run-llama/llama_index · error · ValueError

The embedding file {file_path} is empty.

Error message

The embedding file {file_path} is empty.

What it means

load_embedding(file_path) reads a CSV-of-floats file produced by save_embedding() and returns the first line. If the file exists but contains zero lines (empty file), the for-loop body never executes and the function raises ValueError — the file was created but no embedding was ever written to it.

Source

Thrown at llama-index-core/llama_index/core/embeddings/utils.py:27

from llama_index.core.callbacks import CallbackManager
from llama_index.core.embeddings.mock_embed_model import MockEmbedding
from llama_index.core.utils import get_cache_dir

EmbedType = Union[BaseEmbedding, "LCEmbeddings", str]


def save_embedding(embedding: List[float], file_path: str) -> None:
    """Save embedding to file."""
    with open(file_path, "w", encoding="utf-8") as f:
        f.write(",".join([str(x) for x in embedding]))


def load_embedding(file_path: str) -> List[float]:
    """Load embedding from file. Will only return first embedding in file."""
    with open(file_path, encoding="utf-8") as f:
        for line in f:
            return [float(x) for x in line.strip().split(",")]
    raise ValueError(f"The embedding file {file_path} is empty.")


def resolve_embed_model(
    embed_model: Optional[EmbedType] = None,
    callback_manager: Optional[CallbackManager] = None,
) -> BaseEmbedding:
    """Resolve embed model."""
    from llama_index.core.settings import Settings

    try:
        from llama_index.core.bridge.langchain import Embeddings as LCEmbeddings
    except ImportError:
        LCEmbeddings = None  # type: ignore

    if embed_model == "default":
        if os.getenv("IS_TESTING"):
            embed_model = MockEmbedding(embed_dim=8)
            embed_model.callback_manager = callback_manager or Settings.callback_manager

View on GitHub (pinned to afd0fef371)

Solutions

1. Check size before loading: if os.path.getsize(path) == 0: regenerate the embedding and save_embedding(vec, path)
2. Re-run the code that produces the embedding (save_embedding) so the file contains one comma-separated line
3. Treat empty cache entries as cache misses and recompute, then overwrite the file

Example fix

// before
vec = load_embedding("emb.txt")  # file is empty -> ValueError

// after
import os
if os.path.getsize("emb.txt") == 0:
    vec = embed_model.get_text_embedding(text)
    save_embedding(vec, "emb.txt")
else:
    vec = load_embedding("emb.txt")

import os
if not os.path.exists(path) or os.path.getsize(path) == 0:
    vec = embed_model.get_text_embedding(text)
    save_embedding(vec, path)
vec = load_embedding(path)

try:
    vec = load_embedding(path)
except ValueError as e:
    if "is empty" in str(e):
        vec = embed_model.get_text_embedding(text)
        save_embedding(vec, path)
    else:
        raise

When it happens

Trigger: Calling load_embedding on a file that is 0 bytes (created by open(..., 'w') without a write, an interrupted save_embedding, or touch); pointing at a truncated/corrupted cache file.

Common situations: A crashed or killed embedding-save job left empty files; filesystem issues truncating cache files; code that pre-creates placeholder files then fails to populate them.

Related errors

• embeddings_cache must be of type BaseKVStore
• embeddings_cache must be defined
• No embeddings to aggregate
• Embedding loading requires a class_name
• Invalid Embedding name: {name}

AI-assisted analysis of run-llama/llama_index@afd0fef371 (2026-08-15). Data as JSON: /api/errors/7a14bdff1436fcc3. Report an issue: GitHub.