What causes "vectorOutput is undefined" in redis/node-redis?

Wrap the pipe() call in try/catch to surface the real upstream error (transformers.js hides it behind the null return).

redis/node-redis · error · Error

vectorOutput is undefined

Error message

vectorOutput is undefined

What it means

Thrown by an explicit guard in the doctest's embedText() helper. The @xenova/transformers pipeline('feature-extraction', ...) call resolved to null/undefined for the input sentence, so vectorOutput.data cannot be read. transformers.js returns null when model loading, inference, or ONNX runtime setup fails internally and swallows the underlying cause.

Source

Thrown at doctests/query-combined.js:25

import { pipeline } from '@xenova/transformers';

function float32Buffer(arr) {
  const floatArray = new Float32Array(arr);
  const float32Buffer = Buffer.from(floatArray.buffer);
  return float32Buffer;
}

async function embedText(sentence) {
  let modelName = 'Xenova/all-MiniLM-L6-v2';
  let pipe = await pipeline('feature-extraction', modelName);

  let vectorOutput = await pipe(sentence, {
      pooling: 'mean',
      normalize: true,
  });

  if (vectorOutput == null) {
    throw new Error('vectorOutput is undefined');
  }

  const embedding = Object.values(vectorOutput.data);

  return embedding;
}

let vector_query = float32Buffer(await embedText('That is a very happy person'));

const client = createClient();
await client.connect().catch(console.error);

// create index
await client.ft.create('idx:bicycle', {
    '$.description': {
      type: SCHEMA_FIELD_TYPE.TEXT,
      AS: 'description'
    },

View on GitHub (pinned to 90fd0652bc)

Solutions

Wrap the pipe() call in try/catch to surface the real upstream error (transformers.js hides it behind the null return).
Verify network access to huggingface.co or pre-download the model into a local cache directory and point env.cacheDir at it.
Pin @xenova/transformers to a version known to work with your Node runtime and the all-MiniLM-L6-v2 model.
Confirm the model id is correct and reachable: test with transformers.js standalone before invoking the doctest.

Example fix

// before
let pipe = await pipeline('feature-extraction', modelName);
let vectorOutput = await pipe(sentence, { pooling: 'mean', normalize: true });
if (vectorOutput == null) throw new Error('vectorOutput is undefined');

// after
const pipe = await pipeline('feature-extraction', modelName);
let vectorOutput;
try {
  vectorOutput = await pipe(sentence, { pooling: 'mean', normalize: true });
} catch (err) {
  throw new Error('feature-extraction inference failed: ' + err.message, { cause: err });
}
if (vectorOutput == null || !('data' in vectorOutput)) {
  throw new Error('vectorOutput is undefined (model load or inference returned nothing)');
}

Defensive patterns

Strategy: try-catch

Validate before calling

import { pipeline, env } from '@xenova/transformers';

async function safePipeline(name) {
  try {
    const pipe = await pipeline('feature-extraction', name);
    if (typeof pipe !== 'function') {
      throw new Error('pipeline did not return a callable');
    }
    return pipe;
  } catch (err) {
    throw new Error(`transformers pipeline init failed for ${name}: ${err.message}`, { cause: err });
  }
}

// call before the doctest runs
const pipe = await safePipeline('Xenova/all-MiniLM-L6-v2');

Type guard

function isTensor(v) {
  return v != null && typeof v === 'object' && 'data' in v && typeof v.data.length === 'number';
}

Try / catch

try {
  vectorOutput = await pipe(sentence, { pooling: 'mean', normalize: true });
} catch (err) {
  throw new Error('feature-extraction failed: ' + err.message, { cause: err });
}
if (!isTensor(vectorOutput)) throw new Error('vectorOutput is undefined');

Prevention

Pre-warm and cache the pipeline once at process startup; reuse the instance.
Pin @xenova/transformers and verify Node/ONNX runtime compatibility.
Set env.cacheDir and pre-download the model so offline runs do not return null.
Wrap pipe() in try/catch — transformers.js swallows inference errors into null returns.

When it happens

Trigger: Running doctests/query-combined.js and the await pipe(sentence, { pooling: 'mean', normalize: true }) call resolves to null/undefined. Happens when the model artifact fails to load, ONNX backend init fails, or inference throws and is caught upstream.

Common situations: HuggingFace model download blocked (offline/corporate proxy); @xenova/transformers version mismatch with the bundled ONNX runtime; Node version incompatibility; model name typo 'Xenova/all-MiniLM-L6-v2'; first-run fetch timeout; WASM threads disabled in the runtime.

Related errors

AI-assisted analysis of redis/node-redis@90fd0652bc (2026-08-11). Data as JSON: /api/errors/6acb6ec303693a29. Report an issue: GitHub.