mastra-ai/mastra · error · Error

No embeddings generated

Error message

No embeddings generated

What it means

generateEmbeddings() consumes the async iterable of batch embeddings from the fastembed model; if no results arrive at all (allResults.length === 0), it throws 'No embeddings generated'. This indicates the underlying model produced zero batches for the given input.

Source

Thrown at packages/fastembed/src/index.ts:44

export async function warmup() {
  await warmupFastEmbedModels();
}

// Shared function to generate embeddings using fastembed
async function generateEmbeddings(values: string[], modelType: FastEmbedModelType) {
  const model = await getCachedModel(modelType);

  // model.embed() returns an AsyncGenerator that processes texts in batches (default size 256)
  const embeddings = model.embed(values);

  const allResults = [];
  for await (const result of embeddings) {
    // result is an array of embeddings, one for each text in the batch
    // We convert each Float32Array embedding to a regular number array
    allResults.push(...result.map(embedding => Array.from(embedding)));
  }

  if (allResults.length === 0) throw new Error('No embeddings generated');

  return {
    embeddings: allResults,
  };
}

// E5 models are asymmetric: queries and passages must be embedded with different prefixes.
async function generatePrefixedEmbeddings(
  values: string[],
  modelType: FastEmbedModelType,
  prefix: 'query' | 'passage',
) {
  return generateEmbeddings(
    values.map(value => `${prefix}: ${value}`),
    modelType,
  );
}

View on GitHub (pinned to 75dd419e61)

Solutions

1. Check the input: ensure the texts/embeddings request contains at least one non-empty item before calling embed.
2. Add a guard in your calling code to skip or error early on empty input arrays.
3. If input is non-empty, verify the fastembed model initialized correctly (model files loaded) — a misloaded model can yield no batches.

Example fix

// before
await model.embed([]);
// after
const texts = getInputTexts();
if (texts.length === 0) return { embeddings: [] };
const { embeddings } = await model.embed(texts);

function assertNonEmptyInputs(texts) {
  if (!Array.isArray(texts) || texts.length === 0) {
    throw new Error('embed() requires at least one input text');
  }
}

function hasItems(x) {
  return Array.isArray(x) && x.length > 0;
}

try {
  const { embeddings } = await model.embed(texts);
  return embeddings;
} catch (e) {
  if (e.message === 'No embeddings generated') {
    return { embeddings: [] }; // or skip this batch
  }
  throw e;
}

When it happens

Trigger: Calling embed/generatePrefixedEmbeddings (via doEmbed) with an empty input array, or with a model/embedder that silently yields nothing for the batch.

Common situations: Empty texts array passed to an AI provider's embed call; upstream filtering removed all inputs; integration where a zero-length batch slips through validation at a higher layer.

Related errors

• RUN_EXPERIMENT_FAILED_NO_DATA_PROVIDED
• Knowledge scope cannot be empty
• Input and output messages cannot be null or empty
• For custom model, modelAbsoluteDirPath is required in FlagEm
• For custom model, modelName is required in FlagEmbedding.ini

AI-assisted analysis of mastra-ai/mastra@75dd419e61 (2026-08-30). Data as JSON: /api/errors/9109b129266d119f. Report an issue: GitHub.