mini-jstorch 1.2.2 → 1.3.2

package/src/MainEngine.js

@@ -1 +1,560 @@
-console.log("MAJOR UPDATE IN 2:22 AM BE READY GUYS! this was empty because this versions only for an major note.")
+
+// MINI JSTORCH ENGINE
+// LICENSE: MIT (C) Rizal 2025
+// 1.2.3 ENGINE VERSIONS
+// IMPORTANT: CORE ENGINE DO NOT EDIT THIS FILES VERY SENSITIVE IT WILL CRASHING YOUR ENGINE SYSTEMS!
+
+// ---------------------- Utilities ----------------------
+function zeros(rows, cols) { return Array.from({length:rows},()=>Array(cols).fill(0)); }
+function ones(rows, cols) { return Array.from({length:rows},()=>Array(cols).fill(1)); }
+function randomMatrix(rows, cols, scale=0.1){ return Array.from({length:rows},()=>Array.from({length:cols},()=> (Math.random()*2-1)*scale)); }
+function transpose(matrix){ return matrix[0].map((_,i)=>matrix.map(row=>row[i])); }
+function addMatrices(a,b){ return a.map((row,i)=>row.map((v,j)=>v+(b[i] && b[i][j]!==undefined?b[i][j]:0))); }
+function dot(a,b){ const res=zeros(a.length,b[0].length); for(let i=0;i<a.length;i++) for(let j=0;j<b[0].length;j++) for(let k=0;k<a[0].length;k++) res[i][j]+=a[i][k]*b[k][j]; return res; }
+function softmax(x){ const m=Math.max(...x); const exps=x.map(v=>Math.exp(v-m)); const s=exps.reduce((a,b)=>a+b,0); return exps.map(v=>v/s); }
+function crossEntropy(pred,target){ const eps=1e-12; return -target.reduce((sum,t,i)=>sum+t*Math.log(pred[i]+eps),0); }
+
+// ---------------------- Tensor ----------------------
+export class Tensor {
+    constructor(data){ this.data=data; this.grad=zeros(data.length,data[0].length); }
+    shape(){ return [this.data.length,this.data[0].length]; }
+    add(t){ return t instanceof Tensor?this.data.map((r,i)=>r.map((v,j)=>v+t.data[i][j])):this.data.map(r=>r.map(v=>v+t)); }
+    sub(t){ return t instanceof Tensor?this.data.map((r,i)=>r.map((v,j)=>v-t.data[i][j])):this.data.map(r=>r.map(v=>v-t)); }
+    mul(t){ return t instanceof Tensor?this.data.map((r,i)=>r.map((v,j)=>v*t.data[i][j])):this.data.map(r=>r.map(v=>v*t)); }
+    matmul(t){ if(t instanceof Tensor) return dot(this.data,t.data); else throw new Error("matmul requires Tensor"); }
+    transpose(){ return transpose(this.data); }
+    flatten(){ return this.data.flat(); }
+    static zeros(r,c){ return new Tensor(zeros(r,c)); }
+    static ones(r,c){ return new Tensor(ones(r,c)); }
+    static random(r,c,scale=0.1){ return new Tensor(randomMatrix(r,c,scale)); }
+}
+
+// ---------------------- Layers ----------------------
+export class Linear {
+    constructor(inputDim,outputDim){
+        this.W=randomMatrix(inputDim,outputDim);
+        this.b=Array(outputDim).fill(0);
+        this.gradW=zeros(inputDim,outputDim);
+        this.gradb=Array(outputDim).fill(0);
+        this.x=null;
+    }
+
+    forward(x){
+        this.x=x;
+        const out=dot(x,this.W);
+        return out.map((row,i)=>row.map((v,j)=>v+this.b[j]));
+    }
+
+    backward(grad){
+        for(let i=0;i<this.W.length;i++) for(let j=0;j<this.W[0].length;j++)
+            this.gradW[i][j]=this.x.reduce((sum,row,k)=>sum+row[i]*grad[k][j],0);
+        for(let j=0;j<this.b.length;j++)
+            this.gradb[j]=grad.reduce((sum,row)=>sum+row[j],0);
+
+        const gradInput=zeros(this.x.length,this.W.length);
+        for(let i=0;i<this.x.length;i++)
+            for(let j=0;j<this.W.length;j++)
+                for(let k=0;k<this.W[0].length;k++)
+                    gradInput[i][j]+=grad[i][k]*this.W[j][k];
+        return gradInput;
+    }
+
+    parameters(){ return [ {param:this.W,grad:this.gradW}, {param:[this.b],grad:[this.gradb]} ]; }
+}
+
+// ---------------------- Conv2D ----------------------
+export class Conv2D {
+    constructor(inC,outC,kernel,stride=1,padding=0){
+        this.inC=inC; this.outC=outC; this.kernel=kernel;
+        this.stride=stride; this.padding=padding;
+        this.W=Array(outC).fill(0).map(()=>Array(inC).fill(0).map(()=>randomMatrix(kernel,kernel)));
+        this.gradW=Array(outC).fill(0).map(()=>Array(inC).fill(0).map(()=>zeros(kernel,kernel)));
+        this.x=null;
+    }
+
+    pad2D(input,pad){
+        return input.map(channel=>{
+            const rows=channel.length+2*pad;
+            const cols=channel[0].length+2*pad;
+            const out=Array.from({length:rows},()=>Array(cols).fill(0));
+            for(let i=0;i<channel.length;i++) for(let j=0;j<channel[0].length;j++) out[i+pad][j+pad]=channel[i][j];
+            return out;
+        });
+    }
+
+    conv2DSingle(input,kernel){
+        const rows=input.length-kernel.length+1;
+        const cols=input[0].length-kernel[0].length+1;
+        const out=zeros(rows,cols);
+        for(let i=0;i<rows;i++) for(let j=0;j<cols;j++)
+            for(let ki=0;ki<kernel.length;ki++) for(let kj=0;kj<kernel[0].length;kj++)
+                out[i][j]+=input[i+ki][j+kj]*kernel[ki][kj];
+        return out;
+    }
+
+    forward(batch){
+        this.x=batch;
+        return batch.map(sample=>{
+            const channelsOut=[];
+            for(let oc=0;oc<this.outC;oc++){
+                let outChan=zeros(sample[0].length,sample[0][0].length);
+                for(let ic=0;ic<this.inC;ic++){
+                    let inputChan=sample[ic];
+                    if(this.padding>0) inputChan=this.pad2D([inputChan],this.padding)[0];
+                    const conv=this.conv2DSingle(inputChan,this.W[oc][ic]);
+                    outChan=addMatrices(outChan,conv);
+                }
+                channelsOut.push(outChan);
+            }
+            return channelsOut;
+        });
+    }
+
+    backward(grad) {
+    const batchSize = this.x.length;
+    const gradInput = this.x.map(sample => sample.map(chan => zeros(chan.length, chan[0].length)));
+    const gradW = this.W.map(oc => oc.map(ic => zeros(this.kernel,this.kernel)));
+
+    for (let b = 0; b < batchSize; b++) {
+        const xPadded = this.pad2D(this.x[b], this.padding);
+        const gradInputPadded = xPadded.map(chan => zeros(chan.length, chan[0].length));
+
+        for (let oc = 0; oc < this.outC; oc++) {
+            for (let ic = 0; ic < this.inC; ic++) {
+                const outGrad = grad[b][oc];
+                const inChan = xPadded[ic];
+
+                // Compute gradW
+                for (let i = 0; i < this.kernel; i++) {
+                    for (let j = 0; j < this.kernel; j++) {
+                        let sum = 0;
+                        for (let y = 0; y < outGrad.length; y++) {
+                            for (let x = 0; x < outGrad[0].length; x++) {
+                                const inY = y * this.stride + i;
+                                const inX = x * this.stride + j;
+                                if (inY < inChan.length && inX < inChan[0].length) {
+                                    sum += inChan[inY][inX] * outGrad[y][x];
+                                }
+                            }
+                        }
+                        gradW[oc][ic][i][j] += sum;
+                    }
+                }
+
+                // Compute gradInput
+                const flippedKernel = this.W[oc][ic].map(row => [...row].reverse()).reverse();
+                for (let y = 0; y < outGrad.length; y++) {
+                    for (let x = 0; x < outGrad[0].length; x++) {
+                        for (let i = 0; i < this.kernel; i++) {
+                            for (let j = 0; j < this.kernel; j++) {
+                                const inY = y * this.stride + i;
+                                const inX = x * this.stride + j;
+                                if (inY < gradInputPadded[ic].length && inX < gradInputPadded[ic][0].length) {
+                                    gradInputPadded[ic][inY][inX] += flippedKernel[i][j] * outGrad[y][x];
+                                }
+                            }
+                        }
+                    }
+                }
+            }
+        }
+
+        // Remove padding from gradInput
+        if (this.padding > 0) {
+            for (let ic = 0; ic < this.inC; ic++) {
+                const padded = gradInputPadded[ic];
+                const cropped = padded.slice(this.padding, padded.length - this.padding)
+                    .map(row => row.slice(this.padding, row.length - this.padding));
+                gradInput[b][ic] = cropped;
+            }
+        } else {
+            for (let ic = 0; ic < this.inC; ic++) gradInput[b][ic] = gradInputPadded[ic];
+        }
+    }
+
+    this.gradW = gradW;
+    return gradInput;
+}
+
+    parameters(){ return this.W.flatMap((w,oc)=>w.map((wc,ic)=>({param:wc,grad:this.gradW[oc][ic]}))); }
+}
+
+// ---------------------- Sequential ----------------------
+export class Sequential {
+    constructor(layers=[]){ this.layers=layers; }
+    forward(x){ return this.layers.reduce((acc,l)=>l.forward(acc), x); }
+    backward(grad){ return this.layers.reduceRight((g,l)=>l.backward(g), grad); }
+    parameters(){ return this.layers.flatMap(l=>l.parameters?l.parameters():[]); }
+}
+
+// ---------------------- Activations ----------------------
+export class ReLU{ constructor(){ this.out=null; } forward(x){ this.out=x.map(r=>r.map(v=>Math.max(0,v))); return this.out; } backward(grad){ return grad.map((r,i)=>r.map((v,j)=>v*(this.out[i][j]>0?1:0))); } }
+export class Sigmoid{ constructor(){ this.out=null; } forward(x){ const fn=v=>1/(1+Math.exp(-v)); this.out=x.map(r=>r.map(fn)); return this.out; } backward(grad){ return grad.map((r,i)=>r.map((v,j)=>v*this.out[i][j]*(1-this.out[i][j]))); } }
+export class Tanh{ constructor(){ this.out=null; } forward(x){ this.out=x.map(r=>r.map(v=>Math.tanh(v))); return this.out; } backward(grad){ return grad.map((r,i)=>r.map((v,j)=>v*(1-this.out[i][j]**2))); } }
+export class LeakyReLU{ constructor(alpha=0.01){ this.alpha=alpha; this.out=null; } forward(x){ this.out=x.map(r=>r.map(v=>v>0?v:v*this.alpha)); return this.out; } backward(grad){ return grad.map((r,i)=>r.map((v,j)=>v*(this.out[i][j]>0?1:this.alpha))); } }
+export class GELU{ constructor(){ this.out=null; } forward(x){ const fn=v=>0.5*v*(1+Math.tanh(Math.sqrt(2/Math.PI)*(v+0.044715*v**3))); this.out=x.map(r=>r.map(fn)); return this.out; } backward(grad){ return grad.map((r,i)=>r.map(v=>v*1)); } }
+
+// ---------------------- Dropout ----------------------
+export class Dropout{ constructor(p=0.5){ this.p=p; } forward(x){ return x.map(r=>r.map(v=>v*Math.random()>=this.p?v:0)); } backward(grad){ return grad.map(r=>r.map(v=>v*(1-this.p))); } }
+
+// ---------------------- Losses ----------------------
+export class MSELoss{ forward(pred,target){ this.pred=pred; this.target=target; const losses=pred.map((row,i)=>row.reduce((sum,v,j)=>sum+(v-target[i][j])**2,0)/row.length); return losses.reduce((a,b)=>a+b,0)/pred.length; } backward(){ return this.pred.map((row,i)=>row.map((v,j)=>2*(v-this.target[i][j])/row.length)); } }
+export class CrossEntropyLoss{ forward(pred,target){ this.pred=pred; this.target=target; const losses=pred.map((p,i)=>crossEntropy(softmax(p),target[i])); return losses.reduce((a,b)=>a+b,0)/pred.length; } backward(){ return this.pred.map((p,i)=>{ const s=softmax(p); return s.map((v,j)=>(v-this.target[i][j])/this.pred.length); }); } }
+
+// ---------------------- Optimizers ----------------------
+export class Adam{
+    constructor(params,lr=0.001,b1=0.9,b2=0.999,eps=1e-8){
+        this.params=params; this.lr=lr; this.beta1=b1; this.beta2=b2; this.eps=eps;
+        this.m=params.map(p=>zeros(p.param.length,p.param[0].length||1));
+        this.v=params.map(p=>zeros(p.param.length,p.param[0].length||1));
+        this.t=0;
+    }
+    step(){
+        this.t++;
+        this.params.forEach((p,idx)=>{
+            for(let i=0;i<p.param.length;i++)
+                for(let j=0;j<(p.param[0].length||1);j++){
+                    const g=p.grad[i][j];
+                    this.m[idx][i][j]=this.beta1*this.m[idx][i][j]+(1-this.beta1)*g;
+                    this.v[idx][i][j]=this.beta2*this.v[idx][i][j]+(1-this.beta2)*g*g;
+                    const mHat=this.m[idx][i][j]/(1-Math.pow(this.beta1,this.t));
+                    const vHat=this.v[idx][i][j]/(1-Math.pow(this.beta2,this.t));
+                    p.param[i][j]-=this.lr*mHat/(Math.sqrt(vHat)+this.eps);
+                }
+        });
+    }
+}
+
+// ---------------------- ELU Activation ----------------------
+export class ELU {
+    constructor(alpha=1.0) {
+        this.alpha = alpha;
+        this.out = null;
+    }
+
+    forward(x) {
+        this.out = x.map(row => 
+            row.map(v => v > 0 ? v : this.alpha * (Math.exp(v) - 1))
+        );
+        return this.out;
+    }
+
+    backward(grad) {
+        return grad.map((row, i) => 
+            row.map((v, j) => 
+                v * (this.out[i][j] > 0 ? 1 : this.alpha * Math.exp(this.out[i][j]))
+            )
+        );
+    }
+}
+
+// ---------------------- Mish Activation ----------------------
+export class Mish {
+    constructor() {
+        this.x = null;
+    }
+
+    forward(x) {
+        this.x = x;
+        return x.map(row => 
+            row.map(v => {
+                // Mish(x) = x * tanh(softplus(x)) = x * tanh(ln(1 + e^x))
+                const softplus = Math.log(1 + Math.exp(v));
+                return v * Math.tanh(softplus);
+            })
+        );
+    }
+
+    backward(grad) {
+        return grad.map((row, i) => 
+            row.map((v, j) => {
+                const x_val = this.x[i][j];
+                
+                // Gradient of Mish: 
+                // δ = ω * (4(x+1) + 4e^2x + e^3x + e^x(4x+6)) / (2e^x + e^2x + 2)^2
+                // where ω = sech^2(softplus(x))
+                
+                const exp_x = Math.exp(x_val);
+                const exp_2x = Math.exp(2 * x_val);
+                const exp_3x = Math.exp(3 * x_val);
+                const softplus = Math.log(1 + exp_x);
+                
+                const sech_softplus = 1 / Math.cosh(softplus);
+                const numerator = 4 * (x_val + 1) + 4 * exp_2x + exp_3x + exp_x * (4 * x_val + 6);
+                const denominator = Math.pow(2 * exp_x + exp_2x + 2, 2);
+                
+                const mish_grad = (sech_softplus * sech_softplus) * (numerator / denominator);
+                return v * mish_grad;
+            })
+        );
+    }
+}
+
+// ---------------------- SiLU Activation ----------------------
+export class SiLU {
+    constructor() {
+        this.x = null;
+    }
+
+    forward(x) {
+        this.x = x;
+        return x.map(row => 
+            row.map(v => v / (1 + Math.exp(-v)))  // x * sigmoid(x)
+        );
+    }
+
+    backward(grad) {
+        return grad.map((row, i) => 
+            row.map((v, j) => {
+                const x_val = this.x[i][j];
+                const sigmoid = 1 / (1 + Math.exp(-x_val));
+                return v * (sigmoid * (1 + x_val * (1 - sigmoid)));
+            })
+        );
+    }
+}
+
+export class SGD{ constructor(params,lr=0.01){ this.params=params; this.lr=lr; } step(){ this.params.forEach(p=>{ for(let i=0;i<p.param.length;i++) for(let j=0;j<(p.param[0].length||1);j++) p.param[i][j]-=this.lr*p.grad[i][j]; }); } }
+
+// ---------------------- BatchNorm2D ----------------------
+export class BatchNorm2d {
+    constructor(numFeatures, eps=1e-5, momentum=0.1, affine=true) {
+        this.numFeatures = numFeatures;
+        this.eps = eps;
+        this.momentum = momentum;
+        this.affine = affine;
+        
+        // Parameters
+        if (affine) {
+            this.weight = Array(numFeatures).fill(1);
+            this.bias = Array(numFeatures).fill(0);
+            this.gradWeight = Array(numFeatures).fill(0);
+            this.gradBias = Array(numFeatures).fill(0);
+        }
+        
+        // Running statistics
+        this.runningMean = Array(numFeatures).fill(0);
+        this.runningVar = Array(numFeatures).fill(1);
+        
+        // Training state
+        this.training = true;
+        this.x = null;
+        this.xCentered = null;
+        this.std = null;
+    }
+
+    forward(x) {
+        // x shape: [batch, channels, height, width]
+        this.x = x;
+        const batchSize = x.length;
+        const channels = x[0].length;
+        
+        if (this.training) {
+            // Calculate mean per channel
+            const means = Array(channels).fill(0);
+            for (let b = 0; b < batchSize; b++) {
+                for (let c = 0; c < channels; c++) {
+                    const channelData = x[b][c];
+                    let sum = 0;
+                    for (let i = 0; i < channelData.length; i++) {
+                        for (let j = 0; j < channelData[0].length; j++) {
+                            sum += channelData[i][j];
+                        }
+                    }
+                    means[c] += sum / (channelData.length * channelData[0].length);
+                }
+            }
+            means.forEach((_, c) => means[c] /= batchSize);
+            
+            // Calculate variance per channel
+            const variances = Array(channels).fill(0);
+            for (let b = 0; b < batchSize; b++) {
+                for (let c = 0; c < channels; c++) {
+                    const channelData = x[b][c];
+                    let sum = 0;
+                    for (let i = 0; i < channelData.length; i++) {
+                        for (let j = 0; j < channelData[0].length; j++) {
+                            sum += Math.pow(channelData[i][j] - means[c], 2);
+                        }
+                    }
+                    variances[c] += sum / (channelData.length * channelData[0].length);
+                }
+            }
+            variances.forEach((_, c) => variances[c] /= batchSize);
+            
+            // Update running statistics
+            for (let c = 0; c < channels; c++) {
+                this.runningMean[c] = this.momentum * means[c] + (1 - this.momentum) * this.runningMean[c];
+                this.runningVar[c] = this.momentum * variances[c] + (1 - this.momentum) * this.runningVar[c];
+            }
+            
+            // Normalize
+            this.xCentered = [];
+            this.std = Array(channels).fill(0).map(() => []);
+            
+            const output = [];
+            for (let b = 0; b < batchSize; b++) {
+                const batchOut = [];
+                for (let c = 0; c < channels; c++) {
+                    const channelData = x[b][c];
+                    const channelOut = zeros(channelData.length, channelData[0].length);
+                    const channelCentered = zeros(channelData.length, channelData[0].length);
+                    const channelStd = Math.sqrt(variances[c] + this.eps);
+                    this.std[c].push(channelStd);
+                    
+                    for (let i = 0; i < channelData.length; i++) {
+                        for (let j = 0; j < channelData[0].length; j++) {
+                            channelCentered[i][j] = channelData[i][j] - means[c];
+                            channelOut[i][j] = channelCentered[i][j] / channelStd;
+                            
+                            // Apply affine transformation if enabled
+                            if (this.affine) {
+                                channelOut[i][j] = channelOut[i][j] * this.weight[c] + this.bias[c];
+                            }
+                        }
+                    }
+                    
+                    batchOut.push(channelOut);
+                    if (b === 0) this.xCentered.push(channelCentered);
+                    else this.xCentered[c] = addMatrices(this.xCentered[c], channelCentered);
+                }
+                output.push(batchOut);
+            }
+            
+            return output;
+        } else {
+            // Inference mode - use running statistics
+            const output = [];
+            for (let b = 0; b < batchSize; b++) {
+                const batchOut = [];
+                for (let c = 0; c < channels; c++) {
+                    const channelData = x[b][c];
+                    const channelOut = zeros(channelData.length, channelData[0].length);
+                    const channelStd = Math.sqrt(this.runningVar[c] + this.eps);
+                    
+                    for (let i = 0; i < channelData.length; i++) {
+                        for (let j = 0; j < channelData[0].length; j++) {
+                            channelOut[i][j] = (channelData[i][j] - this.runningMean[c]) / channelStd;
+                            
+                            // Apply affine transformation if enabled
+                            if (this.affine) {
+                                channelOut[i][j] = channelOut[i][j] * this.weight[c] + this.bias[c];
+                            }
+                        }
+                    }
+                    
+                    batchOut.push(channelOut);
+                }
+                output.push(batchOut);
+            }
+            
+            return output;
+        }
+    }
+
+    backward(gradOutput) {
+        if (!this.training) {
+            throw new Error("Backward should only be called in training mode");
+        }
+        
+        const batchSize = gradOutput.length;
+        const channels = gradOutput[0].length;
+        
+        // Initialize gradients
+        const gradInput = this.x.map(batch => 
+            batch.map(channel => 
+                zeros(channel.length, channel[0].length)
+            )
+        );
+        
+        if (this.affine) {
+            this.gradWeight.fill(0);
+            this.gradBias.fill(0);
+        }
+        
+        for (let c = 0; c < channels; c++) {
+            let sumGradWeight = 0;
+            let sumGradBias = 0;
+            
+            for (let b = 0; b < batchSize; b++) {
+                const channelGrad = gradOutput[b][c];
+                const channelData = this.x[b][c];
+                
+                // Calculate gradients for bias and weight
+                if (this.affine) {
+                    for (let i = 0; i < channelGrad.length; i++) {
+                        for (let j = 0; j < channelGrad[0].length; j++) {
+                            sumGradBias += channelGrad[i][j];
+                            sumGradWeight += channelGrad[i][j] * (this.xCentered[c][i][j] / this.std[c][b]);
+                        }
+                    }
+                }
+                
+                // Calculate gradient for input
+                const n = channelData.length * channelData[0].length;
+                const stdInv = 1 / this.std[c][b];
+                
+                for (let i = 0; i < channelGrad.length; i++) {
+                    for (let j = 0; j < channelGrad[0].length; j++) {
+                        let grad = channelGrad[i][j];
+                        
+                        if (this.affine) {
+                            grad *= this.weight[c];
+                        }
+                        
+                        grad *= stdInv;
+                        gradInput[b][c][i][j] = grad;
+                    }
+                }
+            }
+            
+            if (this.affine) {
+                this.gradWeight[c] = sumGradWeight / batchSize;
+                this.gradBias[c] = sumGradBias / batchSize;
+            }
+        }
+        
+        return gradInput;
+    }
+
+    parameters() {
+        if (!this.affine) return [];
+        return [
+            { param: [this.weight], grad: [this.gradWeight] },
+            { param: [this.bias], grad: [this.gradBias] }
+        ];
+    }
+
+    train() { this.training = true; }
+    eval() { this.training = false; }
+}
+
+// ---------------------- Model Save/Load ----------------------
+export function saveModel(model){
+    if(!(model instanceof Sequential)) throw new Error("saveModel supports only Sequential");
+    const weights=model.layers.map(layer=>({weights:layer.W||null,biases:layer.b||null}));
+    return JSON.stringify(weights);
+}
+
+export function loadModel(model,json){
+    if(!(model instanceof Sequential)) throw new Error("loadModel supports only Sequential");
+    const weights=JSON.parse(json);
+    model.layers.forEach((layer,i)=>{
+        if(layer.W && weights[i].weights) layer.W=weights[i].weights;
+        if(layer.b && weights[i].biases) layer.b=weights[i].biases;
+    });
+}
+
+// ---------------------- Advanced Utils ----------------------
+export function flattenBatch(batch){ return batch.flat(2); }
+export function stack(tensors){ return tensors.map(t=>t.data); }
+export function eye(n){ return Array.from({length:n},(_,i)=>Array.from({length:n},(_,j)=>i===j?1:0)); }
+export function concat(a,b,axis=0){ /* concat along axis */ if(axis===0) return [...a,...b]; if(axis===1) return a.map((row,i)=>[...row,...b[i]]); }
+export function reshape(tensor, rows, cols) {
+    let flat = tensor.data.flat(); // flatten dulu
+    if(flat.length < rows*cols) throw new Error("reshape size mismatch");
+    const out = Array.from({length: rows}, (_, i) =>
+        flat.slice(i*cols, i*cols + cols)
+    );
+    return out;
+}

package/src/startup.cpu

@@ -1,12 +1,15 @@
 // you can delete this files this files are not important for the engine runtime.
 
 e=run=[cpu[runtime]]
-e.set.runtime('beta')
-e.rnt()
-e.set()
-e.register('vanilla',expe='Experiments.js',main='MainEngine.js')
-l=e.prog('asm')
+devices=e.getdata[devices[5]]
+env.set.runtime('beta')
+env.rnt()
+env.set()
+env.register('vanilla',expe='Experiments.js',main='MainEngine.js',tgver=latest)
+resources=e.find(tag='resources')
+resources.ld(env)
+l=env.prog('asm')
 r=l.gv=[0xCAFEBABE]
-eng=e.load(register,r,'vanilla')
-eng.boot(e,r,'dp')
-eng.load()
+eng=env.load(register,r,'vanilla')
+eng.boot(env,r,'dp')
+eng.load(resources,runtime,devices)