@genai-fi/nanogpt 0.7.1 → 0.7.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/dist/Generator.js +7 -7
- package/dist/NanoGPTModel.js +8 -8
- package/dist/{RealDiv-CVYNbZxu.js → RealDiv-Dy0p8Bvo.js} +7 -7
- package/dist/{Reshape-CEsEp0AI.js → Reshape-DH5srBP0.js} +2 -2
- package/dist/{Reshape-Do18N3gO.js → Reshape-DvudQDvJ.js} +1 -1
- package/dist/TeachableLLM.js +5 -5
- package/dist/{TiedEmbedding-ccLBFiZi.js → TiedEmbedding-BxOerUmB.js} +4 -4
- package/dist/{axis_util-5DTW2tFV.js → axis_util-BzbKo31C.js} +1 -1
- package/dist/backend.js +2 -2
- package/dist/{backend_util-C9Ut8n0Q.js → backend_util-TE7aTPhZ.js} +4 -4
- package/dist/{broadcast_to-Ba9h_8DO.js → broadcast_to-CdbwV-Dj.js} +2 -2
- package/dist/{concat-CbXTetof.js → concat-CsxrgovM.js} +1 -1
- package/dist/{dataset-U3PrjwgU.js → dataset-CtdBYwjo.js} +3 -3
- package/dist/{dropout-DPfPgWWe.js → dropout-DYs5QFGQ.js} +1 -1
- package/dist/{gather-Bbh8DHhM.js → gather-CMMy2KEG.js} +1 -1
- package/dist/{gelu-BFwVnd1r.js → gelu-C-dPj6Ku.js} +1 -1
- package/dist/{gpgpu_math-DffelNS-.js → gpgpu_math-DGNLNL4I.js} +2 -2
- package/dist/{index-UdZhlibC.js → index-BoWRt-10.js} +4 -4
- package/dist/{index-DYD_yPa-.js → index-CLthM0TO.js} +10 -10
- package/dist/{kernel_funcs_utils-CXDy3EN7.js → kernel_funcs_utils-BYKWV8Aa.js} +3 -3
- package/dist/layers/BaseLayer.js +2 -2
- package/dist/layers/CausalSelfAttention.js +6 -6
- package/dist/layers/MLP.js +5 -5
- package/dist/layers/RMSNorm.js +3 -3
- package/dist/layers/RoPECache.js +4 -4
- package/dist/layers/TiedEmbedding.js +5 -5
- package/dist/layers/TransformerBlock.js +1 -1
- package/dist/loader/loadTransformers.js +1 -1
- package/dist/loader/oldZipLoad.js +5 -5
- package/dist/{log_sum_exp-BnmCkHWl.js → log_sum_exp-DbjkV734.js} +5 -5
- package/dist/main.js +5 -5
- package/dist/{mat_mul-dwmZz69e.js → mat_mul-8m8pfdcx.js} +1 -1
- package/dist/{max-ByjEGoFx.js → max-Ddnnb5xe.js} +1 -1
- package/dist/{mulmat_packed_gpu-IGPBp6h9.js → mulmat_packed_gpu-VSekgsNv.js} +1 -1
- package/dist/{ones-C8Mfln6-.js → ones-Dj0SDhHf.js} +2 -2
- package/dist/ops/adamAdjust.js +1 -1
- package/dist/ops/adamMoments.js +1 -1
- package/dist/ops/appendCache.js +3 -3
- package/dist/ops/attentionMask.js +1 -1
- package/dist/ops/cpu/adamAdjust.js +1 -1
- package/dist/ops/cpu/adamMoments.js +2 -2
- package/dist/ops/cpu/appendCache.js +2 -2
- package/dist/ops/cpu/attentionMask.js +5 -5
- package/dist/ops/cpu/fusedSoftmax.js +2 -2
- package/dist/ops/cpu/gatherSub.js +3 -3
- package/dist/ops/cpu/gelu.js +1 -1
- package/dist/ops/cpu/matMulGelu.js +2 -2
- package/dist/ops/cpu/matMulMul.js +1 -1
- package/dist/ops/cpu/mulDropout.js +1 -1
- package/dist/ops/cpu/normRMS.js +1 -1
- package/dist/ops/cpu/qkv.js +3 -3
- package/dist/ops/cpu/rope.js +5 -5
- package/dist/ops/cpu/scatterSub.js +5 -5
- package/dist/ops/fusedSoftmax.js +1 -1
- package/dist/ops/gatherSub.js +1 -1
- package/dist/ops/gelu.js +2 -2
- package/dist/ops/grads/attentionMask.js +1 -1
- package/dist/ops/grads/fusedSoftmax.js +2 -2
- package/dist/ops/grads/gelu.js +2 -2
- package/dist/ops/grads/matMulGelu.js +1 -1
- package/dist/ops/grads/normRMS.js +1 -1
- package/dist/ops/grads/qkv.js +1 -1
- package/dist/ops/grads/rope.js +1 -1
- package/dist/ops/matMulGelu.js +1 -1
- package/dist/ops/matMulMul.js +1 -1
- package/dist/ops/mulDrop.js +1 -1
- package/dist/ops/normRMS.js +1 -1
- package/dist/ops/qkv.js +1 -1
- package/dist/ops/rope.js +4 -4
- package/dist/ops/scatterSub.js +1 -1
- package/dist/ops/webgl/adamAdjust.js +2 -2
- package/dist/ops/webgl/adamMoments.js +7 -5
- package/dist/ops/webgl/appendCache.js +1 -1
- package/dist/ops/webgl/attentionMask.js +1 -1
- package/dist/ops/webgl/fusedSoftmax.js +4 -4
- package/dist/ops/webgl/gatherSub.js +1 -1
- package/dist/ops/webgl/gelu.js +2 -2
- package/dist/ops/webgl/log.js +3 -3
- package/dist/ops/webgl/matMulGelu.js +4 -4
- package/dist/ops/webgl/matMulMul.js +1 -1
- package/dist/ops/webgl/mulDropout.js +1 -1
- package/dist/ops/webgl/normRMS.js +2 -2
- package/dist/ops/webgl/qkv.js +1 -1
- package/dist/ops/webgl/rope.js +1 -1
- package/dist/ops/webgl/scatterSub.js +1 -1
- package/dist/ops/webgpu/adamAdjust.js +15 -13
- package/dist/ops/webgpu/adamMoments.js +18 -11
- package/dist/ops/webgpu/appendCache.js +18 -15
- package/dist/ops/webgpu/attentionMask.js +24 -18
- package/dist/ops/webgpu/gatherSub.js +17 -30
- package/dist/ops/webgpu/gelu.js +3 -3
- package/dist/ops/webgpu/normRMS.js +16 -8
- package/dist/ops/webgpu/normRMSGrad.js +25 -20
- package/dist/ops/webgpu/qkv.js +23 -19
- package/dist/ops/webgpu/rope.js +37 -24
- package/dist/ops/webgpu/scatterSub.js +16 -14
- package/dist/ops/webgpu/utils/reductions.js +4 -4
- package/dist/{ops-aRTXR2Sr.js → ops-BFGCx8Ri.js} +15 -15
- package/dist/{random_width-DbSpgl4o.js → random_width-sZORGo5k.js} +22 -22
- package/dist/{range-D9CZhVlR.js → range-CRuAh-gd.js} +1 -1
- package/dist/{reciprocal-CGB48wZB.js → reciprocal-BvGAyKyu.js} +1 -1
- package/dist/{register_all_kernels-DnbAyBXt.js → register_all_kernels-BwDSRN-f.js} +30 -30
- package/dist/{reshape-BR0eoLYN.js → reshape-CdBq1WJ6.js} +1 -1
- package/dist/{scatter_nd_util-OjyAxku2.js → scatter_nd_util-DUstGbU1.js} +1 -1
- package/dist/{selu_util-Ce6pu9IM.js → selu_util-BJEXVvjX.js} +3 -3
- package/dist/{shared-Czipaeb6.js → shared-B8ztnyEk.js} +6 -6
- package/dist/{shared-DS5waSIY.js → shared-wS99K7_n.js} +1 -1
- package/dist/{sin-CiBxrDqX.js → sin-BeA3tsEd.js} +1 -1
- package/dist/{slice-BHbDHObE.js → slice-BiOsknYS.js} +1 -1
- package/dist/{softmax-JMEIUo2J.js → softmax-Bv_6lyMX.js} +1 -1
- package/dist/{split-CRU0PjVV.js → split-B-dikLRw.js} +1 -1
- package/dist/{stack-ikk2Y8_P.js → stack-B17UN2nn.js} +1 -1
- package/dist/{sum-NLYbiDag.js → sum-66ew2byf.js} +1 -1
- package/dist/{tensor-Do9PKbIE.js → tensor-JwS7ZYY6.js} +1 -1
- package/dist/{tensor2d-CWHxHpLh.js → tensor2d-wxPAnDQy.js} +1 -1
- package/dist/training/Adam.js +2 -2
- package/dist/training/AdamExt.js +1 -1
- package/dist/training/DatasetBuilder.js +35 -32
- package/dist/training/FullTrainer.js +1 -1
- package/dist/training/Trainer.js +2 -2
- package/dist/training/sparseCrossEntropy.js +4 -4
- package/dist/utilities/dummy.js +2 -2
- package/dist/utilities/generate.js +3 -3
- package/dist/utilities/multinomialCPU.js +2 -2
- package/dist/utilities/performance.js +1 -1
- package/dist/utilities/profile.js +1 -1
- package/dist/utilities/safetensors.js +2 -2
- package/dist/utilities/weights.js +2 -2
- package/dist/{variable-BTBkayv_.js → variable-BuddVFLa.js} +1 -1
- package/dist/{webgpu_program-WaoMq-WD.js → webgpu_program-PFzf1hAQ.js} +1 -1
- package/dist/{webgpu_util-DhSeP4b6.js → webgpu_util-D____QpY.js} +1 -1
- package/dist/{zeros-DnPT2nD4.js → zeros--BdLQ3oG.js} +1 -1
- package/package.json +1 -1
|
@@ -1,35 +1,35 @@
|
|
|
1
|
-
import { B as oe, C as se, E as le, cl as ps, cm as hs, co as en, cp as fs, cq as ms, cs as gs, I as xs, cr as Cs, ct as bs, bp as To, l as Y, dd as tn, cz as sn, n as Do, z as nn, br as Eo, cD as Is, cE as ys, cF as $s, aq as Nc, cI as on, y as vc, a as an, k as Ne, dl as ks, cL as rn, cM as ws, X as It, bw as Fo, cQ as Ns, dg as cn, dj as ln, b as Po, bA as Oo, ds as un, cS as vs, dv as dn, dw as pn, al as xe, dx as S, aT as Sc, o as Rc, T as Ao, dy as Tc, cT as Dc, x as Ec, q as Fc, aU as Pc, c as Oc, ac as L, d2 as _o, d3 as Lo, e as Me, aO as Vt, d7 as Vo, aM as Wo, d4 as Ac, aL as Wt, aj as Ke, d8 as Mo, as as hn, an as _c, j as z, bg as Ge, ap as Lc, at as fn, au as mn, K as Vc,
|
|
2
|
-
import { e as Dr, s as Er, l as Fr, a as Pr, b as Or, m as Ar, c as _r, r as Gs, f as $l, g as kl, h as wl, i as Nl, j as vl, k as Sl, n as Rl, o as Tl, q as Dl, t as El, u as Fl, v as Pl, w as Ol, x as Al, y as _l, z as Ll, A as Vl, B as Wl, C as Ml, p as Bl, F as Gl, G as zl, H as Ul, I as Hl, J as Xl, K as Kl, L as ql } from "./random_width-
|
|
3
|
-
import { r as re } from "./reshape-
|
|
4
|
-
import { b as jl } from "./broadcast_to-
|
|
5
|
-
import { c as os } from "./concat-
|
|
6
|
-
import { c as Yl, s as Ql } from "./sin-
|
|
7
|
-
import { w as Lr, o as Vr, n as ss, a as zs, r as Us, p as Wr, f as Zl, g as Jl, d as eu, b as tu, e as su, k as nu, m as ou, q as au, v as ru, x as iu, s as cu, c as lu, t as uu, u as du } from "./ops-
|
|
8
|
-
import { m as Tt } from "./mat_mul-
|
|
9
|
-
import { w as Mr, k as Ye, h as Ve, x as pu, z as hu, A as fu, y as mu, D as gu, v as xu, i as ot, j as lt, d as De, e as $t, g as Rs, S as Br, a as Gr, B as zr, C as Ur, p as Cu, q as bu, r as Iu, u as yu } from "./selu_util-
|
|
10
|
-
import { e as $u, l as ku, a as wu } from "./log_sum_exp-
|
|
11
|
-
import { f as Nu } from "./dropout-
|
|
12
|
-
import { g as vu } from "./gather-
|
|
13
|
-
import { c as Hs, z as Hr } from "./zeros
|
|
14
|
-
import { s as as } from "./slice-
|
|
15
|
-
import { m as Su } from "./max-
|
|
16
|
-
import { r as Ru } from "./reciprocal-
|
|
17
|
-
import { s as Xs } from "./split-
|
|
18
|
-
import { s as Tu } from "./softmax-
|
|
19
|
-
import { s as Du } from "./stack-
|
|
20
|
-
import { s as Eu } from "./sum-
|
|
21
|
-
import { a as j, u as ue, c as ut, i as dt, b as Fu, d as Pt, t as Pe, e as kt, f as gt, g as Xr, r as Ot, h as Be, j as Pu, k as Ou, l as On, z as Au, m as An, n as Kr, o as _u, p as Lu, q as Vu, v as Wu, w as Mu, x as Bu, y as Gu, A as zu, B as Uu, C as Hu, D as ft, E as Xu, F as Ku, G as qr, H as qu, I as ju, J as Yu, K as Qu, L as Zu, M as Ju, N as ed, O as td, P as sd, Q as nd, R as od, S as ad, T as rd, U as id, V as cd, W as ld, X as ud, Y as dd, Z as pd, _ as hd, $ as fd, a0 as md, a1 as gd, a2 as xd, a3 as Cd, a4 as bd, a5 as Id, a6 as yd, a7 as $d, a8 as kd, a9 as wd, aa as Nd, ab as vd, ac as Sd, ad as Rd, ae as Td, af as Dd, ag as Ed, ah as Fd, ai as Pd } from "./shared-
|
|
22
|
-
import { m as xt, A as Ts, B as Ds, C as Es, D as jr, F as Yr, H as Qr, G as je, I as Zr, J as Jr, K as ei, L as ti, M as si, z as ni, y as oi, x as ai, w as ri, u as ii, E as ci, T as li, U as Od, V as Ad, W as _d, X as Ld, Y as Vd, Z as Wd, N as ui, O as di, P as pi, Q as hi, c as fi, f as Md, S as Bd, _ as Gd } from "./backend_util-
|
|
23
|
-
import { g as ye, a as $e, b as Re, c as Ee, e as Oe, h as Fs } from "./axis_util-
|
|
1
|
+
import { B as oe, C as se, E as le, cl as ps, cm as hs, co as en, cp as fs, cq as ms, cs as gs, I as xs, cr as Cs, ct as bs, bp as To, l as Y, dd as tn, cz as sn, n as Do, z as nn, br as Eo, cD as Is, cE as ys, cF as $s, aq as Nc, cI as on, y as vc, a as an, k as Ne, dl as ks, cL as rn, cM as ws, X as It, bw as Fo, cQ as Ns, dg as cn, dj as ln, b as Po, bA as Oo, ds as un, cS as vs, dv as dn, dw as pn, al as xe, dx as S, aT as Sc, o as Rc, T as Ao, dy as Tc, cT as Dc, x as Ec, q as Fc, aU as Pc, c as Oc, ac as L, d2 as _o, d3 as Lo, e as Me, aO as Vt, d7 as Vo, aM as Wo, d4 as Ac, aL as Wt, aj as Ke, d8 as Mo, as as hn, an as _c, j as z, bg as Ge, ap as Lc, at as fn, au as mn, K as Vc, a3 as Wc, J as Bo, a1 as ne, cd as Go, cn as zo, bE as Uo, p as me, a7 as st, bF as Ho, bG as Xo, bH as Ko, ab as ct, bI as qo, cu as jo, cv as Yo, bL as Qo, bK as Zo, aP as Jo, dc as ea, bM as gn, cx as ta, aY as sa, Q as na, bN as oa, by as Te, c9 as aa, bO as ra, bP as ia, cy as ca, bQ as la, a8 as xn, bR as Cn, b6 as ua, bS as da, u as nt, H as Mc, bT as pa, bU as ha, bV as fa, cb as ma, cc as ga, de as xa, bo as Ss, bi as Ft, bl as bn, cA as Kn, dz as mt, dA as Ca, cB as qn, df as Bc, Y as Gc, aR as ba, cC as Ia, bW as In, aV as ya, az as $a, ao as yt, d9 as ka, da as wa, b7 as Na, ca as va, di as Sa, a5 as Ra, G as Ta, dk as Da, bX as yn, b0 as $n, bZ as kn, cH as Ea, U as zc, b_ as Fa, b$ as Pa, cJ as Oa, cK as Aa, dm as _a, c0 as La, aS as Va, bh as Wa, S as Ma, ad as Ba, b9 as Ga, bb as za, ba as Ua, c1 as Ha, cU as Xa, c2 as Ka, P as qa, a2 as ja, c3 as Ya, cN as wn, dB as Qa, dC as Za, dD as Ja, R as er, V as Nn, bc as tr, cO as sr, bd as nr, cP as or, c5 as ar, b8 as rr, b3 as vn, a4 as ir, dn as cr, aQ as lr, c6 as Sn, a9 as Rn, c7 as Tn, bY as Dn, c4 as ur, dE as dr, dF as pr, dp as hr, dq as fr, dr as mr, L as gr, cR as xr, av as En, dt as Cr, dG as br, dH as Ir, bJ as Fn, du as yr, N as $r, be as kr, ax as Pn, b4 as wr, c8 as Nr, f as vr, dI as Sr, dJ as jn, d5 as Yn, d6 as Uc, t as Hc, dK as Xc, dL as Kc, bu as qc, b5 as jc, bf as Yc, bm as Qc, W as Zc, bn as Jc, bv as el, b2 as tl, bj as sl, bk as nl, aI as Rr, bq as ol, ae as al, ak as rl, bs as il, dh as cl, aW as ll, aX as ul, aZ as dl, a_ as pl, af as hl, bt as fl, b1 as ml, bC as gl, a$ as xl, ay as Tr, bx as Cl, ar as bl, bz as Il, bB as yl } from "./index-BoWRt-10.js";
|
|
2
|
+
import { e as Dr, s as Er, l as Fr, a as Pr, b as Or, m as Ar, c as _r, r as Gs, f as $l, g as kl, h as wl, i as Nl, j as vl, k as Sl, n as Rl, o as Tl, q as Dl, t as El, u as Fl, v as Pl, w as Ol, x as Al, y as _l, z as Ll, A as Vl, B as Wl, C as Ml, p as Bl, F as Gl, G as zl, H as Ul, I as Hl, J as Xl, K as Kl, L as ql } from "./random_width-sZORGo5k.js";
|
|
3
|
+
import { r as re } from "./reshape-CdBq1WJ6.js";
|
|
4
|
+
import { b as jl } from "./broadcast_to-CdbwV-Dj.js";
|
|
5
|
+
import { c as os } from "./concat-CsxrgovM.js";
|
|
6
|
+
import { c as Yl, s as Ql } from "./sin-BeA3tsEd.js";
|
|
7
|
+
import { w as Lr, o as Vr, n as ss, a as zs, r as Us, p as Wr, f as Zl, g as Jl, d as eu, b as tu, e as su, k as nu, m as ou, q as au, v as ru, x as iu, s as cu, c as lu, t as uu, u as du } from "./ops-BFGCx8Ri.js";
|
|
8
|
+
import { m as Tt } from "./mat_mul-8m8pfdcx.js";
|
|
9
|
+
import { w as Mr, k as Ye, h as Ve, x as pu, z as hu, A as fu, y as mu, D as gu, v as xu, i as ot, j as lt, d as De, e as $t, g as Rs, S as Br, a as Gr, B as zr, C as Ur, p as Cu, q as bu, r as Iu, u as yu } from "./selu_util-BJEXVvjX.js";
|
|
10
|
+
import { e as $u, l as ku, a as wu } from "./log_sum_exp-DbjkV734.js";
|
|
11
|
+
import { f as Nu } from "./dropout-DYs5QFGQ.js";
|
|
12
|
+
import { g as vu } from "./gather-CMMy2KEG.js";
|
|
13
|
+
import { c as Hs, z as Hr } from "./zeros--BdLQ3oG.js";
|
|
14
|
+
import { s as as } from "./slice-BiOsknYS.js";
|
|
15
|
+
import { m as Su } from "./max-Ddnnb5xe.js";
|
|
16
|
+
import { r as Ru } from "./reciprocal-BvGAyKyu.js";
|
|
17
|
+
import { s as Xs } from "./split-B-dikLRw.js";
|
|
18
|
+
import { s as Tu } from "./softmax-Bv_6lyMX.js";
|
|
19
|
+
import { s as Du } from "./stack-B17UN2nn.js";
|
|
20
|
+
import { s as Eu } from "./sum-66ew2byf.js";
|
|
21
|
+
import { a as j, u as ue, c as ut, i as dt, b as Fu, d as Pt, t as Pe, e as kt, f as gt, g as Xr, r as Ot, h as Be, j as Pu, k as Ou, l as On, z as Au, m as An, n as Kr, o as _u, p as Lu, q as Vu, v as Wu, w as Mu, x as Bu, y as Gu, A as zu, B as Uu, C as Hu, D as ft, E as Xu, F as Ku, G as qr, H as qu, I as ju, J as Yu, K as Qu, L as Zu, M as Ju, N as ed, O as td, P as sd, Q as nd, R as od, S as ad, T as rd, U as id, V as cd, W as ld, X as ud, Y as dd, Z as pd, _ as hd, $ as fd, a0 as md, a1 as gd, a2 as xd, a3 as Cd, a4 as bd, a5 as Id, a6 as yd, a7 as $d, a8 as kd, a9 as wd, aa as Nd, ab as vd, ac as Sd, ad as Rd, ae as Td, af as Dd, ag as Ed, ah as Fd, ai as Pd } from "./shared-B8ztnyEk.js";
|
|
22
|
+
import { m as xt, A as Ts, B as Ds, C as Es, D as jr, F as Yr, H as Qr, G as je, I as Zr, J as Jr, K as ei, L as ti, M as si, z as ni, y as oi, x as ai, w as ri, u as ii, E as ci, T as li, U as Od, V as Ad, W as _d, X as Ld, Y as Vd, Z as Wd, N as ui, O as di, P as pi, Q as hi, c as fi, f as Md, S as Bd, _ as Gd } from "./backend_util-TE7aTPhZ.js";
|
|
23
|
+
import { g as ye, a as $e, b as Re, c as Ee, e as Oe, h as Fs } from "./axis_util-BzbKo31C.js";
|
|
24
24
|
import { s as zd } from "./index-C4L8Cm77.js";
|
|
25
25
|
import { n as mi, b as gi, a as xi } from "./non_max_suppression_impl-CsEgBuMA.js";
|
|
26
|
-
import { c as wt } from "./scatter_nd_util-
|
|
27
|
-
import { h as Qn, j as Ud, k as Hd, l as Xd, m as Kd, n as qd, o as jd, P as _n, p as ze, u as We, q as Ci, c as bi, T as _e, E as Ii, g as yi, a as $i, r as Yd, s as Qd, t as Z, v as Mt, w as Zd, x as Zn, y as Jd, z as ep, A as Bt, B as tp, C as sp, D as Ks, F as jt, G as Yt, H as np, I as op, J as Jn, K as ap, L as rp, M as Vs, N as ip, O as cp, Q as lp, R as Qt, S as Ws, U as up, f as Ce, V as we, W as Zt, X as Jt, Y as dp, d as eo, e as to, i as ki, Z as pp, _ as hp, $ as fp, a0 as mp, a1 as gp, a2 as xp, a3 as Gt } from "./gpgpu_math-
|
|
28
|
-
import { s as wi, a as Cp, t as Ni, b as bp, c as Ip, d as vi, e as yp, n as $p, f as kp, g as wp, h as Np, i as vp, j as Sp, k as Rp, l as Tp, o as Dp, p as Ep, q as Fp, r as Pp, u as Op, v as Ap, w as _p, x as Lp, y as Vp, z as Wp, A as Mp, B as Bp, C as Gp, D as zp, E as Up, F as Hp, G as Xp, H as Kp, I as qp, J as jp, K as Yp, L as Qp, M as Si, N as Zp, O as Jp, P as eh, Q as th, R as sh, S as nh, T as oh, U as ah, V as rh, W as ih } from "./shared-
|
|
29
|
-
import { a as Se, c as ch, U as it, d as tt, e as qe, A as so, f as Nt, B as Ln, h as Vn, m as At, u as ae, C as Ue, b as ke, i as Le, j as Wn, k as pt, l as vt, n as lh, o as uh, p as dh, q as ph } from "./kernel_funcs_utils-
|
|
30
|
-
import { R as hh, r as H, a as fh } from "./Reshape-
|
|
31
|
-
import { M as Ri } from "./mulmat_packed_gpu-
|
|
32
|
-
import { t as Ti, s as Mn, a as zt, m as mh, r as gh, b as xh, c as Ch, d as bh } from "./RealDiv-
|
|
26
|
+
import { c as wt } from "./scatter_nd_util-DUstGbU1.js";
|
|
27
|
+
import { h as Qn, j as Ud, k as Hd, l as Xd, m as Kd, n as qd, o as jd, P as _n, p as ze, u as We, q as Ci, c as bi, T as _e, E as Ii, g as yi, a as $i, r as Yd, s as Qd, t as Z, v as Mt, w as Zd, x as Zn, y as Jd, z as ep, A as Bt, B as tp, C as sp, D as Ks, F as jt, G as Yt, H as np, I as op, J as Jn, K as ap, L as rp, M as Vs, N as ip, O as cp, Q as lp, R as Qt, S as Ws, U as up, f as Ce, V as we, W as Zt, X as Jt, Y as dp, d as eo, e as to, i as ki, Z as pp, _ as hp, $ as fp, a0 as mp, a1 as gp, a2 as xp, a3 as Gt } from "./gpgpu_math-DGNLNL4I.js";
|
|
28
|
+
import { s as wi, a as Cp, t as Ni, b as bp, c as Ip, d as vi, e as yp, n as $p, f as kp, g as wp, h as Np, i as vp, j as Sp, k as Rp, l as Tp, o as Dp, p as Ep, q as Fp, r as Pp, u as Op, v as Ap, w as _p, x as Lp, y as Vp, z as Wp, A as Mp, B as Bp, C as Gp, D as zp, E as Up, F as Hp, G as Xp, H as Kp, I as qp, J as jp, K as Yp, L as Qp, M as Si, N as Zp, O as Jp, P as eh, Q as th, R as sh, S as nh, T as oh, U as ah, V as rh, W as ih } from "./shared-wS99K7_n.js";
|
|
29
|
+
import { a as Se, c as ch, U as it, d as tt, e as qe, A as so, f as Nt, B as Ln, h as Vn, m as At, u as ae, C as Ue, b as ke, i as Le, j as Wn, k as pt, l as vt, n as lh, o as uh, p as dh, q as ph } from "./kernel_funcs_utils-BYKWV8Aa.js";
|
|
30
|
+
import { R as hh, r as H, a as fh } from "./Reshape-DH5srBP0.js";
|
|
31
|
+
import { M as Ri } from "./mulmat_packed_gpu-VSekgsNv.js";
|
|
32
|
+
import { t as Ti, s as Mn, a as zt, m as mh, r as gh, b as xh, c as Ch, d as bh } from "./RealDiv-Dy0p8Bvo.js";
|
|
33
33
|
/**
|
|
34
34
|
* @license
|
|
35
35
|
* Copyright 2018 Google LLC. All Rights Reserved.
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { j as p, a1 as w } from "./index-
|
|
1
|
+
import { j as p, a1 as w } from "./index-BoWRt-10.js";
|
|
2
2
|
function k(o, t, r) {
|
|
3
3
|
const n = t.rank > 1 ? t.shape[t.rank - 1] : 1, e = t.rank > 1 ? t.rank - 1 : 1, h = `Must have updates.shape = indices.shape[:batchDim] + shape[sliceDim:], got updates.shape: ${r.shape}, indices.shape: ${t.shape}, shape: ${o}, sliceDim: ${n}, and batchDim: ${e}.`;
|
|
4
4
|
if (r.rank < e)
|
|
@@ -1,6 +1,6 @@
|
|
|
1
|
-
import { l as N, aq as H, B as _, C as S, E as L, ar as te, as as ne, an as se, ap as re, at as ie, au as oe, av as le, a as ue, aw as fe, ax as O } from "./index-
|
|
2
|
-
import { r as ce } from "./reshape-
|
|
3
|
-
import { s as ae } from "./sum-
|
|
1
|
+
import { l as N, aq as H, B as _, C as S, E as L, ar as te, as as ne, an as se, ap as re, at as ie, au as oe, av as le, a as ue, aw as fe, ax as O } from "./index-BoWRt-10.js";
|
|
2
|
+
import { r as ce } from "./reshape-CdBq1WJ6.js";
|
|
3
|
+
import { s as ae } from "./sum-66ew2byf.js";
|
|
4
4
|
/**
|
|
5
5
|
* @license
|
|
6
6
|
* Copyright 2020 Google LLC. All Rights Reserved.
|
|
@@ -1,9 +1,9 @@
|
|
|
1
|
-
import { l as Lt, bf as Ce, j as V, k as Pt, a1 as H, bg as et, aa as ht, bh as dt, bi as at, a6 as De, a7 as st, am as We, b2 as $e, bj as Ue, bk as ze, bl as Be, bm as qt, al as K, bn as At, bo as W, bp as _t, bq as Vt, ae as Ct, br as Dt, ak as Wt, bs as $t, aW as Ut, aX as zt, aZ as Bt, a_ as jt, af as Gt, bt as Zt, b1 as Kt, bu as Ht, a$ as je, ao as Ge, bv as Xt, b5 as Ze, bw as Ke, p as Jt, u as He, t as Xe, bx as Qt, by as It, ar as Yt, W as Je, bz as te, bA as ee, bB as ne, aM as pt, bC as se, bD as ot } from "./index-
|
|
2
|
-
import { f as it, m as kt, g as Qe, a as Ye, R as tn, v as en, d as nn, e as sn, h as on, i as rn, j as an, k as ln, l as cn, n as un, o as hn, p as fn, q as Nt, r as dn, s as gn, t as mn } from "./backend_util-
|
|
3
|
-
import { g as pn, a as wn, e as In, c as bn } from "./axis_util-
|
|
4
|
-
import { b as xn } from "./broadcast_to-
|
|
5
|
-
import { r as En } from "./reshape-
|
|
6
|
-
import { p as Fn, q as yn, r as kn, u as Nn } from "./selu_util-
|
|
1
|
+
import { l as Lt, bf as Ce, j as V, k as Pt, a1 as H, bg as et, aa as ht, bh as dt, bi as at, a6 as De, a7 as st, am as We, b2 as $e, bj as Ue, bk as ze, bl as Be, bm as qt, al as K, bn as At, bo as W, bp as _t, bq as Vt, ae as Ct, br as Dt, ak as Wt, bs as $t, aW as Ut, aX as zt, aZ as Bt, a_ as jt, af as Gt, bt as Zt, b1 as Kt, bu as Ht, a$ as je, ao as Ge, bv as Xt, b5 as Ze, bw as Ke, p as Jt, u as He, t as Xe, bx as Qt, by as It, ar as Yt, W as Je, bz as te, bA as ee, bB as ne, aM as pt, bC as se, bD as ot } from "./index-BoWRt-10.js";
|
|
2
|
+
import { f as it, m as kt, g as Qe, a as Ye, R as tn, v as en, d as nn, e as sn, h as on, i as rn, j as an, k as ln, l as cn, n as un, o as hn, p as fn, q as Nt, r as dn, s as gn, t as mn } from "./backend_util-TE7aTPhZ.js";
|
|
3
|
+
import { g as pn, a as wn, e as In, c as bn } from "./axis_util-BzbKo31C.js";
|
|
4
|
+
import { b as xn } from "./broadcast_to-CdbwV-Dj.js";
|
|
5
|
+
import { r as En } from "./reshape-CdBq1WJ6.js";
|
|
6
|
+
import { p as Fn, q as yn, r as kn, u as Nn } from "./selu_util-BJEXVvjX.js";
|
|
7
7
|
import { g as vn } from "./_commonjsHelpers-ByX85dGu.js";
|
|
8
8
|
function Sn(e, t) {
|
|
9
9
|
for (var n = 0; n < t.length; n++) {
|
package/dist/training/Adam.js
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { adamAdjust as b } from "../ops/adamAdjust.js";
|
|
2
2
|
import { adamMoments as d } from "../ops/adamMoments.js";
|
|
3
|
-
import { O as g, e as h, t as o, d as B } from "../index-
|
|
4
|
-
import { z as M } from "../zeros
|
|
3
|
+
import { O as g, e as h, t as o, d as B } from "../index-BoWRt-10.js";
|
|
4
|
+
import { z as M } from "../zeros--BdLQ3oG.js";
|
|
5
5
|
/**
|
|
6
6
|
* @license
|
|
7
7
|
* Copyright 2018 Google LLC. All Rights Reserved.
|
package/dist/training/AdamExt.js
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { t as
|
|
2
|
-
import { d as
|
|
1
|
+
import { t as g } from "../index-BoWRt-10.js";
|
|
2
|
+
import { d as u, i as d } from "../dataset-CtdBYwjo.js";
|
|
3
3
|
import "../index-Tf7vU29b.js";
|
|
4
4
|
/**
|
|
5
5
|
* @license
|
|
@@ -18,57 +18,60 @@ import "../index-Tf7vU29b.js";
|
|
|
18
18
|
*
|
|
19
19
|
* =============================================================================
|
|
20
20
|
*/
|
|
21
|
-
function
|
|
22
|
-
return
|
|
23
|
-
const t = await
|
|
24
|
-
return
|
|
21
|
+
function z(r) {
|
|
22
|
+
return u(async () => {
|
|
23
|
+
const t = await r();
|
|
24
|
+
return d(() => t.next());
|
|
25
25
|
});
|
|
26
26
|
}
|
|
27
|
-
const
|
|
28
|
-
async function y(
|
|
29
|
-
const s = await Promise.all(
|
|
30
|
-
|
|
27
|
+
const S = 8;
|
|
28
|
+
async function y(r, t) {
|
|
29
|
+
const s = await Promise.all(r.map((e) => t.encode(e))), o = t.eosToken >= 0, a = s.map((e) => o ? [...e, t.eosToken] : e).flat();
|
|
30
|
+
for (const e of a)
|
|
31
|
+
if (e < 0 || e >= t.vocabSize)
|
|
32
|
+
throw new Error(`Invalid token index ${e} found in tokenised data`);
|
|
33
|
+
return a;
|
|
31
34
|
}
|
|
32
35
|
class w {
|
|
33
36
|
tokenizer;
|
|
34
37
|
blockSize;
|
|
35
38
|
pageSize;
|
|
36
39
|
constructor(t, s = 128) {
|
|
37
|
-
this.tokenizer = t, this.blockSize = s, this.pageSize = s *
|
|
40
|
+
this.tokenizer = t, this.blockSize = s, this.pageSize = s * S;
|
|
38
41
|
}
|
|
39
42
|
// Create dataset from text files
|
|
40
|
-
async createTextDataset(t, s = 32,
|
|
43
|
+
async createTextDataset(t, s = 32, o, a) {
|
|
41
44
|
if (t.length < this.blockSize + 1)
|
|
42
45
|
throw new Error(`Not enough tokens (${t.length}) for block size ${this.blockSize}`);
|
|
43
|
-
if (
|
|
46
|
+
if (o && o.size > t.length / this.pageSize / 2)
|
|
44
47
|
throw new Error("Too many masked pages - would leave insufficient training data");
|
|
45
|
-
const
|
|
46
|
-
if (
|
|
47
|
-
const
|
|
48
|
+
const e = (function* () {
|
|
49
|
+
if (o && a) {
|
|
50
|
+
const i = Array.from(o);
|
|
48
51
|
for (; ; ) {
|
|
49
|
-
const
|
|
50
|
-
if (
|
|
52
|
+
const c = Math.floor(Math.random() * i.length), l = Math.floor(Math.random() * this.pageSize), n = i[c] * this.pageSize + l;
|
|
53
|
+
if (n + this.blockSize + 1 > t.length)
|
|
51
54
|
continue;
|
|
52
|
-
const h = t.slice(
|
|
53
|
-
yield { xs: h, ys:
|
|
55
|
+
const h = t.slice(n, n + this.blockSize), f = t.slice(n + 1, n + this.blockSize + 1);
|
|
56
|
+
yield { xs: h, ys: f };
|
|
54
57
|
}
|
|
55
58
|
} else
|
|
56
59
|
for (; ; ) {
|
|
57
|
-
const
|
|
58
|
-
if (
|
|
59
|
-
const
|
|
60
|
-
if (h && !
|
|
60
|
+
const i = Math.floor(Math.random() * (t.length - this.blockSize - 1));
|
|
61
|
+
if (o) {
|
|
62
|
+
const n = Math.floor(i / this.pageSize), h = o.has(n);
|
|
63
|
+
if (h && !a || !h && a)
|
|
61
64
|
continue;
|
|
62
65
|
}
|
|
63
|
-
const
|
|
64
|
-
yield { xs:
|
|
66
|
+
const c = t.slice(i, i + this.blockSize), l = t.slice(i + 1, i + this.blockSize + 1);
|
|
67
|
+
yield { xs: c, ys: l };
|
|
65
68
|
}
|
|
66
69
|
}).bind(this);
|
|
67
|
-
return
|
|
68
|
-
const
|
|
69
|
-
return
|
|
70
|
-
xs:
|
|
71
|
-
ys:
|
|
70
|
+
return z(e).batch(s).map((i) => {
|
|
71
|
+
const c = i;
|
|
72
|
+
return g(() => ({
|
|
73
|
+
xs: c.xs.cast("int32"),
|
|
74
|
+
ys: c.ys.cast("int32")
|
|
72
75
|
// this.tf.oneHot(batchData.ys.cast('int32'), this.tokenizer.vocabSize),
|
|
73
76
|
}));
|
|
74
77
|
}).prefetch(2);
|
|
@@ -76,6 +79,6 @@ class w {
|
|
|
76
79
|
}
|
|
77
80
|
export {
|
|
78
81
|
w as DatasetBuilder,
|
|
79
|
-
|
|
82
|
+
S as PAGE_FACTOR,
|
|
80
83
|
y as flattenTokens
|
|
81
84
|
};
|
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { generateText as w } from "../utilities/generate.js";
|
|
2
2
|
import T from "./Trainer.js";
|
|
3
3
|
import L from "./Evaluator.js";
|
|
4
|
-
import { d as h } from "../index-
|
|
4
|
+
import { d as h } from "../index-BoWRt-10.js";
|
|
5
5
|
import x from "../utilities/profile.js";
|
|
6
6
|
const y = {
|
|
7
7
|
desiredLoss: 0.01,
|
package/dist/training/Trainer.js
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
import { DatasetBuilder as h, flattenTokens as p, PAGE_FACTOR as g } from "./DatasetBuilder.js";
|
|
2
2
|
import u from "./AdamExt.js";
|
|
3
|
-
import { t as f, v as y, d as c } from "../index-
|
|
4
|
-
import { z as m } from "../zeros
|
|
3
|
+
import { t as f, v as y, d as c } from "../index-BoWRt-10.js";
|
|
4
|
+
import { z as m } from "../zeros--BdLQ3oG.js";
|
|
5
5
|
class x {
|
|
6
6
|
constructor(t, e, a = 1e-3) {
|
|
7
7
|
this.tokenizer = e, this.model = t, this.learningRate = a, this.resetOptimizer(), this.datasetBuilder = new h(e, t.config.gpt.blockSize);
|
|
@@ -1,9 +1,9 @@
|
|
|
1
1
|
import { gatherSub as x } from "../ops/gatherSub.js";
|
|
2
2
|
import { scatterSub as L } from "../ops/scatterSub.js";
|
|
3
|
-
import { y, t as u, z as C, c as E } from "../index-
|
|
4
|
-
import { s as G } from "../softmax-
|
|
5
|
-
import { m as z } from "../max-
|
|
6
|
-
import { l as v } from "../log_sum_exp-
|
|
3
|
+
import { y, t as u, z as C, c as E } from "../index-BoWRt-10.js";
|
|
4
|
+
import { s as G } from "../softmax-Bv_6lyMX.js";
|
|
5
|
+
import { m as z } from "../max-Ddnnb5xe.js";
|
|
6
|
+
import { l as v } from "../log_sum_exp-DbjkV734.js";
|
|
7
7
|
function k(t, s) {
|
|
8
8
|
return u(() => {
|
|
9
9
|
const n = t.shape[t.shape.length - 1], c = t.shape.slice(0, -1).reduce((o, e) => o * e, 1), h = t.shape.length > 2 ? t.reshape([c, n]) : t, p = s.shape.length > 1 ? s.reshape([c]).cast("int32") : s.cast("int32"), r = z(h, -1, !0), a = E(h, r), m = v(a, -1);
|
package/dist/utilities/dummy.js
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import { m as y, v as P, e as S } from "../index-
|
|
2
|
-
import { z as i } from "../zeros
|
|
1
|
+
import { m as y, v as P, e as S } from "../index-BoWRt-10.js";
|
|
2
|
+
import { z as i } from "../zeros--BdLQ3oG.js";
|
|
3
3
|
async function w(s) {
|
|
4
4
|
const t = i([1, s.config.gpt.blockSize], "int32"), [e, n] = s.forward({ training: !1 }, t);
|
|
5
5
|
await e.data(), e.dispose(), n && n.dispose(), t.dispose();
|
|
@@ -1,6 +1,6 @@
|
|
|
1
|
-
import "../index-
|
|
2
|
-
import { t as m } from "../tensor2d-
|
|
3
|
-
import { c as u } from "../concat-
|
|
1
|
+
import "../index-BoWRt-10.js";
|
|
2
|
+
import { t as m } from "../tensor2d-wxPAnDQy.js";
|
|
3
|
+
import { c as u } from "../concat-CsxrgovM.js";
|
|
4
4
|
async function v(o, r, a, c, f) {
|
|
5
5
|
if (c <= 0)
|
|
6
6
|
throw new Error("Length must be a positive integer");
|
|
@@ -1,5 +1,5 @@
|
|
|
1
|
-
import "../index-
|
|
2
|
-
import { t as p } from "../tensor-
|
|
1
|
+
import "../index-BoWRt-10.js";
|
|
2
|
+
import { t as p } from "../tensor-JwS7ZYY6.js";
|
|
3
3
|
function h(n) {
|
|
4
4
|
const e = n.reduce((s, o) => s + o.length, 0), a = new Float32Array(e);
|
|
5
5
|
let t = 0;
|
|
@@ -1,4 +1,4 @@
|
|
|
1
|
-
import { B as m, C as r,
|
|
1
|
+
import { B as m, C as r, a2 as l, E as c, a6 as i, F as p, a7 as u, j as f } from "./index-BoWRt-10.js";
|
|
2
2
|
/**
|
|
3
3
|
* @license
|
|
4
4
|
* Copyright 2020 Google LLC. All Rights Reserved.
|