modelfusion 0.122.0 → 0.124.0
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.
- package/CHANGELOG.md +56 -0
- package/README.md +42 -2
- package/index.cjs +0 -1
- package/index.d.ts +0 -1
- package/index.js +0 -1
- package/model-function/ModelCallEvent.d.ts +3 -2
- package/model-function/classify/Classifier.cjs +2 -0
- package/model-function/classify/Classifier.d.ts +10 -0
- package/model-function/classify/Classifier.js +1 -0
- package/model-function/classify/ClassifyEvent.cjs +2 -0
- package/model-function/classify/ClassifyEvent.d.ts +20 -0
- package/model-function/classify/ClassifyEvent.js +1 -0
- package/model-function/classify/EmbeddingSimilarityClassifier.cjs +97 -0
- package/model-function/classify/EmbeddingSimilarityClassifier.d.ts +40 -0
- package/model-function/classify/EmbeddingSimilarityClassifier.js +93 -0
- package/model-function/classify/classify.cjs +27 -0
- package/model-function/classify/classify.d.ts +17 -0
- package/model-function/classify/classify.js +23 -0
- package/{classifier → model-function/classify}/index.cjs +4 -1
- package/model-function/classify/index.d.ts +4 -0
- package/model-function/classify/index.js +4 -0
- package/model-function/index.cjs +1 -0
- package/model-function/index.d.ts +1 -0
- package/model-function/index.js +1 -0
- package/model-provider/mistral/MistralTextEmbeddingModel.d.ts +13 -13
- package/model-provider/ollama/OllamaChatModel.d.ts +9 -9
- package/model-provider/openai/AbstractOpenAITextEmbeddingModel.cjs +82 -0
- package/model-provider/openai/AbstractOpenAITextEmbeddingModel.d.ts +91 -0
- package/model-provider/openai/AbstractOpenAITextEmbeddingModel.js +78 -0
- package/model-provider/openai/OpenAIFacade.cjs +18 -18
- package/model-provider/openai/OpenAIFacade.d.ts +18 -18
- package/model-provider/openai/OpenAIFacade.js +18 -18
- package/model-provider/openai/OpenAITextEmbeddingModel.cjs +3 -68
- package/model-provider/openai/OpenAITextEmbeddingModel.d.ts +4 -82
- package/model-provider/openai/OpenAITextEmbeddingModel.js +3 -68
- package/model-provider/openai/index.cjs +1 -0
- package/model-provider/openai/index.d.ts +1 -0
- package/model-provider/openai/index.js +1 -0
- package/model-provider/openai-compatible/OpenAICompatibleChatModel.d.ts +2 -0
- package/model-provider/openai-compatible/OpenAICompatibleCompletionModel.d.ts +2 -0
- package/model-provider/openai-compatible/OpenAICompatibleFacade.cjs +28 -7
- package/model-provider/openai-compatible/OpenAICompatibleFacade.d.ts +24 -6
- package/model-provider/openai-compatible/OpenAICompatibleFacade.js +26 -6
- package/model-provider/openai-compatible/OpenAICompatibleTextEmbeddingModel.cjs +27 -0
- package/model-provider/openai-compatible/OpenAICompatibleTextEmbeddingModel.d.ts +18 -0
- package/model-provider/openai-compatible/OpenAICompatibleTextEmbeddingModel.js +23 -0
- package/model-provider/openai-compatible/index.cjs +1 -0
- package/model-provider/openai-compatible/index.d.ts +1 -0
- package/model-provider/openai-compatible/index.js +1 -0
- package/package.json +1 -1
- package/classifier/SemanticClassifier.cjs +0 -81
- package/classifier/SemanticClassifier.d.ts +0 -25
- package/classifier/SemanticClassifier.js +0 -77
- package/classifier/index.d.ts +0 -1
- package/classifier/index.js +0 -1
package/CHANGELOG.md
CHANGED
@@ -1,5 +1,61 @@
|
|
1
1
|
# Changelog
|
2
2
|
|
3
|
+
## v0.124.0 - 2024-01-13
|
4
|
+
|
5
|
+
### Added
|
6
|
+
|
7
|
+
- [Embedding-support for OpenAI-compatible providers](https://modelfusion.dev/integration/model-provider/openaicompatible/#embed-text). You can for example use the Together AI embedding endpoint:
|
8
|
+
|
9
|
+
```ts
|
10
|
+
import { embed, openaicompatible } from "modelfusion";
|
11
|
+
|
12
|
+
const embedding = await embed({
|
13
|
+
model: openaicompatible.TextEmbedder({
|
14
|
+
api: openaicompatible.TogetherAIApi(),
|
15
|
+
provider: "openaicompatible-togetherai",
|
16
|
+
model: "togethercomputer/m2-bert-80M-8k-retrieval",
|
17
|
+
}),
|
18
|
+
value: "At first, Nox didn't know what to do with the pup.",
|
19
|
+
});
|
20
|
+
```
|
21
|
+
|
22
|
+
## v0.123.0 - 2024-01-13
|
23
|
+
|
24
|
+
### Added
|
25
|
+
|
26
|
+
- `classify` model function ([docs](https://modelfusion.dev/guide/function/classify)) for classifying values. The `SemanticClassifier` has been renamed to `EmbeddingSimilarityClassifier` and can be used in conjunction with `classify`:
|
27
|
+
|
28
|
+
```ts
|
29
|
+
import { classify, EmbeddingSimilarityClassifier, openai } from "modelfusion";
|
30
|
+
|
31
|
+
const classifier = new EmbeddingSimilarityClassifier({
|
32
|
+
embeddingModel: openai.TextEmbedder({ model: "text-embedding-ada-002" }),
|
33
|
+
similarityThreshold: 0.82,
|
34
|
+
clusters: [
|
35
|
+
{
|
36
|
+
name: "politics" as const,
|
37
|
+
values: [
|
38
|
+
"they will save the country!",
|
39
|
+
// ...
|
40
|
+
],
|
41
|
+
},
|
42
|
+
{
|
43
|
+
name: "chitchat" as const,
|
44
|
+
values: [
|
45
|
+
"how's the weather today?",
|
46
|
+
// ...
|
47
|
+
],
|
48
|
+
},
|
49
|
+
],
|
50
|
+
});
|
51
|
+
|
52
|
+
// strongly typed result:
|
53
|
+
const result = await classify({
|
54
|
+
model: classifier,
|
55
|
+
value: "don't you love politics?",
|
56
|
+
});
|
57
|
+
```
|
58
|
+
|
3
59
|
## v0.122.0 - 2024-01-13
|
4
60
|
|
5
61
|
### Changed
|
package/README.md
CHANGED
@@ -22,7 +22,7 @@
|
|
22
22
|
- **Built for production**: ModelFusion is fully tree-shakeable, can be used in serverless environments, and only uses a minimal set of dependencies.
|
23
23
|
|
24
24
|
> [!NOTE]
|
25
|
-
> ModelFusion is getting closer to a stable v1, which is expected in
|
25
|
+
> ModelFusion is getting closer to a stable v1, which is expected in Q2/2024. The main API is now mostly stable, but until version 1.0 there may be breaking changes. Feedback and suggestions are welcome.
|
26
26
|
|
27
27
|
## Quick Install
|
28
28
|
|
@@ -291,6 +291,8 @@ Providers: [OpenAI (Whisper)](https://modelfusion.dev/integration/model-provider
|
|
291
291
|
Create embeddings for text and other values. Embeddings are vectors that represent the essence of the values in the context of the model.
|
292
292
|
|
293
293
|
```ts
|
294
|
+
import { embed, embedMany, openai } from "modelfusion";
|
295
|
+
|
294
296
|
// embed single value:
|
295
297
|
const embedding = await embed({
|
296
298
|
model: openai.TextEmbedder({ model: "text-embedding-ada-002" }),
|
@@ -307,7 +309,44 @@ const embeddings = await embedMany({
|
|
307
309
|
});
|
308
310
|
```
|
309
311
|
|
310
|
-
Providers: [OpenAI](https://modelfusion.dev/integration/model-provider/openai), [Llama.cpp](https://modelfusion.dev/integration/model-provider/llamacpp), [Ollama](https://modelfusion.dev/integration/model-provider/ollama), [Mistral](https://modelfusion.dev/integration/model-provider/mistral), [Hugging Face](https://modelfusion.dev/integration/model-provider/huggingface), [Cohere](https://modelfusion.dev/integration/model-provider/cohere)
|
312
|
+
Providers: [OpenAI](https://modelfusion.dev/integration/model-provider/openai), [OpenAI compatible](https://modelfusion.dev/integration/model-provider/openaicompatible), [Llama.cpp](https://modelfusion.dev/integration/model-provider/llamacpp), [Ollama](https://modelfusion.dev/integration/model-provider/ollama), [Mistral](https://modelfusion.dev/integration/model-provider/mistral), [Hugging Face](https://modelfusion.dev/integration/model-provider/huggingface), [Cohere](https://modelfusion.dev/integration/model-provider/cohere)
|
313
|
+
|
314
|
+
### [Classify Value](https://modelfusion.dev/guide/function/classify)
|
315
|
+
|
316
|
+
Classifies a value into a category.
|
317
|
+
|
318
|
+
```ts
|
319
|
+
import { classify, EmbeddingSimilarityClassifier, openai } from "modelfusion";
|
320
|
+
|
321
|
+
const classifier = new EmbeddingSimilarityClassifier({
|
322
|
+
embeddingModel: openai.TextEmbedder({ model: "text-embedding-ada-002" }),
|
323
|
+
similarityThreshold: 0.82,
|
324
|
+
clusters: [
|
325
|
+
{
|
326
|
+
name: "politics" as const,
|
327
|
+
values: [
|
328
|
+
"they will save the country!",
|
329
|
+
// ...
|
330
|
+
],
|
331
|
+
},
|
332
|
+
{
|
333
|
+
name: "chitchat" as const,
|
334
|
+
values: [
|
335
|
+
"how's the weather today?",
|
336
|
+
// ...
|
337
|
+
],
|
338
|
+
},
|
339
|
+
],
|
340
|
+
});
|
341
|
+
|
342
|
+
// strongly typed result:
|
343
|
+
const result = await classify({
|
344
|
+
model: classifier,
|
345
|
+
value: "don't you love politics?",
|
346
|
+
});
|
347
|
+
```
|
348
|
+
|
349
|
+
Classifiers: [EmbeddingSimilarityClassifier](https://modelfusion.dev/guide/function/classify#embeddingsimilarityclassifier)
|
311
350
|
|
312
351
|
### [Tokenize Text](https://modelfusion.dev/guide/function/tokenize-text)
|
313
352
|
|
@@ -552,6 +591,7 @@ modelfusion.setLogFormat("detailed-object"); // log full events
|
|
552
591
|
- [Generate transcription](https://modelfusion.dev/guide/function/generation-transcription)
|
553
592
|
- [Tokenize Text](https://modelfusion.dev/guide/function/tokenize-text)
|
554
593
|
- [Embed Value](https://modelfusion.dev/guide/function/embed)
|
594
|
+
- [Classify Value](https://modelfusion.dev/guide/function/classify)
|
555
595
|
- [Tools](https://modelfusion.dev/guide/tools)
|
556
596
|
- [Use Tool](https://modelfusion.dev/guide/tools/use-tool)
|
557
597
|
- [Use Tools](https://modelfusion.dev/guide/tools/use-tools)
|
package/index.cjs
CHANGED
@@ -14,7 +14,6 @@ var __exportStar = (this && this.__exportStar) || function(m, exports) {
|
|
14
14
|
for (var p in m) if (p !== "default" && !Object.prototype.hasOwnProperty.call(exports, p)) __createBinding(exports, m, p);
|
15
15
|
};
|
16
16
|
Object.defineProperty(exports, "__esModule", { value: true });
|
17
|
-
__exportStar(require("./classifier/index.cjs"), exports);
|
18
17
|
__exportStar(require("./core/index.cjs"), exports);
|
19
18
|
__exportStar(require("./model-function/index.cjs"), exports);
|
20
19
|
__exportStar(require("./model-provider/index.cjs"), exports);
|
package/index.d.ts
CHANGED
package/index.js
CHANGED
@@ -2,6 +2,7 @@ import { BaseFunctionFinishedEvent, BaseFunctionStartedEvent } from "../core/Fun
|
|
2
2
|
import { ToolCallGenerationFinishedEvent, ToolCallGenerationStartedEvent } from "../tool/generate-tool-call/ToolCallGenerationEvent.js";
|
3
3
|
import { ToolCallsGenerationFinishedEvent, ToolCallsGenerationStartedEvent } from "../tool/generate-tool-calls/ToolCallsGenerationEvent.js";
|
4
4
|
import { ModelInformation } from "./ModelInformation.js";
|
5
|
+
import { ClassifyFinishedEvent, ClassifyStartedEvent } from "./classify/ClassifyEvent.js";
|
5
6
|
import { EmbeddingFinishedEvent, EmbeddingStartedEvent } from "./embed/EmbeddingEvent.js";
|
6
7
|
import { ImageGenerationFinishedEvent, ImageGenerationStartedEvent } from "./generate-image/ImageGenerationEvent.js";
|
7
8
|
import { SpeechGenerationFinishedEvent, SpeechGenerationStartedEvent, SpeechStreamingFinishedEvent, SpeechStreamingStartedEvent } from "./generate-speech/SpeechGenerationEvent.js";
|
@@ -52,5 +53,5 @@ export interface BaseModelCallFinishedEvent extends BaseFunctionFinishedEvent {
|
|
52
53
|
*/
|
53
54
|
result: BaseModelCallFinishedEventResult;
|
54
55
|
}
|
55
|
-
export type ModelCallStartedEvent = EmbeddingStartedEvent | ImageGenerationStartedEvent | SpeechGenerationStartedEvent | SpeechStreamingStartedEvent | StructureGenerationStartedEvent | StructureStreamingStartedEvent | TextGenerationStartedEvent | TextStreamingStartedEvent | ToolCallGenerationStartedEvent | ToolCallsGenerationStartedEvent | TranscriptionStartedEvent;
|
56
|
-
export type ModelCallFinishedEvent = EmbeddingFinishedEvent | ImageGenerationFinishedEvent | SpeechGenerationFinishedEvent | SpeechStreamingFinishedEvent | StructureGenerationFinishedEvent | StructureStreamingFinishedEvent | TextGenerationFinishedEvent | TextStreamingFinishedEvent | ToolCallGenerationFinishedEvent | ToolCallsGenerationFinishedEvent | TranscriptionFinishedEvent;
|
56
|
+
export type ModelCallStartedEvent = ClassifyStartedEvent | EmbeddingStartedEvent | ImageGenerationStartedEvent | SpeechGenerationStartedEvent | SpeechStreamingStartedEvent | StructureGenerationStartedEvent | StructureStreamingStartedEvent | TextGenerationStartedEvent | TextStreamingStartedEvent | ToolCallGenerationStartedEvent | ToolCallsGenerationStartedEvent | TranscriptionStartedEvent;
|
57
|
+
export type ModelCallFinishedEvent = ClassifyFinishedEvent | EmbeddingFinishedEvent | ImageGenerationFinishedEvent | SpeechGenerationFinishedEvent | SpeechStreamingFinishedEvent | StructureGenerationFinishedEvent | StructureStreamingFinishedEvent | TextGenerationFinishedEvent | TextStreamingFinishedEvent | ToolCallGenerationFinishedEvent | ToolCallsGenerationFinishedEvent | TranscriptionFinishedEvent;
|
@@ -0,0 +1,10 @@
|
|
1
|
+
import { FunctionCallOptions } from "../../core/FunctionOptions.js";
|
2
|
+
import { Model, ModelSettings } from "../Model.js";
|
3
|
+
export interface ClassifierSettings extends ModelSettings {
|
4
|
+
}
|
5
|
+
export interface Classifier<VALUE, CLASS extends string | null, SETTINGS extends ClassifierSettings = ClassifierSettings> extends Model<SETTINGS> {
|
6
|
+
doClassify(value: VALUE, options: FunctionCallOptions): PromiseLike<{
|
7
|
+
rawResponse: unknown | undefined;
|
8
|
+
class: CLASS;
|
9
|
+
}>;
|
10
|
+
}
|
@@ -0,0 +1 @@
|
|
1
|
+
export {};
|
@@ -0,0 +1,20 @@
|
|
1
|
+
import { BaseModelCallFinishedEvent, BaseModelCallStartedEvent } from "../ModelCallEvent.js";
|
2
|
+
export interface ClassifyStartedEvent extends BaseModelCallStartedEvent {
|
3
|
+
functionType: "classify";
|
4
|
+
input: unknown | Array<unknown>;
|
5
|
+
}
|
6
|
+
export type ClassifyFinishedEventResult = {
|
7
|
+
status: "success";
|
8
|
+
rawResponse: unknown;
|
9
|
+
value: unknown;
|
10
|
+
} | {
|
11
|
+
status: "error";
|
12
|
+
error: unknown;
|
13
|
+
} | {
|
14
|
+
status: "abort";
|
15
|
+
};
|
16
|
+
export interface ClassifyFinishedEvent extends BaseModelCallFinishedEvent {
|
17
|
+
functionType: "classify";
|
18
|
+
input: unknown;
|
19
|
+
result: ClassifyFinishedEventResult;
|
20
|
+
}
|
@@ -0,0 +1 @@
|
|
1
|
+
export {};
|
@@ -0,0 +1,97 @@
|
|
1
|
+
"use strict";
|
2
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
3
|
+
exports.EmbeddingSimilarityClassifier = void 0;
|
4
|
+
const cosineSimilarity_js_1 = require("../../util/cosineSimilarity.cjs");
|
5
|
+
const embed_js_1 = require("../embed/embed.cjs");
|
6
|
+
/**
|
7
|
+
* Classifies values based on their distance to the values from a set of clusters.
|
8
|
+
* When the distance is below a certain threshold, the value is classified as belonging to the cluster,
|
9
|
+
* and the cluster name is returned. Otherwise, the value is classified as null.
|
10
|
+
*/
|
11
|
+
class EmbeddingSimilarityClassifier {
|
12
|
+
constructor(settings) {
|
13
|
+
Object.defineProperty(this, "settings", {
|
14
|
+
enumerable: true,
|
15
|
+
configurable: true,
|
16
|
+
writable: true,
|
17
|
+
value: void 0
|
18
|
+
});
|
19
|
+
Object.defineProperty(this, "modelInformation", {
|
20
|
+
enumerable: true,
|
21
|
+
configurable: true,
|
22
|
+
writable: true,
|
23
|
+
value: {
|
24
|
+
provider: "modelfusion",
|
25
|
+
modelName: "EmbeddingSimilarityClassifier",
|
26
|
+
}
|
27
|
+
});
|
28
|
+
Object.defineProperty(this, "embeddings", {
|
29
|
+
enumerable: true,
|
30
|
+
configurable: true,
|
31
|
+
writable: true,
|
32
|
+
value: void 0
|
33
|
+
});
|
34
|
+
this.settings = settings;
|
35
|
+
}
|
36
|
+
async getEmbeddings(options) {
|
37
|
+
if (this.embeddings != null) {
|
38
|
+
return this.embeddings;
|
39
|
+
}
|
40
|
+
const embeddings = [];
|
41
|
+
for (const cluster of this.settings.clusters) {
|
42
|
+
const clusterEmbeddings = await (0, embed_js_1.embedMany)({
|
43
|
+
model: this.settings.embeddingModel,
|
44
|
+
values: cluster.values,
|
45
|
+
...options,
|
46
|
+
});
|
47
|
+
for (let i = 0; i < clusterEmbeddings.length; i++) {
|
48
|
+
embeddings.push({
|
49
|
+
embedding: clusterEmbeddings[i],
|
50
|
+
clusterValue: cluster.values[i],
|
51
|
+
clusterName: cluster.name,
|
52
|
+
});
|
53
|
+
}
|
54
|
+
}
|
55
|
+
this.embeddings = embeddings; // lazy caching
|
56
|
+
return embeddings;
|
57
|
+
}
|
58
|
+
async doClassify(value, options) {
|
59
|
+
const valueEmbedding = await (0, embed_js_1.embed)({
|
60
|
+
model: this.settings.embeddingModel,
|
61
|
+
value,
|
62
|
+
...options,
|
63
|
+
});
|
64
|
+
const clusterEmbeddings = await this.getEmbeddings(options);
|
65
|
+
const allMatches = [];
|
66
|
+
for (const embedding of clusterEmbeddings) {
|
67
|
+
const similarity = (0, cosineSimilarity_js_1.cosineSimilarity)(valueEmbedding, embedding.embedding);
|
68
|
+
if (similarity >= this.settings.similarityThreshold) {
|
69
|
+
allMatches.push({
|
70
|
+
similarity,
|
71
|
+
clusterValue: embedding.clusterValue,
|
72
|
+
clusterName: embedding.clusterName,
|
73
|
+
});
|
74
|
+
}
|
75
|
+
}
|
76
|
+
// sort (highest similarity first)
|
77
|
+
allMatches.sort((a, b) => b.similarity - a.similarity);
|
78
|
+
return {
|
79
|
+
class: allMatches.length > 0
|
80
|
+
? allMatches[0].clusterName
|
81
|
+
: null,
|
82
|
+
rawResponse: undefined,
|
83
|
+
};
|
84
|
+
}
|
85
|
+
get settingsForEvent() {
|
86
|
+
const eventSettingProperties = [
|
87
|
+
"clusters",
|
88
|
+
"embeddingModel",
|
89
|
+
"similarityThreshold",
|
90
|
+
];
|
91
|
+
return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
|
92
|
+
}
|
93
|
+
withSettings(additionalSettings) {
|
94
|
+
return new EmbeddingSimilarityClassifier(Object.assign({}, this.settings, additionalSettings));
|
95
|
+
}
|
96
|
+
}
|
97
|
+
exports.EmbeddingSimilarityClassifier = EmbeddingSimilarityClassifier;
|
@@ -0,0 +1,40 @@
|
|
1
|
+
import { FunctionCallOptions } from "../../core/FunctionOptions.js";
|
2
|
+
import { Vector } from "../../core/Vector.js";
|
3
|
+
import { EmbeddingModel } from "../embed/EmbeddingModel.js";
|
4
|
+
import { Classifier, ClassifierSettings } from "./Classifier.js";
|
5
|
+
export interface ValueCluster<VALUE, NAME extends string> {
|
6
|
+
name: NAME;
|
7
|
+
values: VALUE[];
|
8
|
+
}
|
9
|
+
export interface EmbeddingSimilarityClassifierSettings<VALUE, CLUSTERS extends Array<ValueCluster<VALUE, string>>> extends ClassifierSettings {
|
10
|
+
clusters: CLUSTERS;
|
11
|
+
embeddingModel: EmbeddingModel<VALUE>;
|
12
|
+
similarityThreshold: number;
|
13
|
+
}
|
14
|
+
/**
|
15
|
+
* Classifies values based on their distance to the values from a set of clusters.
|
16
|
+
* When the distance is below a certain threshold, the value is classified as belonging to the cluster,
|
17
|
+
* and the cluster name is returned. Otherwise, the value is classified as null.
|
18
|
+
*/
|
19
|
+
export declare class EmbeddingSimilarityClassifier<VALUE, CLUSTERS extends Array<ValueCluster<VALUE, string>>> implements Classifier<VALUE, ClusterNames<CLUSTERS> | null, EmbeddingSimilarityClassifierSettings<VALUE, CLUSTERS>> {
|
20
|
+
readonly settings: EmbeddingSimilarityClassifierSettings<VALUE, CLUSTERS>;
|
21
|
+
readonly modelInformation: {
|
22
|
+
provider: string;
|
23
|
+
modelName: string;
|
24
|
+
};
|
25
|
+
private embeddings;
|
26
|
+
constructor(settings: EmbeddingSimilarityClassifierSettings<VALUE, CLUSTERS>);
|
27
|
+
getEmbeddings(options: FunctionCallOptions): Promise<{
|
28
|
+
embedding: Vector;
|
29
|
+
clusterValue: VALUE;
|
30
|
+
clusterName: string;
|
31
|
+
}[]>;
|
32
|
+
doClassify(value: VALUE, options: FunctionCallOptions): Promise<{
|
33
|
+
class: ClusterNames<CLUSTERS> | null;
|
34
|
+
rawResponse: undefined;
|
35
|
+
}>;
|
36
|
+
get settingsForEvent(): Partial<EmbeddingSimilarityClassifierSettings<VALUE, CLUSTERS>>;
|
37
|
+
withSettings(additionalSettings: Partial<EmbeddingSimilarityClassifierSettings<VALUE, CLUSTERS>>): this;
|
38
|
+
}
|
39
|
+
type ClusterNames<CLUSTERS> = CLUSTERS extends Array<ValueCluster<unknown, infer NAME>> ? NAME : never;
|
40
|
+
export {};
|
@@ -0,0 +1,93 @@
|
|
1
|
+
import { cosineSimilarity } from "../../util/cosineSimilarity.js";
|
2
|
+
import { embed, embedMany } from "../embed/embed.js";
|
3
|
+
/**
|
4
|
+
* Classifies values based on their distance to the values from a set of clusters.
|
5
|
+
* When the distance is below a certain threshold, the value is classified as belonging to the cluster,
|
6
|
+
* and the cluster name is returned. Otherwise, the value is classified as null.
|
7
|
+
*/
|
8
|
+
export class EmbeddingSimilarityClassifier {
|
9
|
+
constructor(settings) {
|
10
|
+
Object.defineProperty(this, "settings", {
|
11
|
+
enumerable: true,
|
12
|
+
configurable: true,
|
13
|
+
writable: true,
|
14
|
+
value: void 0
|
15
|
+
});
|
16
|
+
Object.defineProperty(this, "modelInformation", {
|
17
|
+
enumerable: true,
|
18
|
+
configurable: true,
|
19
|
+
writable: true,
|
20
|
+
value: {
|
21
|
+
provider: "modelfusion",
|
22
|
+
modelName: "EmbeddingSimilarityClassifier",
|
23
|
+
}
|
24
|
+
});
|
25
|
+
Object.defineProperty(this, "embeddings", {
|
26
|
+
enumerable: true,
|
27
|
+
configurable: true,
|
28
|
+
writable: true,
|
29
|
+
value: void 0
|
30
|
+
});
|
31
|
+
this.settings = settings;
|
32
|
+
}
|
33
|
+
async getEmbeddings(options) {
|
34
|
+
if (this.embeddings != null) {
|
35
|
+
return this.embeddings;
|
36
|
+
}
|
37
|
+
const embeddings = [];
|
38
|
+
for (const cluster of this.settings.clusters) {
|
39
|
+
const clusterEmbeddings = await embedMany({
|
40
|
+
model: this.settings.embeddingModel,
|
41
|
+
values: cluster.values,
|
42
|
+
...options,
|
43
|
+
});
|
44
|
+
for (let i = 0; i < clusterEmbeddings.length; i++) {
|
45
|
+
embeddings.push({
|
46
|
+
embedding: clusterEmbeddings[i],
|
47
|
+
clusterValue: cluster.values[i],
|
48
|
+
clusterName: cluster.name,
|
49
|
+
});
|
50
|
+
}
|
51
|
+
}
|
52
|
+
this.embeddings = embeddings; // lazy caching
|
53
|
+
return embeddings;
|
54
|
+
}
|
55
|
+
async doClassify(value, options) {
|
56
|
+
const valueEmbedding = await embed({
|
57
|
+
model: this.settings.embeddingModel,
|
58
|
+
value,
|
59
|
+
...options,
|
60
|
+
});
|
61
|
+
const clusterEmbeddings = await this.getEmbeddings(options);
|
62
|
+
const allMatches = [];
|
63
|
+
for (const embedding of clusterEmbeddings) {
|
64
|
+
const similarity = cosineSimilarity(valueEmbedding, embedding.embedding);
|
65
|
+
if (similarity >= this.settings.similarityThreshold) {
|
66
|
+
allMatches.push({
|
67
|
+
similarity,
|
68
|
+
clusterValue: embedding.clusterValue,
|
69
|
+
clusterName: embedding.clusterName,
|
70
|
+
});
|
71
|
+
}
|
72
|
+
}
|
73
|
+
// sort (highest similarity first)
|
74
|
+
allMatches.sort((a, b) => b.similarity - a.similarity);
|
75
|
+
return {
|
76
|
+
class: allMatches.length > 0
|
77
|
+
? allMatches[0].clusterName
|
78
|
+
: null,
|
79
|
+
rawResponse: undefined,
|
80
|
+
};
|
81
|
+
}
|
82
|
+
get settingsForEvent() {
|
83
|
+
const eventSettingProperties = [
|
84
|
+
"clusters",
|
85
|
+
"embeddingModel",
|
86
|
+
"similarityThreshold",
|
87
|
+
];
|
88
|
+
return Object.fromEntries(Object.entries(this.settings).filter(([key]) => eventSettingProperties.includes(key)));
|
89
|
+
}
|
90
|
+
withSettings(additionalSettings) {
|
91
|
+
return new EmbeddingSimilarityClassifier(Object.assign({}, this.settings, additionalSettings));
|
92
|
+
}
|
93
|
+
}
|
@@ -0,0 +1,27 @@
|
|
1
|
+
"use strict";
|
2
|
+
Object.defineProperty(exports, "__esModule", { value: true });
|
3
|
+
exports.classify = void 0;
|
4
|
+
const executeStandardCall_js_1 = require("../executeStandardCall.cjs");
|
5
|
+
async function classify({ model, value, fullResponse, ...options }) {
|
6
|
+
const callResponse = await (0, executeStandardCall_js_1.executeStandardCall)({
|
7
|
+
functionType: "classify",
|
8
|
+
input: value,
|
9
|
+
model,
|
10
|
+
options,
|
11
|
+
generateResponse: async (options) => {
|
12
|
+
const result = await model.doClassify(value, options);
|
13
|
+
return {
|
14
|
+
rawResponse: result.rawResponse,
|
15
|
+
extractedValue: result.class,
|
16
|
+
};
|
17
|
+
},
|
18
|
+
});
|
19
|
+
return fullResponse
|
20
|
+
? {
|
21
|
+
class: callResponse.value,
|
22
|
+
rawResponse: callResponse.rawResponse,
|
23
|
+
metadata: callResponse.metadata,
|
24
|
+
}
|
25
|
+
: callResponse.value;
|
26
|
+
}
|
27
|
+
exports.classify = classify;
|
@@ -0,0 +1,17 @@
|
|
1
|
+
import { FunctionOptions } from "../../core/FunctionOptions.js";
|
2
|
+
import { ModelCallMetadata } from "../ModelCallMetadata.js";
|
3
|
+
import { Classifier, ClassifierSettings } from "./Classifier.js";
|
4
|
+
export declare function classify<VALUE, CLASS extends string | null>(args: {
|
5
|
+
model: Classifier<VALUE, CLASS, ClassifierSettings>;
|
6
|
+
value: VALUE;
|
7
|
+
fullResponse?: false;
|
8
|
+
} & FunctionOptions): Promise<CLASS>;
|
9
|
+
export declare function classify<VALUE, CLASS extends string | null>(args: {
|
10
|
+
model: Classifier<VALUE, CLASS, ClassifierSettings>;
|
11
|
+
value: VALUE;
|
12
|
+
fullResponse: true;
|
13
|
+
} & FunctionOptions): Promise<{
|
14
|
+
class: CLASS;
|
15
|
+
rawResponse: unknown;
|
16
|
+
metadata: ModelCallMetadata;
|
17
|
+
}>;
|
@@ -0,0 +1,23 @@
|
|
1
|
+
import { executeStandardCall } from "../executeStandardCall.js";
|
2
|
+
export async function classify({ model, value, fullResponse, ...options }) {
|
3
|
+
const callResponse = await executeStandardCall({
|
4
|
+
functionType: "classify",
|
5
|
+
input: value,
|
6
|
+
model,
|
7
|
+
options,
|
8
|
+
generateResponse: async (options) => {
|
9
|
+
const result = await model.doClassify(value, options);
|
10
|
+
return {
|
11
|
+
rawResponse: result.rawResponse,
|
12
|
+
extractedValue: result.class,
|
13
|
+
};
|
14
|
+
},
|
15
|
+
});
|
16
|
+
return fullResponse
|
17
|
+
? {
|
18
|
+
class: callResponse.value,
|
19
|
+
rawResponse: callResponse.rawResponse,
|
20
|
+
metadata: callResponse.metadata,
|
21
|
+
}
|
22
|
+
: callResponse.value;
|
23
|
+
}
|
@@ -14,4 +14,7 @@ var __exportStar = (this && this.__exportStar) || function(m, exports) {
|
|
14
14
|
for (var p in m) if (p !== "default" && !Object.prototype.hasOwnProperty.call(exports, p)) __createBinding(exports, m, p);
|
15
15
|
};
|
16
16
|
Object.defineProperty(exports, "__esModule", { value: true });
|
17
|
-
__exportStar(require("./
|
17
|
+
__exportStar(require("./Classifier.cjs"), exports);
|
18
|
+
__exportStar(require("./ClassifyEvent.cjs"), exports);
|
19
|
+
__exportStar(require("./EmbeddingSimilarityClassifier.cjs"), exports);
|
20
|
+
__exportStar(require("./classify.cjs"), exports);
|
package/model-function/index.cjs
CHANGED
@@ -20,6 +20,7 @@ __exportStar(require("./ModelCallEvent.cjs"), exports);
|
|
20
20
|
__exportStar(require("./ModelCallMetadata.cjs"), exports);
|
21
21
|
__exportStar(require("./ModelInformation.cjs"), exports);
|
22
22
|
__exportStar(require("./PromptTemplate.cjs"), exports);
|
23
|
+
__exportStar(require("./classify/index.cjs"), exports);
|
23
24
|
__exportStar(require("./embed/EmbeddingEvent.cjs"), exports);
|
24
25
|
__exportStar(require("./embed/EmbeddingModel.cjs"), exports);
|
25
26
|
__exportStar(require("./embed/embed.cjs"), exports);
|
@@ -4,6 +4,7 @@ export * from "./ModelCallEvent.js";
|
|
4
4
|
export * from "./ModelCallMetadata.js";
|
5
5
|
export * from "./ModelInformation.js";
|
6
6
|
export * from "./PromptTemplate.js";
|
7
|
+
export * from "./classify/index.js";
|
7
8
|
export * from "./embed/EmbeddingEvent.js";
|
8
9
|
export * from "./embed/EmbeddingModel.js";
|
9
10
|
export * from "./embed/embed.js";
|
package/model-function/index.js
CHANGED
@@ -4,6 +4,7 @@ export * from "./ModelCallEvent.js";
|
|
4
4
|
export * from "./ModelCallMetadata.js";
|
5
5
|
export * from "./ModelInformation.js";
|
6
6
|
export * from "./PromptTemplate.js";
|
7
|
+
export * from "./classify/index.js";
|
7
8
|
export * from "./embed/EmbeddingEvent.js";
|
8
9
|
export * from "./embed/EmbeddingModel.js";
|
9
10
|
export * from "./embed/embed.js";
|
@@ -32,16 +32,16 @@ export declare class MistralTextEmbeddingModel extends AbstractModel<MistralText
|
|
32
32
|
doEmbedValues(texts: string[], options: FunctionCallOptions): Promise<{
|
33
33
|
rawResponse: {
|
34
34
|
object: string;
|
35
|
-
model: string;
|
36
|
-
usage: {
|
37
|
-
prompt_tokens: number;
|
38
|
-
total_tokens: number;
|
39
|
-
};
|
40
35
|
data: {
|
41
36
|
object: string;
|
42
37
|
embedding: number[];
|
43
38
|
index: number;
|
44
39
|
}[];
|
40
|
+
model: string;
|
41
|
+
usage: {
|
42
|
+
prompt_tokens: number;
|
43
|
+
total_tokens: number;
|
44
|
+
};
|
45
45
|
id: string;
|
46
46
|
};
|
47
47
|
embeddings: number[][];
|
@@ -77,29 +77,29 @@ declare const MistralTextEmbeddingResponseSchema: z.ZodObject<{
|
|
77
77
|
}>;
|
78
78
|
}, "strip", z.ZodTypeAny, {
|
79
79
|
object: string;
|
80
|
-
model: string;
|
81
|
-
usage: {
|
82
|
-
prompt_tokens: number;
|
83
|
-
total_tokens: number;
|
84
|
-
};
|
85
80
|
data: {
|
86
81
|
object: string;
|
87
82
|
embedding: number[];
|
88
83
|
index: number;
|
89
84
|
}[];
|
90
|
-
id: string;
|
91
|
-
}, {
|
92
|
-
object: string;
|
93
85
|
model: string;
|
94
86
|
usage: {
|
95
87
|
prompt_tokens: number;
|
96
88
|
total_tokens: number;
|
97
89
|
};
|
90
|
+
id: string;
|
91
|
+
}, {
|
92
|
+
object: string;
|
98
93
|
data: {
|
99
94
|
object: string;
|
100
95
|
embedding: number[];
|
101
96
|
index: number;
|
102
97
|
}[];
|
98
|
+
model: string;
|
99
|
+
usage: {
|
100
|
+
prompt_tokens: number;
|
101
|
+
total_tokens: number;
|
102
|
+
};
|
103
103
|
id: string;
|
104
104
|
}>;
|
105
105
|
export type MistralTextEmbeddingResponse = z.infer<typeof MistralTextEmbeddingResponseSchema>;
|