Comments (7)
Hi @YoungSharp,
The compression algorithm loss is a generic entity in our API and it ensures that any compression algorithm (or a combination thereof) works correctly, including the cases when it requires adding a penalty term to your main loss function. Some algorithms do not have such a requirement, such as quantization and magnitude sparsity, and for them, this loss term is 0 by default.
I suppose you are using one of those algorithms to compress your model? In that case, this behavior is expected. Do you have any other problems with your model fine-tuning procedure?
from nncf.
Hi @vanyalzr
thanks for replay, I was using quantization algorithm to compress my model.
If compression_loss = 0 is expected. I have another problem. By using mmdetection onnx exporter to export my compressed model to onnx model. i find onnx model weight does not looks like been quantized.
quantization config:
input_size = 448 ENABLE_COMPRESSION = True nncf_config = dict(compression=[dict(algorithm='quantization', initializer=dict(range=dict(num_init_steps=10)))], log_dir=work_dir)
Base model is efficientnet-b3, the first conv layer weights is below.
[ [ [ [ 0.3603667616844177, -0.5374307632446289, 0.001952380989678204 ], [ 0.6092172265052795, -0.5033608675003052, -0.051245443522930145 ], [ 0.015762008726596832, -0.0003829538472928107, 0.06971737742424011 ] ], [ [ 1.3720226287841797, -1.4568500518798828, -0.012406611815094948 ], [ 1.2627625465393066, -1.3461413383483887, -0.05189552903175354 ], [ 0.15398229658603668, 0.03449089452624321, 0.08602666854858398 ] ], [ [ 0.15650884807109833, -0.2866743206977844, 0.022628290578722954 ], [ 0.3569646179676056, -0.3498978614807129, -0.016743917018175125 ], [ 0.00387936201877892, -0.011319339275360107, 0.04645385965704918 ] ] ], [ [ [ 0.02379395067691803, 0.06416888535022736, 0.03441089391708374 ], [ -1.3542351722717285, -0.014414608478546143, 0.002027014037594199 ], [ -0.06681133806705475, 0.042854949831962585, 0.017452307045459747 ] ], [ [ 0.11277637630701065, -0.04625503346323967, 0.03549974411725998 ], [ -0.6202108263969421, 0.038116853684186935, 0.01379059161990881 ], [ 0.04088037833571434, 0.029540065675973892, 0.032375186681747437 ] ], [ [ 0.025897245854139328, 0.08049408346414566, 0.025638259947299957 ], [ -0.14927999675273895, 0.1072687953710556, 0.012837743386626244 ], [ -0.008325441740453243, 0.055185429751873016, 0.021701721474528313 ] ] ], [ [ [ 0.8385065793991089, 1.1928296089172363, 0.28208914399147034 ], [ 1.0477042198181152, 1.5227644443511963, 0.28695428371429443 ], [ 0.3125009536743164, 0.3149097263813019, -0.02771533466875553 ] ], [ [ -0.5980737209320068, -0.9167705178260803, -0.3587746322154999 ], [ -0.761725664138794, -1.0621089935302734, -0.38351115584373474 ], [ -0.2125093787908554, -0.3627846837043762, -0.06995101273059845 ] ], [ [ -0.25617456436157227, -0.3751916289329529, -0.10125856101512909 ], [ -0.33208248019218445, -0.5034953355789185, -0.06561792641878128 ], [ -0.0245540589094162, -0.04001801460981369, -0.014760272577404976 ] ] ], [ [ [ 0.016360608860850334, -0.3531891703605652, 0.03414047136902809 ], [ 0.09571370482444763, -0.08611265569925308, 0.0549452118575573 ], [ 0.06656917184591293, 0.013940572738647461, 0.04504940286278725 ] ], [ [ -0.27257680892944336, -1.6714335680007935, -0.09305314719676971 ], [ -0.09260738641023636, -0.17229221761226654, -0.039645005017519 ], [ -0.03724232316017151, -0.03694054111838341, -0.06745678186416626 ] ], [ [ 0.034323159605264664, -0.22521057724952698, 0.009854068048298359 ], [ 0.0744839608669281, 0.015600468963384628, 0.013111146166920662 ], [ 0.0033704806119203568, 0.050526298582553864, 0.016064150258898735 ] ] ], [ [ [ 0.7831529974937439, 0.08436092734336853, -0.05339992418885231 ], [ -0.08739947527647018, -0.06245709955692291, -0.03166312724351883 ], [ -0.016141800209879875, -0.04678699001669884, -0.03231404349207878 ] ], [ [ 1.3972235918045044, -0.03711560368537903, 0.03354016691446304 ], [ -0.015348229557275772, 0.07899950444698334, -0.00017750493134371936 ], [ 0.06583419442176819, 0.016041243448853493, 0.05606088414788246 ] ], [ [ 0.23624137043952942, 0.008777287788689137, 0.019887495785951614 ], [ -0.05615142360329628, -0.05744911730289459, -0.0127479936927557 ], [ 0.008637518621981144, -0.015697035938501358, -0.017470166087150574 ] ] ], [ [ [ 0.2583921253681183, 0.22606436908245087, -0.012629884295165539 ], [ 0.132790207862854, 0.23452122509479523, -0.05039222165942192 ], [ 0.10378012806177139, -0.00000580014875595225, 0.18373656272888184 ] ], [ [ 0.31156283617019653, 0.3494468629360199, 0.031576622277498245 ], [ 0.23437529802322388, -0.02659810520708561, 0.010435579344630241 ], [ 0.038272518664598465, 0.06256204843521118, -0.045725829899311066 ] ], [ [ -0.6872433423995972, -0.6243017911911011, -0.11017423123121262 ], [ -0.5409526824951172, -0.29402488470077515, -0.025335142388939857 ], [ -0.13306793570518494, -0.05541832372546196, -0.08850867301225662 ] ] ], [ [ [ 0.04774850606918335, -0.008904614485800266, 0.10948716104030609 ], [ -0.048879027366638184, -0.23575983941555023, 0.10958265513181686 ], [ 0.05797131732106209, 0.08702291548252106, -0.012438972480595112 ] ], [ [ -0.0015515829436481, -0.09604127705097198, 0.08865989744663239 ], [ -0.12140224128961563, -0.36449360847473145, 0.121602363884449 ], [ 0.031265731900930405, 0.13444119691848755, 0.00010512386506889015 ] ], [ [ -0.5229217410087585, -0.5190733671188354, -0.24117131531238556 ], [ -0.6028639078140259, -0.7049451470375061, -0.22954948246479034 ], [ -0.20107924938201904, -0.20751909911632538, -0.029963400214910507 ] ] ], [ [ [ 0.8163717985153198, -0.7485648393630981, -0.1100936308503151 ], [ 0.35844388604164124, -0.3826475143432617, 0.0250982865691185 ], [ 0.050180524587631226, -0.04203492030501366, -0.006629049777984619 ] ], [ [ 1.70492684841156, -2.180729627609253, -0.0662025734782219 ], [ 1.4882888793945312, -1.1406952142715454, 0.04071129485964775 ], [ 0.10040322691202164, -0.06366609781980515, 0.031120089814066887 ] ], [ [ 0.3348841071128845, -0.27060186862945557, -0.10768462717533112 ], [ 0.1670825481414795, -0.1599094718694687, -0.0003990217810496688 ], [ 0.040604714304208755, -0.04026969522237778, 0.002382713370025158 ] ] ], [ [ [ 0.4193877577781677, 0.5757598876953125, 0.2693241536617279 ], [ 0.48072150349617004, 0.5980150103569031, 0.27179908752441406 ], [ 0.29319897294044495, 0.2774105966091156, 0.08206933736801147 ] ], [ [ 0.030476650223135948, 0.1904967576265335, -0.12982256710529327 ], [ 0.20038177073001862, 0.2153308242559433, -0.10618191212415695 ], [ -0.12990781664848328, -0.14786118268966675, -0.1167616918683052 ] ], [ [ -0.03919680416584015, 0.20057658851146698, -0.1334412395954132 ], [ 0.11582577973604202, 0.16829605400562286, -0.09570185840129852 ], [ -0.10197532922029495, -0.0907364934682846, 0.015070403926074505 ] ] ], [ [ [ -0.42778220772743225, -0.5743306875228882, -0.1154937893152237 ], [ -0.5544014573097229, -0.6663929224014282, -0.040096547454595566 ], [ -0.04417478293180466, -0.05133568122982979, 0.021676132455468178 ] ], [ [ 0.7522192001342773, 0.8633449673652649, 0.3881658613681793 ], [ 0.7469053864479065, 0.9316321015357971, 0.4451141357421875 ], [ 0.2874215841293335, 0.3667811155319214, 0.13646817207336426 ] ], [ [ -0.16900835931301117, -0.45734903216362, -0.25859203934669495 ], [ -0.36283189058303833, -0.5795014500617981, -0.29318103194236755 ], [ -0.019782477989792824, -0.212410107254982, -0.1644723117351532 ] ] ], [ [ [ 0.019307604059576988, -0.03504905849695206, -0.26730766892433167 ], [ -0.05913460999727249, 0.08116823434829712, 0.35277870297431946 ], [ 0.04194584861397743, 0.05111641809344292, -0.03491010144352913 ] ], [ [ 0.06728450208902359, -0.045907534658908844, -1.0424983501434326 ], [ -0.04568254202604294, 0.04519440606236458, 0.9118595719337463 ], [ 0.10482602566480637, 0.006077687256038189, -0.12225450575351715 ] ], [ [ -0.01152153592556715, -0.011266634799540043, -0.17383696138858795 ], [ -0.0345972441136837, 0.041663769632577896, 0.2710641622543335 ], [ 0.006854290142655373, 0.005612744949758053, -0.003332367865368724 ] ] ], [ [ [ 0.009159397333860397, -0.023808464407920837, 0.014016272500157356 ], [ -0.05235389247536659, -0.6219943165779114, -0.037211500108242035 ], [ -0.014956335537135601, -0.03987602889537811, 0.021333735436201096 ] ], [ [ -0.028101911768317223, 0.038176778703927994, -0.01768294908106327 ], [ 0.019760092720389366, -0.17838062345981598, 0.025721170008182526 ], [ -0.010215799324214458, 0.016402726992964745, 0.0010135842021554708 ] ], [ [ -0.0003243161190766841, 0.01808064430952072, 0.0008097233949229121 ], [ 0.012368757277727127, -0.2519064247608185, 0.001306490390561521 ], [ -0.028058134019374847, -0.00843738205730915, -0.033520009368658066 ] ] ], [ [ [ -0.398679256439209, -0.21350063383579254, 0.32981982827186584 ], [ -0.3684980869293213, -0.086346335709095, 0.19863522052764893 ], [ -0.013920797035098076, 0.1418110430240631, 0.19478055834770203 ] ], [ [ -0.2506774663925171, -0.0494108684360981, 0.18298570811748505 ], [ -0.325312077999115, -0.16771648824214935, 0.22042930126190186 ], [ 0.03484612703323364, 0.13724569976329803, 0.11828869581222534 ] ], [ [ -0.07061946392059326, -0.13229741156101227, 0.15138204395771027 ], [ -0.12568449974060059, -0.06493557244539261, 0.09958207607269287 ], [ -0.0021261044312268496, 0.07625563442707062, 0.06583329290151596 ] ] ], [ [ [ -0.1981441229581833, -0.10210604220628738, -0.004398610442876816 ], [ 0.6559460759162903, 0.29809072613716125, 0.2626305818557739 ], [ -0.0855003148317337, -0.696674108505249, -0.16444475948810577 ] ], [ [ -0.2445569485425949, -0.23681049048900604, -0.06361504644155502 ], [ 0.987004280090332, 0.4560335874557495, 0.39156538248062134 ], [ -0.5248368382453918, -0.5445421934127808, -0.29503196477890015 ] ], [ [ -0.08954138308763504, -0.08140949159860611, -0.01067945547401905 ], [ 0.4982849657535553, 0.2158895581960678, 0.18374638259410858 ], [ 0.060414861887693405, -0.5450943112373352, -0.14695507287979126 ] ] ], [ [ [ -0.8562104105949402, -0.9906357526779175, -0.5031142234802246 ], [ -0.9071540832519531, -0.9781914353370667, -0.4355827867984772 ], [ -0.5464594960212708, -0.500324010848999, -0.22130411863327026 ] ], [ [ 0.14020834863185883, 0.0802011489868164, 0.046848174184560776 ], [ 0.10126226395368576, 0.08568456023931503, 0.011780144646763802 ], [ 0.04818765074014664, 0.014645327813923359, -0.02486075460910797 ] ], [ [ 0.8083459138870239, 0.8701810836791992, 0.4228171706199646 ], [ 0.8594750761985779, 0.9532227516174316, 0.436838299036026 ], [ 0.45302775502204895, 0.4736107885837555, 0.20611847937107086 ] ] ], [ [ [ 0.058796171098947525, 0.08335628360509872, -0.0711517184972763 ], [ 0.0573323555290699, 0.21111132204532623, -0.03765500336885452 ], [ -0.033855997025966644, -0.09221343696117401, -0.05786622688174248 ] ], [ [ 0.6328456401824951, 0.8780357241630554, 0.32857784628868103 ], [ 0.7684171795845032, 1.0956979990005493, 0.4332287013530731 ], [ 0.42248716950416565, 0.45601850748062134, 0.17461632192134857 ] ], [ [ -0.7750219106674194, -0.8816672563552856, -0.32532528042793274 ], [ -0.9552940130233765, -1.00095534324646, -0.3210221827030182 ], [ -0.4230894446372986, -0.41579219698905945, -0.1073332279920578 ] ] ], [ [ [ -0.46355435252189636, -0.2625730037689209, 0.03156619891524315 ], [ -0.05102786049246788, -0.146687850356102, 0.07387083768844604 ], [ 0.20603887736797333, 0.0518198199570179, 0.004210569895803928 ] ], [ [ -0.625877857208252, -0.7585473656654358, -0.07889989018440247 ], [ -0.337210088968277, -0.6023625135421753, 0.019004467874765396 ], [ 0.20417436957359314, 0.01892290636897087, 0.1541927605867386 ] ], [ [ -0.2109488993883133, -0.1699456125497818, 0.09547708183526993 ], [ 0.014258615672588348, -0.0939343050122261, 0.15772822499275208 ], [ 0.2345331460237503, 0.03321376070380211, 0.04866067320108414 ] ] ], [ [ [ -0.08066356182098389, 0.06699484586715698, -0.05335163325071335 ], [ 0.137163445353508, 0.07116510719060898, -0.06270071119070053 ], [ -0.06939078122377396, -0.04035351425409317, -0.028031544759869576 ] ], [ [ -0.06085171177983284, 0.07180242985486984, -0.07156968861818314 ], [ 0.05532224103808403, 0.05118855461478233, -0.03401333466172218 ], [ -0.03794176131486893, -0.05365779623389244, 0.0020153732039034367 ] ], [ [ 0.5407347679138184, 0.5712735652923584, 0.126934215426445 ], [ 0.7878959774971008, 0.7312890291213989, 0.16087837517261505 ], [ 0.13191452622413635, 0.12024389207363129, -0.02063019573688507 ] ] ], [ [ [ 0.1828603297472, 0.3707127571105957, 0.09805241227149963 ], [ 0.19478116929531097, 0.481236070394516, 0.062155742198228836 ], [ 0.0438610278069973, 0.042197905480861664, 0.033447228372097015 ] ], [ [ -0.7129276990890503, -1.2270729541778564, -0.32946935296058655 ], [ -0.9773596525192261, -1.42634916305542, -0.285745769739151 ], [ -0.17941023409366608, -0.2296290248632431, -0.16497860848903656 ] ], [ [ 0.7074711918830872, 1.0014268159866333, 0.21635089814662933 ], [ 0.7776404023170471, 1.0897234678268433, 0.24309903383255005 ], [ 0.15569670498371124, 0.17921270430088043, 0.05190783739089966 ] ] ], [ [ [ 0.6448550820350647, 0.7757762670516968, 0.265536904335022 ], [ 0.855246365070343, 1.1213781833648682, 0.3977590799331665 ], [ 0.42601892352104187, 0.4510995149612427, 0.13173000514507294 ] ], [ [ -0.7698758840560913, -0.9989760518074036, -0.36500924825668335 ], [ -0.9064593315124512, -1.2775191068649292, -0.3780517578125 ], [ -0.32685261964797974, -0.3724842369556427, -0.142035111784935 ] ], [ [ 0.10142450779676437, 0.049159713089466095, 0.062290340662002563 ], [ -0.06228981167078018, -0.08224049210548401, -0.010450776666402817 ], [ 0.042704418301582336, 0.020593175664544106, 0.06695406883955002 ] ] ], [ [ [ -0.2333219051361084, -0.05861935392022133, 0.017658160999417305 ], [ 0.2726060152053833, 0.002925069769844413, -0.012226058170199394 ], [ 0.08304442465305328, -0.04444706439971924, -0.012449493631720543 ] ], [ [ -2.6253952980041504, -0.07974566519260406, -0.07114512473344803 ], [ 2.6212098598480225, 0.03813294321298599, 0.08351016044616699 ], [ 0.09143855422735214, -0.03474245220422745, -0.0011877852957695723 ] ], [ [ -0.12611427903175354, -0.0277447160333395, 0.0174252912402153 ], [ 0.13798558712005615, 0.02108697034418583, -0.025645820423960686 ], [ 0.06430352479219437, -0.03507714346051216, 0.0019475150620564818 ] ] ], [ [ [ -0.09095511585474014, -0.21872882544994354, -0.06964129209518433 ], [ -0.1523389220237732, -0.4586048424243927, -0.1487767994403839 ], [ -0.03353796526789665, -0.06752968579530716, -0.06365496665239334 ] ], [ [ 0.49509748816490173, 0.9233545660972595, 0.21358178555965424 ], [ 0.7725897431373596, 1.5698148012161255, 0.38938605785369873 ], [ 0.1232072189450264, 0.24568314850330353, -0.020314032211899757 ] ], [ [ -0.4995216727256775, -0.7280101776123047, -0.1727304756641388 ], [ -0.6577193140983582, -1.1066330671310425, -0.2805374562740326 ], [ -0.1264062374830246, -0.2190459668636322, 0.055434923619031906 ] ] ], [ [ [ -0.04300277680158615, -0.0655662938952446, -0.029654234647750854 ], [ -0.06257188320159912, -1.0924845933914185, -0.060749709606170654 ], [ -0.06638696044683456, -0.11511973291635513, -0.02709292061626911 ] ], [ [ 0.007220647297799587, -0.016386747360229492, 0.0067110443487763405 ], [ 0.02901639975607395, -0.7643899917602539, -0.05630763620138168 ], [ -0.013474308885633945, -0.10103007405996323, 0.010158869437873363 ] ], [ [ 0.03740652650594711, -0.0008063737186603248, 0.036541346460580826 ], [ -0.02309894561767578, -0.3653051555156708, -0.0035656869877129793 ], [ 0.02682569809257984, -0.02334381826221943, 0.032951291650533676 ] ] ], [ [ [ -0.6283855438232422, -1.0346978902816772, -0.4618341624736786 ], [ -0.990109384059906, -1.3493326902389526, -0.5677844882011414 ], [ -0.440218061208725, -0.5083809494972229, -0.22494329512119293 ] ], [ [ 0.5865362882614136, 0.9011220932006836, 0.4359624683856964 ], [ 0.8290916681289673, 1.1895203590393066, 0.5677972435951233 ], [ 0.44348475337028503, 0.4978010356426239, 0.2861088812351227 ] ], [ [ 0.22518539428710938, 0.24377207458019257, 0.18839772045612335 ], [ 0.23825781047344208, 0.24001827836036682, 0.10789249837398529 ], [ 0.12714509665966034, 0.12732839584350586, 0.04583052918314934 ] ] ], [ [ [ 0.14288341999053955, 0.1661665439605713, 0.017716268077492714 ], [ 0.21468505263328552, 0.4363313615322113, 0.15240894258022308 ], [ 0.02158236876130104, 0.1273614466190338, -0.07062479853630066 ] ], [ [ 0.22774554789066315, 0.12395881116390228, 0.1334252804517746 ], [ 0.2157345414161682, 0.07453242689371109, 0.15068846940994263 ], [ 0.10964358597993851, 0.17018309235572815, 0.1537860631942749 ] ], [ [ -0.31842097640037537, -0.48083704710006714, -0.07941428571939468 ], [ -0.49105679988861084, -0.7597746849060059, -0.35145196318626404 ], [ -0.19168363511562347, -0.2859923243522644, -0.10497064143419266 ] ] ], [ [ [ -0.23247992992401123, -0.16225722432136536, -0.0362224318087101 ], [ -0.24673308432102203, -0.17584271728992462, -0.007466368842869997 ], [ -0.04854286462068558, -0.11079102754592896, 0.0030275911558419466 ] ], [ [ -0.664275050163269, -0.41647112369537354, -0.11373399198055267 ], [ -0.7338221669197083, -0.480450838804245, 0.020727945491671562 ], [ -0.18460418283939362, -0.11569029837846756, 0.022433390840888023 ] ], [ [ -0.09195003658533096, -0.09647324681282043, -0.04394334927201271 ], [ -0.0414799265563488, -0.08566100895404816, 0.010918430052697659 ], [ -0.04652433097362518, -0.039986759424209595, 0.02866440825164318 ] ] ], [ [ [ -0.08907525986433029, -0.5057278871536255, -0.08514060825109482 ], [ 0.007414326071739197, 0.5591298937797546, 0.06905894726514816 ], [ -0.025935346260666847, 0.027297768741846085, -0.025238165631890297 ] ], [ [ 0.20276440680027008, -1.336762547492981, -0.12318651378154755 ], [ -0.1847866177558899, 1.3915544748306274, 0.09633810818195343 ], [ -0.015326730906963348, -0.10923270881175995, -0.07494840025901794 ] ], [ [ -0.06998037546873093, -0.2446768879890442, -0.08565784990787506 ], [ -0.029606150463223457, 0.32551082968711853, 0.06208110973238945 ], [ -0.032088473439216614, 0.017849473282694817, -0.02085847593843937 ] ] ], [ [ [ 0.09651488810777664, 0.23205925524234772, -0.08183037489652634 ], [ -0.029768701642751694, 0.14294277131557465, -0.05942956730723381 ], [ -0.0415957011282444, 0.112301766872406, 0.12354004383087158 ] ], [ [ 0.5116007924079895, 0.8406188488006592, 0.019703593105077744 ], [ 0.08330079168081284, 0.6935448050498962, -0.007170010358095169 ], [ 0.060804642736911774, -0.03487224876880646, 0.07630860805511475 ] ], [ [ 0.1472375988960266, 0.10639087855815887, -0.014882327057421207 ], [ 0.09490340203046799, 0.17497366666793823, -0.07323182374238968 ], [ -0.028872594237327576, 0.07445451617240906, 0.11657644063234329 ] ] ], [ [ [ -0.7229280471801758, -1.2723137140274048, -0.2044282853603363 ], [ 0.3907982110977173, 0.36343222856521606, 0.08707531541585922 ], [ 0.4489656388759613, 0.8129950761795044, 0.17313021421432495 ] ], [ [ -1.8849279880523682, -2.66172194480896, -0.8309814929962158 ], [ 0.359889954328537, 0.611809253692627, 0.1255803108215332 ], [ 1.5214461088180542, 2.21639084815979, 0.6208829879760742 ] ], [ [ -0.24490311741828918, -0.3867005705833435, -0.12782146036624908 ], [ 0.15008841454982758, 0.1402387022972107, 0.0680820569396019 ], [ 0.18753619492053986, 0.1987757831811905, 0.10601679980754852 ] ] ], [ [ [ 0.12476927787065506, 0.09450821578502655, -0.0021741928067058325 ], [ 0.23859353363513947, 0.4203175902366638, 0.05808408558368683 ], [ 0.027211686596274376, -0.058450207114219666, -0.11174662411212921 ] ], [ [ 0.5247494578361511, 0.43590083718299866, 0.007708608638495207 ], [ 0.48485320806503296, 0.7540436387062073, 0.13374803960323334 ], [ -0.0437060222029686, -0.13196393847465515, -0.1502121388912201 ] ], [ [ 0.06071457639336586, 0.02188105881214142, 0.02326376922428608 ], [ 0.1268029510974884, 0.12796497344970703, 0.021176433190703392 ], [ -0.011505565606057644, -0.044241104274988174, -0.07049383223056793 ] ] ], [ [ [ 0.046929191797971725, 0.005567304790019989, 0.11009315401315689 ], [ 1.1094356775283813, 0.06104682385921478, 0.001989752519875765 ], [ 0.13448961079120636, -0.029986880719661713, 0.06342284381389618 ] ], [ [ -0.02683747559785843, 0.07784740626811981, -0.0380820594727993 ], [ -0.05361492559313774, -0.009224426001310349, -0.025265684351325035 ], [ -0.04171985760331154, 0.016504906117916107, -0.010941273532807827 ] ], [ [ -0.012030511163175106, -0.026031576097011566, -0.01272142119705677 ], [ -0.05199361592531204, -0.003316761925816536, 0.0119745759293437 ], [ -0.04280630871653557, 0.017966631799936295, -0.025569753721356392 ] ] ], [ [ [ 0.05673861503601074, -0.3442171812057495, 0.29814037680625916 ], [ 0.07890255004167557, -0.3074108362197876, 0.2072567343711853 ], [ 0.03108009323477745, -0.06109263002872467, 0.007638668175786734 ] ], [ [ -0.05523533746600151, -2.350207805633545, 2.323551654815674 ], [ -0.04087957367300987, -2.1407501697540283, 2.204890251159668 ], [ 0.04314588010311127, -0.03548639640212059, -0.01842305436730385 ] ], [ [ 0.020133597776293755, -0.18985797464847565, 0.14316506683826447 ], [ 0.05472446233034134, -0.17922285199165344, 0.11863597482442856 ], [ 0.02496812306344509, -0.054119326174259186, 0.015111591666936874 ] ] ], [ [ [ -0.36826759576797485, 0.05346129834651947, -0.013111489824950695 ], [ 0.24502187967300415, 0.10179971158504486, 0.018803991377353668 ], [ 0.052893638610839844, 0.05131584778428078, -0.03583928197622299 ] ], [ [ -0.2062772661447525, 0.2702265977859497, 0.041720759123563766 ], [ -0.10007712244987488, 0.2137908637523651, 0.04379701986908913 ], [ 0.022239530459046364, 0.08246369659900665, -0.04936861991882324 ] ], [ [ -0.25231802463531494, 0.011950662359595299, -0.009359505027532578 ], [ 0.31619271636009216, 0.031017551198601723, -0.05029178410768509 ], [ 0.06575247645378113, -0.036072805523872375, -0.1268322914838791 ] ] ], [ [ [ -0.18031398952007294, -0.3078305423259735, -0.016197869554162025 ], [ -0.1326381117105484, -0.314590722322464, 0.04581848531961441 ], [ 0.001434437115676701, -0.023663945496082306, 0.029290681704878807 ] ], [ [ -0.4272288978099823, -0.7436265349388123, -0.047925107181072235 ], [ -0.26583242416381836, -0.7907370924949646, 0.04620717465877533 ], [ 0.004439732991158962, -0.0781710296869278, 0.1213066577911377 ] ], [ [ -0.10199767351150513, -0.14107772707939148, 0.011785715818405151 ], [ -0.0456390306353569, -0.13385623693466187, 0.017242588102817535 ], [ 0.002430260181427002, -0.026987910270690918, 0.04776272177696228 ] ] ], [ [ [ -0.013949154876172543, 0.03946760296821594, -0.01679888367652893 ], [ 0.03412811830639839, 0.7154437899589539, 0.04671855643391609 ], [ -0.015230119228363037, 0.04908209666609764, -0.023905234411358833 ] ], [ [ -0.0017834395403042436, 0.05842486023902893, 0.029108626767992973 ], [ -0.00008539984264643863, 0.24045227468013763, 0.008705812506377697 ], [ 0.01104763988405466, 0.08698895573616028, 0.0353543646633625 ] ], [ [ 0.014890934340655804, 0.0023694531992077827, -0.016901109367609024 ], [ 0.011412251740694046, 0.36613452434539795, -0.015485808253288269 ], [ -0.040096547454595566, -0.06679188460111618, -0.025204453617334366 ] ] ], [ [ [ 0.21904590725898743, 0.568728506565094, 0.07251344621181488 ], [ -0.09646545350551605, -0.6750847697257996, -0.09228291362524033 ], [ 0.05512077733874321, -0.09400464594364166, 0.008183701895177364 ] ], [ [ 0.13221536576747894, 1.3762810230255127, 0.03968409076333046 ], [ -0.09568770974874496, -1.4706544876098633, -0.08553091436624527 ], [ 0.02074281871318817, -0.043814875185489655, 0.044483982026576996 ] ], [ [ 0.14102880656719208, 0.3832210600376129, 0.04342801868915558 ], [ -0.05153796821832657, -0.37412646412849426, -0.0797242671251297 ], [ 0.03593098372220993, -0.09172195196151733, 0.028739754110574722 ] ] ], [ [ [ -0.04175552725791931, 0.03395365551114082, 0.006025323644280434 ], [ -0.017435219138860703, 0.15353664755821228, 0.03521445766091347 ], [ -0.02121744304895401, 0.03729405254125595, -0.013773733749985695 ] ], [ [ 0.008787543512880802, 0.0701339989900589, 0.027702312916517258 ], [ 0.1079016700387001, 2.0528509616851807, 0.03867277130484581 ], [ 0.031473398208618164, 0.05735735967755318, 0.016289005056023598 ] ], [ [ -0.03804020211100578, 0.03467007726430893, -0.024132180958986282 ], [ 0.0001444255030946806, 0.32950520515441895, 0.017702477052807808 ], [ -0.03270495682954788, 0.017887180671095848, -0.03365739807486534 ] ] ], [ [ [ -0.03998103737831116, 0.06323116272687912, 0.009569073095917702 ], [ -0.4411902129650116, -0.38171955943107605, -0.03586957976222038 ], [ 0.47440284490585327, 0.27364158630371094, 0.006098921410739422 ] ], [ [ -0.037365417927503586, -0.048531826585531235, 0.0796576589345932 ], [ -2.3132498264312744, -2.153818368911743, -0.06571041792631149 ], [ 2.241406202316284, 2.201124906539917, 0.03273104876279831 ] ], [ [ -0.0036466571036726236, 0.022369517013430595, 0.02207104302942753 ], [ -0.15473444759845734, -0.19669309258460999, -0.04969623684883118 ], [ 0.2113747000694275, 0.10266326367855072, -0.01015976071357727 ] ] ], [ [ [ -0.36172571778297424, -0.03300758823752403, -0.1474684327840805 ], [ -0.062240809202194214, 0.5624579191207886, 0.07320359349250793 ], [ -0.22507846355438232, 0.14275968074798584, 0.056338150054216385 ] ], [ [ -2.554978132247925, -0.20979070663452148, -0.09245805442333221 ], [ 0.08752262592315674, 2.512366533279419, 0.11505569517612457 ], [ -0.11767906695604324, 0.22719387710094452, 0.03759007900953293 ] ], [ [ -0.18318217992782593, 0.061619628220796585, -0.07983869314193726 ], [ -0.0391593836247921, 0.23017121851444244, 0.0532136894762516 ], [ -0.14887578785419464, 0.0837387889623642, 0.01188741810619831 ] ] ], [ [ [ -0.3290765583515167, -0.1621301919221878, 0.016724800691008568 ], [ -0.2662808895111084, -0.11034732311964035, 0.01069191936403513 ], [ 0.04809742420911789, 0.06221426650881767, 0.015402339398860931 ] ], [ [ -0.6337690949440002, -0.4286348223686218, 0.07476869970560074 ], [ -0.7615185976028442, -0.5187259316444397, 0.07840161770582199 ], [ -0.03377047926187515, 0.16467365622520447, 0.10605861246585846 ] ], [ [ -0.16727682948112488, -0.06927219033241272, 0.006499067414551973 ], [ -0.1113610491156578, -0.03952357545495033, -0.0021834103390574455 ], [ 0.04850565269589424, 0.04377703741192818, 0.04426499456167221 ] ] ] ]
from nncf.
The ONNX model you exported would contain weights in FP32 precision but also additional Quantizer nodes in the model graph - FakeQuantize nodes (or QuantizeLinear/DequantizeLinear node pairs, if you enabled that option). Weights in the original precision would be passed through those quantization nodes first and you would get the discretized weight tensor as a result. You should also have such quantizer nodes for the activation tensors. You can check the presence of these nodes with a model graph visualization tool like Netron.
OpenVINO runtime can interpret a model graph with these quantizer nodes and execute the actual convolution operations in INT8 precision efficiently.
from nncf.
After using Netron to visualize my onnx network I can't find any FakeQuantize nodes.
Can you tell me how to enable those option to get FakeQuantize nodes?
input_size = 448 ENABLE_COMPRESSION = True nncf_config = dict(compression=[dict(algorithm='quantization', initializer=dict(range=dict(num_init_steps=10)))], log_dir=work_dir)
is current used config.
from nncf.
Hi!
I also got a bit confused about the saved compressed model since I used different techniques, and for example for the filter pruning one the model size still the same after the fine-tuning process and the inference time it incresed a bit (~2 ms) but when I checked the wieght values it changed, do you have any idea why this happend and just to clarify what I did ( I load the pre-trained model {Encoder-Decoder) and I compress the encoder part and use the wrapped model for fine-tuning the whole model) ?
from nncf.
@YoungSharp FakeQuantize nodes should be inserted in the model graph automatically upon the execution of the wrap_nncf_model
function in the mmdetection patch we provided, since you specified the quantization algo in the config. You can notice the print(*get_all_modules(model).keys(), sep="\n")
statement right after wrap_nncf_model
is executed. You should see the names of all the modules of the model that are wrapped by nncf in your output. Also you should be able to see that info in the logs written to the work_dir
folder.
Did you modify the mmdetection patch from third_party_integration/mmdetection
beyond the config specification? Could you please provide, if possible, the patch you used for mmdetection, in case you modified it, and the config used so that we could reproduce your issue?
from nncf.
Hi @RanyaJumah!
Regarding your question on filter pruning - the .pth checkpoint file of the pruned model that you get after fine-tuning contains additional binary masks that are used to determine which filters to zero out, along with the model weights of their original shape. Hence the .pth file is similar in size to the original one. Now when you infer this model, the masks are being applied, but you still have these zero filters that are used in the convolution operations.
In order to get an actually compressed model with fewer FLOPs, we provide capabilities for ONNX export via the export_model
method of the compression controller (see #29 for details). The zero filters are discarded there upon the model export, so that the weight and activation tensors actually change their shapes.
If you wish to infer your compressed model in PyTorch directly, you can simply modify the export_model
method of the FilterPruningController
, so that the redundant filters and masks are discarded, but the PyTorch model is returned. If you will use this PyTorch export, we welcome you to open a PR and contribute this capability to the main codebase.
from nncf.
Related Issues (20)
- Compressed models that call torch.is_floating_point() during inference are traced with runtime error.
- nncf + ultralytics yolov8 training-time compression HOT 7
- Ultralytics yolov8 QAT example HOT 1
- [Good First Issue] [NNCF] Make NNCF common utils code pass mypy checks HOT 23
- [Good First Issue] [NNCF] Make NNCF common accuracy aware training code pass mypy checks HOT 17
- [Good First Issue] [NNCF] Make NNCF common tensor statistics code pass mypy checks HOT 9
- Thanks to our Contributors HOT 1
- [Good First Issue][NNCF]: Add INT8 weight compression conformance test for Tinyllama-1.1b PyTorch model HOT 19
- [Good First Issue][NNCF]: Fixing NNCFGraph export for visualization in Netron HOT 6
- Why doesn't the size and precision of the model change after INT4 quantization? HOT 2
- [Good First Issue][NNCF]: Optimize memory footprint by removing redundant collected statistics HOT 8
- [Good First Issue][NNCF]: Dump actual_subset_size to ov.Model HOT 8
- [Good First Issue][NNCF]: dump the ignored scope more gracefully HOT 4
- [Good First Issue][NNCF]: check number of u8, u4 constants in weight compression tests HOT 10
- PTQ of Fast R-CNN crashes in PyTorch backend HOT 1
- [Good First Issue][NNCF]: fix invalid error reporting in JSON schema HOT 19
- [Good First Issue][NNCF]: Add tests for torch device utils HOT 5
- [Good First Issue][NNCF]: Remove compress_to_fp16=False from examples HOT 3
- AttributeError: 'list' object has no attribute 'keys' when executing yolov8_quantize_with_accuracy_control example HOT 4
- The question about function create_compressed_model():RuntimeError: CUDA error: device-side assert triggered HOT 3
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from nncf.