@stellarapp/tfjs-stellar 1.0.0 → 1.0.2

package/dist/utils.test.js

@@ -0,0 +1,73 @@
+import * as tf from "@tensorflow/tfjs";
+import { getScaleShape, getRandomCropStart } from "@/utils";
+import { causal } from "@/masks";
+// avoid TFJS node message during Jest testing
+tf.env().set('IS_NODE', false);
+describe("test custom TFJS utility functions", () => {
+    test("crop an image using the same shape, results in same shape", async () => {
+        // cropping an image of the same shape
+        const img_size = [133, 84];
+        const target_size = [133, 84];
+        expect(getRandomCropStart(img_size, target_size)).toEqual([0, 0, 0]);
+    });
+    it("should throw when crop is larger than image", async () => {
+        expect(() => getRandomCropStart([128, 128], [1000, 2000])).toThrow();
+    });
+    test("cropped image shape", async () => {
+        // cropping from wide to tall image
+        for (let i = 0; i < 100; i++) {
+            const img_size = [4923, 832];
+            const target_size = [333, 739];
+            const [crop_start_h, crop_start_w, channels] = getRandomCropStart(img_size, target_size);
+            expect(crop_start_h).toBeLessThanOrEqual(img_size[0] - target_size[0]);
+            expect(crop_start_w).toBeLessThanOrEqual(img_size[1] - target_size[1]);
+        }
+        // cropping from tall to wide image
+        for (let i = 0; i < 100; i++) {
+            const img_size = [381, 999];
+            const target_size = [300, 157];
+            const [crop_start_h, crop_start_w, channels] = getRandomCropStart(img_size, target_size);
+            expect(crop_start_h).toBeLessThanOrEqual(img_size[0] - target_size[0]);
+            expect(crop_start_w).toBeLessThanOrEqual(img_size[1] - target_size[1]);
+        }
+    });
+    test("scale 1:1, results in the same shape", async () => {
+        const scale = getScaleShape([256, 256], [256, 256]);
+        expect(scale).toEqual([256, 256]);
+    });
+    test("scaled image shape", async () => {
+        // scaling squares result in squares
+        const scale1 = getScaleShape([256, 256], [128, 128]);
+        expect(scale1).toEqual([128, 128]);
+        const scale2 = getScaleShape([128, 128], [256, 256]);
+        expect(scale2).toEqual([256, 256]);
+        const scale3 = getScaleShape([123, 123], [321, 321]);
+        expect(scale3).toEqual([321, 321]);
+        const scale4 = getScaleShape([321, 321], [123, 123]);
+        expect(scale4).toEqual([123, 123]);
+        // scaling rectangles result in rectangles
+        const scale5 = getScaleShape([640, 480], [1280, 960]);
+        expect(scale5).toEqual([1280, 960]);
+        const scale6 = getScaleShape([480, 640], [960, 1280]);
+        expect(scale6).toEqual([960, 1280]);
+        const [scale7_h, scale7_w] = getScaleShape([777, 555], [555, 333]);
+        expect(scale7_h).toBeGreaterThan(scale7_w);
+        const [scale8_h, scale8_w] = getScaleShape([555, 777], [333, 555]);
+        expect(scale8_h).toBeLessThan(scale8_w);
+    });
+    test("causal attention map", async () => {
+        const seq_len = 4;
+        const causal_mask = causal(seq_len, seq_len);
+        const _ = -1e7;
+        const expected_mask = tf.tensor([
+            [0, _, _, _],
+            [0, 0, _, _],
+            [0, 0, 0, _],
+            [0, 0, 0, 0]
+        ]);
+        // this might fail due to precision issues on the masked positions,
+        // in which case use less <= to 6 or 12 (number of masked positions x2)
+        expect((await causal_mask.sub(expected_mask).sum().data())[0]).toEqual(0);
+    });
+});
+//# sourceMappingURL=utils.test.js.map

package/dist/utils.test.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"utils.test.js","sourceRoot":"","sources":["../src/utils.test.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,MAAM,kBAAkB,CAAC;AACvC,OAAO,EAAE,aAAa,EAAE,kBAAkB,EAAE,MAAM,SAAS,CAAC;AAC5D,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AAEjC,8CAA8C;AAC9C,EAAE,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,SAAS,EAAE,KAAK,CAAC,CAAC;AAG/B,QAAQ,CAAC,oCAAoC,EAAE,GAAG,EAAE;IAEhD,IAAI,CAAC,2DAA2D,EAAE,KAAK,IAAI,EAAE;QACzE,sCAAsC;QACtC,MAAM,QAAQ,GAAG,CAAC,GAAG,EAAE,EAAE,CAAqB,CAAC;QAC/C,MAAM,WAAW,GAAG,CAAC,GAAG,EAAE,EAAE,CAAqB,CAAC;QAElD,MAAM,CAAC,kBAAkB,CAAC,QAAQ,EAAE,WAAW,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC;IACzE,CAAC,CAAC,CAAC;IAGH,EAAE,CAAC,6CAA6C,EAAE,KAAK,IAAI,EAAE;QACzD,MAAM,CAAC,GAAG,EAAE,CAAC,kBAAkB,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,CAAC,CAAC,OAAO,EAAE,CAAC;IACzE,CAAC,CAAC,CAAA;IAGF,IAAI,CAAC,qBAAqB,EAAE,KAAK,IAAI,EAAE;QACnC,mCAAmC;QACnC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;YAC3B,MAAM,QAAQ,GAAG,CAAC,IAAI,EAAE,GAAG,CAAqB,CAAC;YACjD,MAAM,WAAW,GAAG,CAAC,GAAG,EAAE,GAAG,CAAqB,CAAC;YAEnD,MAAM,CAAC,YAAY,EAAE,YAAY,EAAE,QAAQ,CAAC,GAAG,kBAAkB,CAAC,QAAQ,EAAE,WAAW,CAAC,CAAA;YAExF,MAAM,CAAC,YAAY,CAAC,CAAC,mBAAmB,CAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC;YACvE,MAAM,CAAC,YAAY,CAAC,CAAC,mBAAmB,CAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC;QAC3E,CAAC;QAED,mCAAmC;QACnC,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,GAAG,EAAE,CAAC,EAAE,EAAE,CAAC;YAC3B,MAAM,QAAQ,GAAG,CAAC,GAAG,EAAE,GAAG,CAAqB,CAAC;YAChD,MAAM,WAAW,GAAG,CAAC,GAAG,EAAE,GAAG,CAAqB,CAAC;YAEnD,MAAM,CAAC,YAAY,EAAE,YAAY,EAAE,QAAQ,CAAC,GAAG,kBAAkB,CAAC,QAAQ,EAAE,WAAW,CAAC,CAAA;YAExF,MAAM,CAAC,YAAY,CAAC,CAAC,mBAAmB,CAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC;YACvE,MAAM,CAAC,YAAY,CAAC,CAAC,mBAAmB,CAAC,QAAQ,CAAC,CAAC,CAAC,GAAG,WAAW,CAAC,CAAC,CAAC,CAAC,CAAC;QAC3E,CAAC;IACL,CAAC,CAAC,CAAC;IAGH,IAAI,CAAC,sCAAsC,EAAE,KAAK,IAAI,EAAE;QACpD,MAAM,KAAK,GAAG,aAAa,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAA;QACnD,MAAM,CAAC,KAAK,CAAC,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;IACtC,CAAC,CAAC,CAAC;IAGH,IAAI,CAAC,oBAAoB,EAAE,KAAK,IAAI,EAAE;QAClC,oCAAoC;QACpC,MAAM,MAAM,GAAG,aAAa,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAA;QACpD,MAAM,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;QAEnC,MAAM,MAAM,GAAG,aAAa,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAA;QACpD,MAAM,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;QAEnC,MAAM,MAAM,GAAG,aAAa,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAA;QACpD,MAAM,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;QAEnC,MAAM,MAAM,GAAG,aAAa,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAA;QACpD,MAAM,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC;QAEnC,0CAA0C;QAC1C,MAAM,MAAM,GAAG,aAAa,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAAA;QACrD,MAAM,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,GAAG,CAAC,CAAC,CAAC;QAEpC,MAAM,MAAM,GAAG,aAAa,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,CAAA;QACrD,MAAM,CAAC,MAAM,CAAC,CAAC,OAAO,CAAC,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,CAAC;QAEpC,MAAM,CAAC,QAAQ,EAAE,QAAQ,CAAC,GAAG,aAAa,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAA;QAClE,MAAM,CAAC,QAAQ,CAAC,CAAC,eAAe,CAAC,QAAQ,CAAC,CAAC;QAE3C,MAAM,CAAC,QAAQ,EAAE,QAAQ,CAAC,GAAG,aAAa,CAAC,CAAC,GAAG,EAAE,GAAG,CAAC,EAAE,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAA;QAClE,MAAM,CAAC,QAAQ,CAAC,CAAC,YAAY,CAAC,QAAQ,CAAC,CAAC;IAC5C,CAAC,CAAC,CAAC;IAGH,IAAI,CAAC,sBAAsB,EAAE,KAAK,IAAI,EAAE;QACpC,MAAM,OAAO,GAAG,CAAC,CAAC;QAClB,MAAM,WAAW,GAAG,MAAM,CAAC,OAAO,EAAE,OAAO,CAAC,CAAC;QAE7C,MAAM,CAAC,GAAG,CAAC,GAAG,CAAC;QACf,MAAM,aAAa,GAAG,EAAE,CAAC,MAAM,CAAC;YAC5B,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;YACZ,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;YACZ,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;YACZ,CAAC,CAAC,EAAE,CAAC,EAAE,CAAC,EAAE,CAAC,CAAC;SACf,CAAC,CAAC;QAEH,mEAAmE;QACnE,uEAAuE;QACvE,MAAM,CAAC,CAAC,MAAM,WAAW,CAAC,GAAG,CAAC,aAAa,CAAC,CAAC,GAAG,EAAE,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,CAAC,CAAC;IAC9E,CAAC,CAAC,CAAC;AAEP,CAAC,CAAC,CAAC"}

package/package.json

@@ -1,18 +1,24 @@
 {
   "name": "@stellarapp/tfjs-stellar",
-  "version": "1.0.0",
+  "version": "1.0.2",
   "description": "An extension of TensorFlow.js for implementing large language models.",
   "license": "ISC",
   "author": "",
   "type": "module",
-  "main": "index.ts",
+  "main": "dist/index.js",
+  "types": "dist/index.d.ts",
+  "files": [
+    "dist"
+  ],
   "scripts": {
-    "test": "npx jest"
+    "test": "npx jest",
+    "build": "tsc"
   },
   "devDependencies": {
     "@tensorflow/tfjs": "^4.22.0",
     "@types/jest": "^30.0.0",
     "@types/node": "^26.0.0",
+    "globals": "^17.6.0",
     "jest": "^30.4.2",
     "ts-jest": "^29.4.11",
     "tsx": "^4.22.4",
@@ -20,5 +26,9 @@
   },
   "peerDependencies": {
     "@tensorflow/tfjs": "*"
+  },
+  "repository": {
+    "type": "git",
+    "url": "https://github.com/rkuang9/tfjs-stellar.git"
   }
-}
+}

package/src/index.ts

@@ -1,93 +0,0 @@
-export * as models from "./models";
-export * as losses from "./losses";
-export * as metrics from "./metrics";
-
-import { MultiHeadAttention, type MultiHeadAttentionArgs } from "@/layers/multihead_attention";
-export { MultiHeadAttention, type MultiHeadAttentionArgs } from "@/layers/multihead_attention";
-
-import { CachedRoPEMultiHeadAttention } from "@/layers/cached_rope_multihead_attention";
-export { CachedRoPEMultiHeadAttention } from "@/layers/cached_rope_multihead_attention";
-
-import { TransformerEncoder, type TransformerEncoderArgs, } from "@/layers/transformer_encoder";
-export { TransformerEncoder, type TransformerEncoderArgs, } from "@/layers/transformer_encoder";
-
-import { TransformerDecoder, type TransformerDecoderArgs, } from "@/layers/transformer_decoder";
-export { TransformerDecoder, type TransformerDecoderArgs, } from "@/layers/transformer_decoder";
-
-import { TokenAndPositionalEmbedding, type TokenAndPositionalEmbeddingArgs } from "@/layers/token_and_positional_embedding";
-export { TokenAndPositionalEmbedding, type TokenAndPositionalEmbeddingArgs } from "@/layers/token_and_positional_embedding";
-
-import { PositionalEncoding, type PositionalEncodingArgs } from "@/layers/positional_encoding";
-export { PositionalEncoding, type PositionalEncodingArgs } from "@/layers/positional_encoding";
-
-import { GPT2DecoderBlock, type GPTDecoderBlockArgs } from "@/layers/gpt_decoder_block";
-export { GPT2DecoderBlock as GPTDecoderBlock, type GPTDecoderBlockArgs } from "@/layers/gpt_decoder_block";
-
-import { LlmModel, type LlmModelArgs } from "@/models/llm_model";
-export { LlmModel, type LlmModelArgs } from "@/models/llm_model";
-
-import { UNetModel, type UNetModelArgs } from "@/models/u_net";
-
-import { RotaryPositionEmbedding, type RotaryPositionEmbeddingArgs } from "@/layers/rotary_position_embedding";
-export { RotaryPositionEmbedding, type RotaryPositionEmbeddingArgs } from "@/layers/rotary_position_embedding";
-
-
-import { GptModel, type GptModelArgs } from "@/models/gpt_model";
-export { GptModel, type GptModelArgs } from "@/models/gpt_model";
-
-
-// The following exports give a keras-like import just like TFJS's tf.layers.<...>
-
-export function llmModel(args: LlmModelArgs) {
-    return new LlmModel(args);
-}
-
-
-export function gptModel(args: GptModelArgs) {
-    return new GptModel(args);
-}
-
-
-export function tokenAndPositionalEmbedding(args: TokenAndPositionalEmbeddingArgs) {
-    return new TokenAndPositionalEmbedding(args);
-}
-
-
-export function transformerEncoder(args: TransformerEncoderArgs) {
-    return new TransformerEncoder(args);
-}
-
-
-export function transformerDecoder(args: TransformerDecoderArgs) {
-    return new TransformerDecoder(args);
-}
-
-
-export function multiheadAttention(args: MultiHeadAttentionArgs) {
-    return new MultiHeadAttention(args);
-}
-
-
-export function cachedRopeMultiheadAttention(args: MultiHeadAttentionArgs) {
-    return new CachedRoPEMultiHeadAttention(args);
-}
-
-
-export function positionalEncoding(args: PositionalEncodingArgs) {
-    return new PositionalEncoding(args);
-}
-
-
-export function gpt2DecoderBlock(args: GPTDecoderBlockArgs) {
-    return new GPT2DecoderBlock(args);
-}
-
-
-export function unetModel(args: UNetModelArgs) {
-    return new UNetModel(args);
-}
-
-
-export function rotaryPositionEmbedding(args: RotaryPositionEmbeddingArgs) {
-    return new RotaryPositionEmbedding(args);
-}

package/src/layers/rotary_position_embedding.test.ts

@@ -1,107 +0,0 @@
-import { RotaryPositionEmbedding } from "@/layers/rotary_position_embedding";
-import * as tf from "@tensorflow/tfjs";
-
-// disables warning for using the faster node backend,
-// https://github.com/tensorflow/tfjs/issues/5349#issuecomment-885170504
-tf.env().set('IS_NODE', false);
-
-
-describe("RotaryPositionEmbedding tests", () => {
-    test("create cache", async () => {
-        const rope = new RotaryPositionEmbedding({ dim: 8, maxSequenceLength: 15 });
-        rope.build([]);
-
-        const expected_cosine_cache = tf.tensor([[[
-            [1, 1, 1, 1, 1, 1, 1, 1],
-            [0.5403022766113281, 0.5403022766113281, 0.9950041770935059, 0.9950041770935059, 0.9999499917030334, 0.9999499917030334, 0.9999995231628418, 0.9999995231628418],
-            [-0.416146844625473, -0.416146844625473, 0.9800665974617004, 0.9800665974617004, 0.9998000264167786, 0.9998000264167786, 0.9999979734420776, 0.9999979734420776],
-            [-0.9899924993515015, -0.9899924993515015, 0.9553365111351013, 0.9553365111351013, 0.9995500445365906, 0.9995500445365906, 0.9999955296516418, 0.9999955296516418],
-            [-0.6536436080932617, -0.6536436080932617, 0.9210609793663025, 0.9210609793663025, 0.9992001056671143, 0.9992001056671143, 0.9999920129776001, 0.9999920129776001],
-            [0.28366219997406006, 0.28366219997406006, 0.8775825500488281, 0.8775825500488281, 0.9987502694129944, 0.9987502694129944, 0.9999874830245972, 0.9999874830245972],
-            [0.9601702690124512, 0.9601702690124512, 0.8253356218338013, 0.8253356218338013, 0.998200535774231, 0.998200535774231, 0.9999819993972778, 0.9999819993972778],
-            [0.7539022564888, 0.7539022564888, 0.7648422122001648, 0.7648422122001648, 0.9975510239601135, 0.9975510239601135, 0.9999755024909973, 0.9999755024909973],
-            [-0.1455000340938568, -0.1455000340938568, 0.6967067122459412, 0.6967067122459412, 0.9968017339706421, 0.9968017339706421, 0.9999679923057556, 0.9999679923057556],
-            [-0.9111302495002747, -0.9111302495002747, 0.6216099262237549, 0.6216099262237549, 0.9959527254104614, 0.9959527254104614, 0.9999595284461975, 0.9999595284461975],
-            [-0.83907151222229, -0.83907151222229, 0.5403022766113281, 0.5403022766113281, 0.9950041770935059, 0.9950041770935059, 0.9999499917030334, 0.9999499917030334],
-            [0.004425697959959507, 0.004425697959959507, 0.4535960853099823, 0.4535960853099823, 0.9939560890197754, 0.9939560890197754, 0.999939501285553, 0.999939501285553],
-            [0.8438539505004883, 0.8438539505004883, 0.3623577058315277, 0.3623577058315277, 0.9928086400032043, 0.9928086400032043, 0.9999279975891113, 0.9999279975891113],
-            [0.9074468016624451, 0.9074468016624451, 0.26749876141548157, 0.26749876141548157, 0.9915618896484375, 0.9915618896484375, 0.9999154806137085, 0.9999154806137085],
-            [0.13673721253871918, 0.13673721253871918, 0.1699671596288681, 0.1699671596288681, 0.9902160167694092, 0.9902160167694092, 0.9999020099639893, 0.9999020099639893]
-        ]]]);
-
-        const expected_sine_cache = tf.tensor([[[
-            [0, 0, 0, 0, 0, 0, 0, 0],
-            [0.8414709568023682, 0.8414709568023682, 0.0998334214091301, 0.0998334214091301, 0.009999833069741726, 0.009999833069741726, 0.0009999999310821295, 0.0009999999310821295],
-            [0.9092974066734314, 0.9092974066734314, 0.19866932928562164, 0.19866932928562164, 0.019998665899038315, 0.019998665899038315, 0.0019999986980110407, 0.0019999986980110407],
-            [0.14112000167369843, 0.14112000167369843, 0.29552021622657776, 0.29552021622657776, 0.029995499178767204, 0.029995499178767204, 0.0029999956022948027, 0.0029999956022948027],
-            [-0.756802499294281, -0.756802499294281, 0.3894183337688446, 0.3894183337688446, 0.03998933359980583, 0.03998933359980583, 0.003999989479780197, 0.003999989479780197],
-            [-0.9589242935180664, -0.9589242935180664, 0.4794255495071411, 0.4794255495071411, 0.04997916519641876, 0.04997916519641876, 0.0049999793991446495, 0.0049999793991446495],
-            [-0.279415488243103, -0.279415488243103, 0.5646424889564514, 0.5646424889564514, 0.059964004904031754, 0.059964004904031754, 0.0059999641962349415, 0.0059999641962349415],
-            [0.6569865942001343, 0.6569865942001343, 0.6442176699638367, 0.6442176699638367, 0.06994284689426422, 0.06994284689426422, 0.0069999429397284985, 0.0069999429397284985],
-            [0.9893582463264465, 0.9893582463264465, 0.7173560857772827, 0.7173560857772827, 0.07991468906402588, 0.07991468906402588, 0.007999914698302746, 0.007999914698302746],
-            [0.41211849451065063, 0.41211849451065063, 0.7833269238471985, 0.7833269238471985, 0.08987854421138763, 0.08987854421138763, 0.008999879471957684, 0.008999879471957684],
-            [-0.5440211296081543, -0.5440211296081543, 0.8414709568023682, 0.8414709568023682, 0.0998334139585495, 0.0998334139585495, 0.0099998340010643, 0.0099998340010643],
-            [-0.9999902248382568, -0.9999902248382568, 0.8912073969841003, 0.8912073969841003, 0.10977829992771149, 0.10977829992771149, 0.010999779216945171, 0.010999779216945171],
-            [-0.5365729331970215, -0.5365729331970215, 0.9320390820503235, 0.9320390820503235, 0.11971220374107361, 0.11971220374107361, 0.011999712325632572, 0.011999712325632572],
-            [0.4201670289039612, 0.4201670289039612, 0.9635581970214844, 0.9635581970214844, 0.12963414192199707, 0.12963414192199707, 0.012999634258449078, 0.012999634258449078],
-            [0.9906073808670044, 0.9906073808670044, 0.9854497313499451, 0.9854497313499451, 0.13954311609268188, 0.13954311609268188, 0.013999543152749538, 0.013999543152749538]
-        ]]]);
-
-        const [cosine_cache, sine_cache] = rope.getWeights();
-
-        expect(await cosine_cache?.sub(expected_cosine_cache).sum().array() as number).toBeLessThanOrEqual(1e-6);
-        expect(await sine_cache?.sub(expected_sine_cache).sum().array() as number).toBeLessThanOrEqual(1e-6);
-    })
-
-
-    test("rotate inputs", async () => {
-        const rope = new RotaryPositionEmbedding({ dim: 8, maxSequenceLength: 15 });
-
-        const x = tf.tensor([[[
-            [0.0766048, 0.5706575, 0.6705932, 0.5273118, 0.4794086, 0.9378104, 0.9888024, 0.6926053],
-            [0.9064133, 0.5875182, 0.1681865, 0.3833345, 0.9901192, 0.4677338, 0.3353315, 0.02699],
-            [0.3033573, 0.4139377, 0.4062586, 0.9705839, 0.3582608, 0.328775, 0.1340587, 0.2193414],
-            [0.5565202, 0.4334963, 0.9912352, 0.3388563, 0.7991487, 0.1911893, 0.1140554, 0.6949552]]]
-        ]); // batch=1, seq = 1, heads=4, embedDim=8
-
-        const expected_output = tf.tensor([[[
-            [0.07660479843616486, 0.57065749168396, 0.6705932021141052, 0.5273118019104004, 0.4794085919857025, 0.9378104209899902, 0.9888023734092712, 0.6926053166389465],
-            [-0.004642367362976074, 1.08015775680542, 0.12907665967941284, 0.39820998907089233, 0.9853923320770264, 0.47761136293411255, 0.33530429005622864, 0.027325313538312912],
-            [-0.5026336908340454, 0.10358311235904694, 0.20533521473407745, 1.0319478511810303, 0.3516140580177307, 0.33587393164634705, 0.1336197406053543, 0.21960905194282532],
-            [-0.6121258735656738, -0.3506217896938324, 0.8468242287635803, 0.6166517734527588, 0.7930541634559631, 0.2150741070508957, 0.11197001487016678, 0.695294201374054]
-        ]]]);
-
-        const output = rope.apply(x) as tf.Tensor;
-
-        expect(await expected_output.sub(output).sum().array() as number).toBeLessThan(1e-6);
-        expect(rope.computeOutputShape(x.shape)).toEqual(x.shape);
-        expect(rope.computeOutputShape([x.shape])).toEqual(x.shape);
-    })
-
-
-    test("expand cache when input sequences are larger than rope's max sequence length", async () => {
-        const dim = 8;
-        const rope = new RotaryPositionEmbedding({ dim, maxSequenceLength: 15, theta: 1_000_000 });
-        const larger_sequence = 20;
-        const even_larger_sequence = 50;
-
-        rope.apply(tf.randomUniform([1, 1, larger_sequence, dim]));
-
-        rope.getWeights().forEach(weight => {
-            expect(weight.shape).toEqual([1, 1, 32, dim]);
-        });
-
-        rope.apply([tf.randomUniform([1, 1, even_larger_sequence, dim])]);
-
-        rope.getWeights().forEach(weight => {
-            expect(weight.shape).toEqual([1, 1, 64, dim]);
-        });
-    })
-
-
-    test("create layer", async () => {
-        // dim must be even
-        expect(() => new RotaryPositionEmbedding({ dim: 7, maxSequenceLength: 15 })).toThrow();
-        expect(() => new RotaryPositionEmbedding({ dim: 8, maxSequenceLength: 25 })).not.toThrow();
-    })
-});

package/src/losses/index.ts

@@ -1 +0,0 @@
-export * from "./dice";

package/src/testing.ts

@@ -1 +0,0 @@
-console.log("test")

package/tsconfig.json

@@ -1,49 +0,0 @@
-{
-  // Visit https://aka.ms/tsconfig to read more about this file
-  "compilerOptions": {
-    // File Layout
-    // "rootDir": "./src",
-    // "outDir": "./dist",
-    // Environment Settings
-    // See also https://aka.ms/tsconfig/module
-    "module": "es2022",
-    "target": "esnext",
-    "types": ["jest"],
-    // For nodejs:
-    // "lib": ["esnext"],
-    // "types": ["node"],
-    // and npm install -D @types/node
-    // Other Outputs
-    "sourceMap": true,
-    "declaration": true,
-    "declarationMap": true,
-    // Stricter Typechecking Options
-    //"noUncheckedIndexedAccess": true,
-    "exactOptionalPropertyTypes": true,
-    // Style Options
-    // "noImplicitReturns": true,
-    // "noImplicitOverride": true,
-    // "noUnusedLocals": true,
-    // "noUnusedParameters": true,
-    // "noFallthroughCasesInSwitch": true,
-    // "noPropertyAccessFromIndexSignature": true,
-    // Recommended Options
-    "strict": true,
-    "jsx": "react-jsx",
-    //"verbatimModuleSyntax": true,
-    "isolatedModules": true,
-    "noUncheckedSideEffectImports": true,
-    "moduleDetection": "force",
-    "skipLibCheck": true,
-    "paths": {
-      "@/*": [
-        "./src/*"
-      ],
-      "e2e/*": [
-        "./e2e/*"
-      ]
-    },
-    "moduleResolution": "bundler",
-    "esModuleInterop": true
-  }
-}