InoteveraGaidhi rinotevera
Activation Steering uye Representation Engineering
Tekinoroji
Nhungamiro yehunyanzvi
SmoothQuant inzira inoita kuti zvikwanise kudzvanya mamodheru emitauro mikuru kusvika pa8-bit integers kune ese maremu uye ma activation pasina kudzidziswazve.
Teerera kugwaro iriAI-yakarondedzerwa · ~2 min
Izvo zvine basa nekuti ma activation mumamodhi mahombe ane akawandisa kunze ayo anowanzo kukanganisa yakaderera-chaiyo math, uye SmoothQuant inovabata.
Paunoderedza modhi kubva pa16-bit inoyangarara kuenda ku8-bit integers, huremu hunodzvanya nyore asi activations inonetsa: mamwe machanera anotakura zvakakosha 10 kusvika 100 akakura kupfuura mamwe, uye kuamanikidza kupinda coarse integer grid kunoparadza kurongeka. SmoothQuant, yakaunzwa naXiao et al. muna 2022, inoona kuti huremu hwakatsetseka uye huri nyore kuverengera uku activation iri spiky. Saka inotamisa kuoma kwemasvomhu: inopatsanura migero yekugonesa nechiyero che-per-channel uye inowanza huremu hunoenderana nechiyero chimwe chete. Iwo maviri mashandiro akadzima, achisiya iyo modhi yekubuda isina kuchinjika, asi ikozvino ese matensor anogara muhushamwari. Mhedzisiro ndeyeW8A8 (8-bit uremu uye activation) inference ine padyo-zero kurasika kwekurasika uye ingangoita 2x kumhanya uye ndangariro kuchengetedza.
Kushandira 70B-parameter LLM paW8A8 pane mashoma maGPU nepakati nepakati ndangariro uye matrix-kuwanza mutengo.
Kugonesa INT8 inference paNVIDIA Hopper/Blackwell tensor cores iyo natively inomhanyisa 8-bit integer math.
Kuendesa machat modhi pane inodhura-inotemerwa makore endpoints uko kupeta kaviri kubuda kunocheka zvakananga bhiri-yechiratidzo.
Kudzvanya maencoder etransformer yekutaura kana kuturikira pamudziyo apo 8-bit kernels inomhanya nekukurumidza uye inotonhorera.
SmoothQuant yakasimbisa kuti activation outliers inotamisa pane isingadzivisike, uye iyo pfungwa ikozvino inosimbisa kugadzirwa INT8 uye FP8 kushanda. Tarisira kutsvedzerera kusanganisirwa nezvirongwa zvakanakisa senge-per-group quantization, kudzidza kuyera, uye 4-bit activation research (e.g. outlier-aware ways). Sezvo FP8 Hardware (Hopper, Blackwell) inokura, kutsetseka-maitiro ekuyeresa acharamba achibikwa mucompiler uye inference-injini mapaipi kuitira kuti quantization igare yakasununguka.
Tsamba yemahara
Nhatu dzakasimbiswa dzeAI nyaya mangwanani ega ega pakati pevhiki, dzakanyorwa neChirungu chakajeka. Yemahara zvachose, hapana ads.
Imwe email zuva rega rega revhiki. Bvisa kunyoresa kamwe chete. Hatimbotengese kana kugovera kero yako.
Zviedze iwe pachako
Mhinduro yepakarepo pamhinduro yega yega, uye chitupa chinogovaniswa chine ID inobatika kana wapedza kosi.
Tsigira dzidzo yemahara yeAI. AI Understanding ndeye 501(c)(3) isingabatsiri - hapana ads, hapana paywall, zvachose. Ita chipo
SmoothQuant inzira inoita kuti zvikwanise kudzvanya mamodheru emitauro mikuru kusvika pa8-bit integers kune ese maremu uye ma activation pasina kudzidziswazve. Izvo zvine basa nekuti ma activation mumamodhi mahombe ane akawandisa kunze ayo anowanzo kukanganisa yakaderera-chaiyo math, uye SmoothQuant inovabata.
SmoothQuant inonangana nehukuru hwekunze hunoonekwa mune dzimwe activation chiteshi, izvo zvikasadaro zvinoparadza chokwadi kana activation ichiverengerwa ku8 bits.
Iyo inopatsanura activation chiteshi nechikero che-per-channel uye inowanza huremu hunofananidzwa nechikero ichocho, saka chigadzirwa hachichinjiki asi ese matensor anova nyore kuyera.
W8A8 inoreva 8-bit quantization yezviremu zvese (W) uye activation (A), iyo hurumende SmoothQuant inogonesa nekurasikirwa kudiki kwechokwadi.
Zviyero zvekutsvedzerera zvinopetwa muhuremu hwekare kana kuti fused op pamberi penguva, saka hapana imwe computation inoitika pakufungidzira.
Alpha (kazhinji 0.5) inoyera iyo inotsvedzerera chinhu, ichifunga kuti yakawanda sei yekuomerwa kweiyo quantization inobviswa kubva kune activation uye pahuremu.
Ramba uchidzidza
Mamwe madhairekitori akasarudzirwa nyaya iyi
InoteveraGaidhi rinotevera
Activation Steering uye Representation Engineering
Tekinoroji