mem0ai/mem0 · error · ValueError

Vertex AI embed_batch() returned {len(all_embeddings)} embed

Error message

Vertex AI embed_batch() returned {len(all_embeddings)} embeddings for {len(texts)} texts using model '{self.config.model}'

What it means

Raised by VertexAIEmbedding.embed_batch after chunking texts into groups of 250 and collecting all vectors: total embeddings must equal total texts. A mismatch means the Vertex AI text-embedding endpoint returned a different number of values for at least one chunk — usually per-request input limits or output_dimensionality interactions with older model versions.

Source

Thrown at mem0/embeddings/vertexai.py:81

        return embeddings[0].values

    def embed_batch(self, texts, memory_action="add"):
        if not texts:
            return []
        embedding_type = "SEMANTIC_SIMILARITY"
        if memory_action is not None:
            if memory_action not in self.embedding_types:
                raise ValueError(f"Invalid memory action: {memory_action}")
            embedding_type = self.embedding_types[memory_action]
        all_embeddings = []
        for i in range(0, len(texts), 250):
            chunk = texts[i : i + 250]
            inputs = [TextEmbeddingInput(text=t, task_type=embedding_type) for t in chunk]
            results = self.model.get_embeddings(texts=inputs, output_dimensionality=self.config.embedding_dims)
            all_embeddings.extend(r.values for r in results)
        if len(all_embeddings) != len(texts):
            raise ValueError(
                f"Vertex AI embed_batch() returned {len(all_embeddings)} embeddings for {len(texts)} texts"
                f" using model '{self.config.model}'"
            )
        return all_embeddings

View on GitHub (pinned to 001c235229)

Solutions

Lower the client-side batch size below the model's per-request input limit (e.g. chunks of 32-64 texts)
Use a current model (text-embedding-004 or newer) whose limits match the 250 chunking
Retry the failing chunk and log per-chunk counts to isolate the request that loses items
Ensure output_dimensionality is supported by the chosen model

Example fix

// before
embs = embedder.embed_batch(texts)  # e.g. 10000 texts

# after
embs = []
for i in range(0, len(texts), 64):
    embs.extend(embedder.embed_batch(texts[i:i+64]))

Defensive patterns

Strategy: retry

Validate before calling

# stay under Vertex per-request input limits; 64 is safe across model versions
for i in range(0, len(texts), 64):
    embedder.embed_batch(texts[i:i+64])

Try / catch

try:
    vecs = embedder.embed_batch(chunk)
except ValueError as e:
    if "embed_batch() returned" in str(e):
        half = max(1, len(chunk)//2)
        vecs = embedder.embed_batch(chunk[:half]) + embedder.embed_batch(chunk[half:])
    else:
        raise

Prevention

Chunk below 250 (the provider's internal chunk size), ideally 64
Prefer text-embedding-004+ with known limits
Log per-chunk counts during bulk ingestion

When it happens

Trigger: Batching more texts than a chunk's API input limit for the model (text-embedding-004 caps inputs per request); using an older model (textembedding-gecko) whose batch limits differ from 250; API responses for a chunk returning fewer values for the requested output_dimensionality.

Common situations: Bulk memory ingestion on Vertex AI; switching model versions without adjusting chunk size; regional endpoint differences in limits.

Related errors

AI-assisted analysis of mem0ai/mem0@001c235229 (2026-08-15). Data as JSON: /api/errors/6e0e2dcf4825883e. Report an issue: GitHub.