Home

distribatch

Batch model inference over DataFrames: LLM generation, text embeddings and speech transcription, on Databricks or anywhere Python runs.

import distribatch as dib

out = dib.GenerateJob(
    model=dib.models.databricks("databricks-meta-llama-3-3-70b-instruct"),
    prompt="Summarize in one sentence:\n{body}",
).run(spark_df, backend="spark", output_col="summary")

One job definition runs unchanged on three backends (sequential local, single-node multiprocessing, cluster-wide spark) with models from three sources (Unity Catalog, a directory or UC Volume with Hugging Face weights, any OpenAI-compatible API). Every row gets its result, an error column and metrics. A failing row never fails the job.

Where to start

I want to…	Read
install it	Installation
see it work in five minutes	Quickstart
understand jobs, engines and backends	Concepts
generate text or structured data	GenerateJob
compute embeddings	EmbedJob
transcribe audio	TranscribeJob
pick where it runs and tune it	Running Jobs, Backends
use UC models, Volumes, endpoints	Model Sources, Engines
understand the output columns and errors	Output Contract
run for hours and resume after failures	Checkpoints and Dedupe
know success rates, tokens and costs	Statistics and Costs
set it up on Databricks	Databricks Guide
fix a problem	Troubleshooting
look up a class or function	API Reference

distribatch is pre-1.0: the API may still change between minor versions. See the changelog in the repository.