Replace toxicity model with multi-label classifier

This commit is contained in:
Jordan Wages 2026-08-26 17:00:49 -05:00
commit 1568c787f8
24 changed files with 2581129 additions and 119717 deletions

View file

@ -3,25 +3,26 @@ import { validateManifest } from "../src/inference/model-metadata";
import { resolveOutputIndex } from "../src/inference/classifier";
const manifest = {
source: "wagesj45/toxic-comment-classifier",
source: "wagesj45/multilabel-toxic-comment-classifier",
revision: "abc123",
architecture: "DistilBertForSequenceClassification",
labels: { toxic: 1, nonToxic: 0, names: { toxic: 1, non_toxic: 0 } },
architecture: "ModernBertForSequenceClassification",
labels: { names: { toxicity: 0, severe_toxicity: 1, obscene: 2, threat: 3, insult: 4, identity_attack: 5, sexual_explicit: 6 } },
scoring: "multi-label-sigmoid" as const,
maxLength: 512,
quantization: "int8-dynamic",
runtime: "onnxruntime-web-wasm"
};
describe("model contract", () => {
it("accepts a binary manifest and resolves common output labels", () => {
it("accepts a multi-label manifest and resolves output labels", () => {
expect(validateManifest(manifest)).toEqual(manifest);
expect(resolveOutputIndex("LABEL_1", manifest)).toBe(1);
expect(resolveOutputIndex("not-toxic", manifest)).toBe(0);
expect(resolveOutputIndex("toxic", manifest)).toBe(1);
expect(resolveOutputIndex("identity-attack", manifest)).toBe(5);
expect(resolveOutputIndex("toxicity", manifest)).toBe(0);
});
it("rejects incomplete metadata", () => {
expect(() => validateManifest({ ...manifest, labels: { toxic: 1, nonToxic: 1, names: {} } })).toThrow();
expect(() => validateManifest({ ...manifest, labels: { names: { ...manifest.labels.names, threat: 1 } } })).toThrow();
expect(() => validateManifest({ ...manifest, maxLength: 0 })).toThrow();
expect(() => validateManifest({ ...manifest, labels: undefined as unknown as typeof manifest.labels })).toThrow();
});