5456000000000001,0.5548500000000001
9
+ chunk_no_conv,0.6969,0.6739999999999999,0.31410000000000005,0.35125000000000006
10
  chunk_svs_no_conv,0.85115,0.8130499999999999,0.10115,0.082975
static/models_data/Qwen2.5-72B-Instruct/matrix.svg CHANGED
static/models_data/Qwen2.5-72B-Instruct/ranks.svg CHANGED
static/models_data/Qwen2.5-72B-Instruct/structure.svg CHANGED
static/models_data/Qwen2.5-7B-Instruct/cfa_metrics.csv CHANGED
@@ -1,10 +1,10 @@
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
- chunk_0,0.7448499999999999,0.6389,0.1069,0.11257499999999998
3
  chunk_1,0.83535,0.8164499999999999,0.10495,0.064675
4
  chunk_2,0.8623,-3.591025,0.09625,0.071425
5
- chunk_3,0.88345,0.886125,0.097575,0.05395
6
- chunk_4,0.9301250000000001,0.9451499999999999,0.089075,0.049049999999999996
7
- chunk_chess_0,0.8341,0.764,0.112125,0.122975
8
- chunk_grammar_1,0.865475,0.877975,0.0904,0.0895
9
  chunk_no_conv,0.780725,0.695675,0.10905000000000001,0.15252499999999997
10
- chunk_svs_no_conv,0.743725,0.6745749999999999,0.120875,0.1343
 
1
  Context chunk,CFI,TLI,SRMR,RMSEA
2
+ chunk_0,0.380775,0.3469,0.5541,0.558875
3
  chunk_1,0.83535,0.8164499999999999,0.10495,0.064675
4
  chunk_2,0.8623,-3.591025,0.09625,0.071425
5
+ chunk_3,0.708525,0.73055,0.318925,0.2823
6
+ chunk_4,0.6801250000000001,0.651375,0.321175,0.29905000000000004
7
+ chunk_chess_0,0.38485,0.332975,0.566875,0.5715749999999999
8
+ chunk_grammar_1,0.6538999999999999,0.678825,0.32212500000000005,0.298825
9
  chunk_no_conv,0.780725,0.695675,0.10905000000000001,0.15252499999999997
10
+ chunk_svs_no_conv,0.539775,0.48119999999999996,0.34257499999999996,0.369425