catniff 0.6.15 → 0.7.0

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.
Files changed (9) hide show
  1. package/dist/lrscheduler.d.ts +13 -0
  2. package/dist/lrscheduler.js +31 -0
  3. package/dist/nn.d.ts +8 -9
  4. package/dist/nn.js +9 -1
  5. package/dist/optim.d.ts +9 -9
  6. package/dist/optim.js +11 -11
  7. package/index.d.ts +1 -0
  8. package/index.js +2 -1
  9. package/package.json +1 -1
package/dist/lrscheduler.d.ts ADDED
@@ -0,0 +1,13 @@
1
+ import { BaseOptimizer } from "./optim";
2
+ export declare class StepLR {
3
+ optimizer: BaseOptimizer;
4
+ stepSize: number;
5
+ gamma: number;
6
+ lastEpoch: number;
7
+ baseLR: number;
8
+ constructor(optimizer: BaseOptimizer, stepSize: number, gamma?: number, lastEpoch?: number);
9
+ step(epoch?: number): void;
10
+ }
11
+ export declare const LRScheduler: {
12
+ StepLR: typeof StepLR;
13
+ };
package/dist/lrscheduler.js ADDED
@@ -0,0 +1,31 @@
1
+ "use strict";
2
+ Object.defineProperty(exports, "__esModule", { value: true });
3
+ exports.LRScheduler = exports.StepLR = void 0;
4
+ class StepLR {
5
+ optimizer;
6
+ stepSize;
7
+ gamma;
8
+ lastEpoch;
9
+ baseLR;
10
+ constructor(optimizer, stepSize, gamma = 0.1, lastEpoch = -1) {
11
+ this.optimizer = optimizer;
12
+ this.stepSize = stepSize;
13
+ this.gamma = gamma;
14
+ this.lastEpoch = lastEpoch;
15
+ this.baseLR = this.optimizer.lr;
16
+ }
17
+ step(epoch) {
18
+ if (typeof epoch === "undefined") {
19
+ this.lastEpoch++;
20
+ epoch = this.lastEpoch;
21
+ }
22
+ else {
23
+ this.lastEpoch = epoch;
24
+ }
25
+ this.optimizer.lr = this.baseLR * this.gamma ** Math.floor(epoch / this.stepSize);
26
+ }
27
+ }
28
+ exports.StepLR = StepLR;
29
+ exports.LRScheduler = {
30
+ StepLR
31
+ };
package/dist/nn.d.ts CHANGED
@@ -1,11 +1,11 @@
1
1
  import { Tensor, TensorValue } from "./core";
2
- declare class Linear {
2
+ export declare class Linear {
3
3
  weight: Tensor;
4
4
  bias?: Tensor;
5
5
  constructor(inFeatures: number, outFeatures: number, bias?: boolean, device?: string);
6
6
  forward(input: Tensor | TensorValue): Tensor;
7
7
  }
8
- declare class RNNCell {
8
+ export declare class RNNCell {
9
9
  weightIH: Tensor;
10
10
  weightHH: Tensor;
11
11
  biasIH?: Tensor;
@@ -13,7 +13,7 @@ declare class RNNCell {
13
13
  constructor(inputSize: number, hiddenSize: number, bias?: boolean, device?: string);
14
14
  forward(input: Tensor | TensorValue, hidden: Tensor | TensorValue): Tensor;
15
15
  }
16
- declare class GRUCell {
16
+ export declare class GRUCell {
17
17
  weightIR: Tensor;
18
18
  weightIZ: Tensor;
19
19
  weightIN: Tensor;
@@ -29,7 +29,7 @@ declare class GRUCell {
29
29
  constructor(inputSize: number, hiddenSize: number, bias?: boolean, device?: string);
30
30
  forward(input: Tensor | TensorValue, hidden: Tensor | TensorValue): Tensor;
31
31
  }
32
- declare class LSTMCell {
32
+ export declare class LSTMCell {
33
33
  weightII: Tensor;
34
34
  weightIF: Tensor;
35
35
  weightIG: Tensor;
@@ -49,7 +49,7 @@ declare class LSTMCell {
49
49
  constructor(inputSize: number, hiddenSize: number, bias?: boolean, device?: string);
50
50
  forward(input: Tensor | TensorValue, hidden: Tensor | TensorValue, cell: Tensor | TensorValue): [Tensor, Tensor];
51
51
  }
52
- declare class LayerNorm {
52
+ export declare class LayerNorm {
53
53
  weight?: Tensor;
54
54
  bias?: Tensor;
55
55
  eps: number;
@@ -57,19 +57,19 @@ declare class LayerNorm {
57
57
  constructor(normalizedShape: number | number[], eps?: number, elementwiseAffine?: boolean, bias?: boolean, device?: string);
58
58
  forward(input: Tensor): Tensor;
59
59
  }
60
- declare class RMSNorm {
60
+ export declare class RMSNorm {
61
61
  weight?: Tensor;
62
62
  eps: number;
63
63
  normalizedShape: number[];
64
64
  constructor(normalizedShape: number | number[], eps?: number, elementwiseAffine?: boolean, device?: string);
65
65
  forward(input: Tensor): Tensor;
66
66
  }
67
- declare class Embedding {
67
+ export declare class Embedding {
68
68
  weight: Tensor;
69
69
  constructor(numEmbeddings: number, embeddingDim: number, device: string);
70
70
  forward(input: Tensor | TensorValue): Tensor;
71
71
  }
72
- declare class MultiheadAttention {
72
+ export declare class MultiheadAttention {
73
73
  qProjection: Linear;
74
74
  kProjection: Linear;
75
75
  vProjection: Linear;
@@ -100,4 +100,3 @@ export declare const nn: {
100
100
  loadStateDict(model: any, stateDict: StateDict, prefix?: string, visited?: WeakSet<object>): void;
101
101
  };
102
102
  };
103
- export {};
package/dist/nn.js CHANGED
@@ -1,6 +1,6 @@
1
1
  "use strict";
2
2
  Object.defineProperty(exports, "__esModule", { value: true });
3
- exports.nn = void 0;
3
+ exports.nn = exports.MultiheadAttention = exports.Embedding = exports.RMSNorm = exports.LayerNorm = exports.LSTMCell = exports.GRUCell = exports.RNNCell = exports.Linear = void 0;
4
4
  const core_1 = require("./core");
5
5
  function linearTransform(input, weight, bias) {
6
6
  let output = input.matmul(weight.t());
@@ -24,6 +24,7 @@ class Linear {
24
24
  return linearTransform(input, this.weight, this.bias);
25
25
  }
26
26
  }
27
+ exports.Linear = Linear;
27
28
  function rnnTransform(input, hidden, inputWeight, hiddenWeight, inputBias, hiddenBias) {
28
29
  let output = input.matmul(inputWeight.t()).add(hidden.matmul(hiddenWeight.t()));
29
30
  if (inputBias) {
@@ -54,6 +55,7 @@ class RNNCell {
54
55
  return rnnTransform(input, hidden, this.weightIH, this.weightHH, this.biasIH, this.biasHH).tanh();
55
56
  }
56
57
  }
58
+ exports.RNNCell = RNNCell;
57
59
  class GRUCell {
58
60
  weightIR;
59
61
  weightIZ;
@@ -93,6 +95,7 @@ class GRUCell {
93
95
  return (z.neg().add(1).mul(n).add(z.mul(hidden)));
94
96
  }
95
97
  }
98
+ exports.GRUCell = GRUCell;
96
99
  class LSTMCell {
97
100
  weightII;
98
101
  weightIF;
@@ -144,6 +147,7 @@ class LSTMCell {
144
147
  return [h, c];
145
148
  }
146
149
  }
150
+ exports.LSTMCell = LSTMCell;
147
151
  class LayerNorm {
148
152
  weight;
149
153
  bias;
@@ -188,6 +192,7 @@ class LayerNorm {
188
192
  return normalized;
189
193
  }
190
194
  }
195
+ exports.LayerNorm = LayerNorm;
191
196
  class RMSNorm {
192
197
  weight;
193
198
  eps;
@@ -224,6 +229,7 @@ class RMSNorm {
224
229
  return normalized;
225
230
  }
226
231
  }
232
+ exports.RMSNorm = RMSNorm;
227
233
  class Embedding {
228
234
  weight;
229
235
  constructor(numEmbeddings, embeddingDim, device) {
@@ -233,6 +239,7 @@ class Embedding {
233
239
  return this.weight.index(input);
234
240
  }
235
241
  }
242
+ exports.Embedding = Embedding;
236
243
  class MultiheadAttention {
237
244
  qProjection;
238
245
  kProjection;
@@ -284,6 +291,7 @@ class MultiheadAttention {
284
291
  return [output, needWeights ? attnWeights : undefined];
285
292
  }
286
293
  }
294
+ exports.MultiheadAttention = MultiheadAttention;
287
295
  const state = {
288
296
  getParameters(model, visited = new WeakSet()) {
289
297
  if (visited.has(model))
package/dist/optim.d.ts CHANGED
@@ -1,7 +1,11 @@
1
1
  import { Tensor } from "./core";
2
- declare abstract class BaseOptimizer {
2
+ export interface BaseOptimizerOptions {
3
+ lr?: number;
4
+ }
5
+ export declare abstract class BaseOptimizer {
3
6
  params: Tensor[];
4
- constructor(params: Tensor[]);
7
+ lr: number;
8
+ constructor(params: Tensor[], options?: BaseOptimizerOptions);
5
9
  zeroGrad(): void;
6
10
  }
7
11
  export interface SGDOptions {
@@ -11,9 +15,8 @@ export interface SGDOptions {
11
15
  weightDecay?: number;
12
16
  nesterov?: boolean;
13
17
  }
14
- declare class SGD extends BaseOptimizer {
18
+ export declare class SGD extends BaseOptimizer {
15
19
  momentumBuffers: Map<Tensor, Tensor>;
16
- lr: number;
17
20
  momentum: number;
18
21
  dampening: number;
19
22
  weightDecay: number;
@@ -27,11 +30,10 @@ export interface AdamOptions {
27
30
  eps?: number;
28
31
  weightDecay?: number;
29
32
  }
30
- declare class Adam extends BaseOptimizer {
33
+ export declare class Adam extends BaseOptimizer {
31
34
  momentumBuffers: Map<Tensor, Tensor>;
32
35
  velocityBuffers: Map<Tensor, Tensor>;
33
36
  stepCount: number;
34
- lr: number;
35
37
  betas: [number, number];
36
38
  eps: number;
37
39
  weightDecay: number;
@@ -44,11 +46,10 @@ export interface AdamWOptions {
44
46
  eps?: number;
45
47
  weightDecay?: number;
46
48
  }
47
- declare class AdamW extends BaseOptimizer {
49
+ export declare class AdamW extends BaseOptimizer {
48
50
  momentumBuffers: Map<Tensor, Tensor>;
49
51
  velocityBuffers: Map<Tensor, Tensor>;
50
52
  stepCount: number;
51
- lr: number;
52
53
  betas: [number, number];
53
54
  eps: number;
54
55
  weightDecay: number;
@@ -61,4 +62,3 @@ export declare const Optim: {
61
62
  Adam: typeof Adam;
62
63
  AdamW: typeof AdamW;
63
64
  };
64
- export {};
package/dist/optim.js CHANGED
@@ -1,11 +1,13 @@
1
1
  "use strict";
2
2
  Object.defineProperty(exports, "__esModule", { value: true });
3
- exports.Optim = void 0;
3
+ exports.Optim = exports.AdamW = exports.Adam = exports.SGD = exports.BaseOptimizer = void 0;
4
4
  const core_1 = require("./core");
5
5
  class BaseOptimizer {
6
6
  params;
7
- constructor(params) {
7
+ lr;
8
+ constructor(params, options) {
8
9
  this.params = params;
10
+ this.lr = options?.lr || 0.001;
9
11
  }
10
12
  zeroGrad() {
11
13
  for (let index = 0; index < this.params.length; index++) {
@@ -14,16 +16,15 @@ class BaseOptimizer {
14
16
  }
15
17
  }
16
18
  }
19
+ exports.BaseOptimizer = BaseOptimizer;
17
20
  class SGD extends BaseOptimizer {
18
21
  momentumBuffers = new Map();
19
- lr;
20
22
  momentum;
21
23
  dampening;
22
24
  weightDecay;
23
25
  nesterov;
24
26
  constructor(params, options) {
25
- super(params);
26
- this.lr = options?.lr || 0.001;
27
+ super(params, options);
27
28
  this.momentum = options?.momentum || 0;
28
29
  this.dampening = options?.dampening || 0;
29
30
  this.weightDecay = options?.weightDecay || 0;
@@ -66,17 +67,16 @@ class SGD extends BaseOptimizer {
66
67
  }
67
68
  }
68
69
  }
70
+ exports.SGD = SGD;
69
71
  class Adam extends BaseOptimizer {
70
72
  momentumBuffers = new Map(); // First moment (m_t)
71
73
  velocityBuffers = new Map(); // Second moment (v_t)
72
74
  stepCount = 0;
73
- lr;
74
75
  betas;
75
76
  eps;
76
77
  weightDecay;
77
78
  constructor(params, options) {
78
- super(params);
79
- this.lr = options?.lr || 0.001;
79
+ super(params, options);
80
80
  this.betas = options?.betas || [0.9, 0.999];
81
81
  this.eps = options?.eps || 1e-8;
82
82
  this.weightDecay = options?.weightDecay || 0;
@@ -126,17 +126,16 @@ class Adam extends BaseOptimizer {
126
126
  }
127
127
  }
128
128
  }
129
+ exports.Adam = Adam;
129
130
  class AdamW extends BaseOptimizer {
130
131
  momentumBuffers = new Map(); // First moment (m_t)
131
132
  velocityBuffers = new Map(); // Second moment (v_t)
132
133
  stepCount = 0;
133
- lr;
134
134
  betas;
135
135
  eps;
136
136
  weightDecay;
137
137
  constructor(params, options) {
138
- super(params);
139
- this.lr = options?.lr || 0.001;
138
+ super(params, options);
140
139
  this.betas = options?.betas || [0.9, 0.999];
141
140
  this.eps = options?.eps || 1e-8;
142
141
  this.weightDecay = options?.weightDecay || 0.01;
@@ -184,6 +183,7 @@ class AdamW extends BaseOptimizer {
184
183
  }
185
184
  }
186
185
  }
186
+ exports.AdamW = AdamW;
187
187
  exports.Optim = {
188
188
  BaseOptimizer,
189
189
  SGD,
package/index.d.ts CHANGED
@@ -1,3 +1,4 @@
1
1
  export * from "./dist/core";
2
2
  export * from "./dist/optim";
3
3
  export * from "./dist/nn";
4
+ export * from "./dist/lrscheduler";
package/index.js CHANGED
@@ -1,5 +1,6 @@
1
1
  module.exports = {
2
2
  ...require("./dist/core"),
3
3
  ...require("./dist/optim"),
4
- ...require("./dist/nn")
4
+ ...require("./dist/nn"),
5
+ ...require("./dist/lrscheduler")
5
6
  };
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "catniff",
3
- "version": "0.6.15",
3
+ "version": "0.7.0",
4
4
  "description": "A small Torch-like deep learning framework for Javascript",
5
5
  "main": "index.js",
6
6
  "scripts": {