[ { "label": "HF BF16", "group": "reference", "ppl": 1.3917625456677853, "bpw_layer": 16.0, "bpw_head": 16.0, "bpw_embed": 16.0, "vram_gb": 51.8809814453125 }, { "label": "Noise floor", "group": "noise_floor", "ppl": 1.3915002431865764, "kld": 0.001020502531901002, "kld_median": 1.282222910958808e-05, "kld_p10": -2.0448380766424634e-08, "kld_p25": 8.19636625237763e-09, "kld_p75": 0.0007068864069879055, "kld_p90": 0.0023693672847002745, "kld_buckets": [ { "conf_range": [ 0.0, 0.25 ], "share": 0.010834397748112679, "mean": 0.003439393127337098, "median": 0.002221998292952776 }, { "conf_range": [ 0.25, 0.5 ], "share": 0.10518290847539902, "mean": 0.0033397581428289413, "median": 0.0018602809868752956 }, { "conf_range": [ 0.5, 0.75 ], "share": 0.14749160408973694, "mean": 0.0028259900864213705, "median": 0.0013211295008659363 }, { "conf_range": [ 0.75, 0.95 ], "share": 0.14585527777671814, "mean": 0.001274451962672174, "median": 0.0004966517444700003 }, { "conf_range": [ 0.95, 1.0 ], "share": 0.5906358361244202, "mean": 4.953622192260809e-05, "median": 6.33895353985281e-08 } ], "bpw_layer": 16.0, "bpw_head": 16.0, "bpw_embed": 16.0, "vram_gb": 51.8809814453125 }, { "label": "2.50bpw-H5", "group": "EXL3", "ppl": 1.5507602966695333, "kld": 0.09578467905521393, "kld_median": 0.005908159539103508, "kld_p10": 1.255859305437923e-09, "kld_p25": 4.354752036306309e-06, "kld_p75": 0.0780472382903099, "kld_p90": 0.2608593702316284, "kld_buckets": [ { "conf_range": [ 0.0, 0.25 ], "share": 0.010834397748112679, "mean": 0.27495020627975464, "median": 0.11884737759828568 }, { "conf_range": [ 0.25, 0.5 ], "share": 0.10518290847539902, "mean": 0.2607549726963043, "median": 0.13331009447574615 }, { "conf_range": [ 0.5, 0.75 ], "share": 0.14749160408973694, "mean": 0.2186451107263565, "median": 0.10543665289878845 }, { "conf_range": [ 0.75, 0.95 ], "share": 0.14585527777671814, "mean": 0.16442078351974487, "median": 0.06520291417837143 }, { "conf_range": [ 0.95, 1.0 ], "share": 0.5906358361244202, "mean": 0.015489807352423668, "median": 1.5676057955715805e-05 } ], "bpw_layer": 2.504164938549567, "bpw_head": 5.00248303396898, "bpw_embed": 16.0, "vram_gb": 10.624184969812632, "on_disk_gb": 14.20649278536439, "unexplained_gb": 0.0, "expected_absent_gb": 3.582307815551758 }, { "label": "3.00bpw-H5", "group": "EXL3", "ppl": 1.4650940553544656, "kld": 0.04725328087806702, "kld_median": 0.0018352975603193045, "kld_p10": -5.42622258237202e-09, "kld_p25": 1.3227120234660106e-06, "kld_p75": 0.035319484770298004, "kld_p90": 0.12741748988628387, "kld_buckets": [ { "conf_range": [ 0.0, 0.25 ], "share": 0.010834397748112679, "mean": 0.14481262862682343, "median": 0.06390627473592758 }, { "conf_range": [ 0.25, 0.5 ], "share": 0.10518290847539902, "mean": 0.13698941469192505, "median": 0.07016156613826752 }, { "conf_range": [ 0.5, 0.75 ], "share": 0.14749160408973694, "mean": 0.11536626517772675, "median": 0.05396163463592529 }, { "conf_range": [ 0.75, 0.95 ], "share": 0.14585527777671814, "mean": 0.07529062777757645, "median": 0.02850371226668358 }, { "conf_range": [ 0.95, 1.0 ], "share": 0.5906358361244202, "mean": 0.005550438072532415, "median": 3.792483312281547e-06 } ], "bpw_layer": 3.004164938549567, "bpw_head": 5.00248303396898, "bpw_embed": 16.0, "vram_gb": 12.088906649500132, "on_disk_gb": 15.67121446505189, "unexplained_gb": 0.0, "expected_absent_gb": 3.582307815551758 }, { "label": "3.50bpw-H5", "group": "EXL3", "ppl": 1.4293462616502994, "kld": 0.02358868531882763, "kld_median": 0.0007844893261790276, "kld_p10": -7.4154287155181464e-09, "kld_p25": 7.421469945256831e-07, "kld_p75": 0.016039883717894554, "kld_p90": 0.06077731400728226, "kld_buckets": [ { "conf_range": [ 0.0, 0.25 ], "share": 0.010834397748112679, "mean": 0.07517395913600922, "median": 0.0326455682516098 }, { "conf_range": [ 0.25, 0.5 ], "share": 0.10518290847539902, "mean": 0.071629099547863, "median": 0.033376604318618774 }, { "conf_range": [ 0.5, 0.75 ], "share": 0.14749160408973694, "mean": 0.0593377947807312, "median": 0.025183163583278656 }, { "conf_range": [ 0.75, 0.95 ], "share": 0.14585527777671814, "mean": 0.035397719591856, "median": 0.012942892499268055 }, { "conf_range": [ 0.95, 1.0 ], "share": 0.5906358361244202, "mean": 0.0022438387386500835, "median": 1.8237283256894443e-06 } ], "bpw_layer": 3.504164938549567, "bpw_head": 5.00248303396898, "bpw_embed": 16.0, "vram_gb": 13.553628329187632, "on_disk_gb": 17.13593614473939, "unexplained_gb": 0.0, "expected_absent_gb": 3.582307815551758 }, { "label": "4.00bpw", "group": "EXL3", "ppl": 1.4083834936765058, "kld": 0.012438196688890457, "kld_median": 0.0003671977901831269, "kld_p10": -9.606677231488447e-09, "kld_p25": 3.046016558982956e-07, "kld_p75": 0.008704855106770992, "kld_p90": 0.03359178081154823, "kld_buckets": [ { "conf_range": [ 0.0, 0.25 ], "share": 0.010834397748112679, "mean": 0.042030107229948044, "median": 0.019715536385774612 }, { "conf_range": [ 0.25, 0.5 ], "share": 0.10518290847539902, "mean": 0.03888596221804619, "median": 0.019420064985752106 }, { "conf_range": [ 0.5, 0.75 ], "share": 0.14749160408973694, "mean": 0.03261873126029968, "median": 0.014236776158213615 }, { "conf_range": [ 0.75, 0.95 ], "share": 0.14585527777671814, "mean": 0.017590878531336784, "median": 0.00685095926746726 }, { "conf_range": [ 0.95, 1.0 ], "share": 0.5906358361244202, "mean": 0.0008735933806747198, "median": 8.953586529969471e-07 } ], "bpw_layer": 4.004164938549566, "bpw_head": 6.00248303396898, "bpw_embed": 16.0, "vram_gb": 15.1749585904181, "on_disk_gb": 18.757266405969858, "unexplained_gb": 0.0, "expected_absent_gb": 3.582307815551758 }, { "label": "5.00bpw", "group": "EXL3", "ppl": 1.3957362507539626, "kld": 0.00400218041613698, "kld_median": 0.00011092517524957657, "kld_p10": -9.554199209560466e-09, "kld_p25": 7.155539094583219e-08, "kld_p75": 0.0028881635516881943, "kld_p90": 0.010395795106887817, "kld_buckets": [ { "conf_range": [ 0.0, 0.25 ], "share": 0.010834397748112679, "mean": 0.014059449546039104, "median": 0.00631443876773119 }, { "conf_range": [ 0.25, 0.5 ], "share": 0.10518290847539902, "mean": 0.013213973492383957, "median": 0.006431023124605417 }, { "conf_range": [ 0.5, 0.75 ], "share": 0.14749160408973694, "mean": 0.01045642513781786, "median": 0.004578147083520889 }, { "conf_range": [ 0.75, 0.95 ], "share": 0.14585527777671814, "mean": 0.005323383025825024, "median": 0.002178037539124489 }, { "conf_range": [ 0.95, 1.0 ], "share": 0.5906358361244202, "mean": 0.00023922021500766277, "median": 3.901281502294296e-07 } ], "bpw_layer": 5.004164938549566, "bpw_head": 6.00248303396898, "bpw_embed": 16.0, "vram_gb": 18.1044019497931, "on_disk_gb": 21.686709765344858, "unexplained_gb": 0.0, "expected_absent_gb": 3.582307815551758 }, { "label": "6.00bpw", "group": "EXL3", "ppl": 1.3924564125630763, "kld": 0.0015886417822912335, "kld_median": 4.18894924223423e-05, "kld_p10": -1.8797473799736508e-08, "kld_p25": 1.78641617054609e-08, "kld_p75": 0.001235467498190701, "kld_p90": 0.0041245222091674805, "kld_buckets": [ { "conf_range": [ 0.0, 0.25 ], "share": 0.010834397748112679, "mean": 0.005647821351885796, "median": 0.0029941126704216003 }, { "conf_range": [ 0.25, 0.5 ], "share": 0.10518290847539902, "mean": 0.005168212112039328, "median": 0.002726401202380657 }, { "conf_range": [ 0.5, 0.75 ], "share": 0.14749160408973694, "mean": 0.004279580432921648, "median": 0.0019542400259524584 }, { "conf_range": [ 0.75, 0.95 ], "share": 0.14585527777671814, "mean": 0.002081291051581502, "median": 0.0008621406741440296 }, { "conf_range": [ 0.95, 1.0 ], "share": 0.5906358361244202, "mean": 8.30867065815255e-05, "median": 1.3826333145061653e-07 } ], "bpw_layer": 6.004164938549566, "bpw_head": 6.00248303396898, "bpw_embed": 16.0, "vram_gb": 21.0338453091681, "on_disk_gb": 24.616153124719858, "unexplained_gb": 0.0, "expected_absent_gb": 3.582307815551758 }, { "label": "thr:RedHatAI/Muse-Glimmer-30B-FP8-block", "group": "threshold", "ppl": 1.3968803033144324, "kld": 0.005431892815977335, "kld_median": 0.00013799371663480997, "kld_p10": -9.245020748949173e-09, "kld_p25": 9.719529714402597e-08, "kld_p75": 0.004438056144863367, "kld_p90": 0.014940274879336357, "kld_buckets": [ { "conf_range": [ 0.0, 0.25 ], "share": 0.010834397748112679, "mean": 0.019723201170563698, "median": 0.012289721518754959 }, { "conf_range": [ 0.25, 0.5 ], "share": 0.10518290847539902, "mean": 0.01760798692703247, "median": 0.010178981348872185 }, { "conf_range": [ 0.5, 0.75 ], "share": 0.14749160408973694, "mean": 0.014376400969922543, "median": 0.007130198180675507 }, { "conf_range": [ 0.75, 0.95 ], "share": 0.14585527777671814, "mean": 0.007329629268497229, "median": 0.003160950494930148 }, { "conf_range": [ 0.95, 1.0 ], "share": 0.5906358361244202, "mean": 0.00029913728940300643, "median": 4.6390141505980864e-07 } ], "bpw_layer": 8.001953125, "bpw_head": 16.0, "bpw_embed": 16.0, "vram_gb": 28.45115613937378 } ]