granite-vision-4.1-4b

by ibm-granite

Specialized VLM for extracting structured data from charts, tables, and forms

39Kdl/month

4Bparams

HuggingFace Use in Pipeline

Identifiers

Model ID

ibm-granite/granite-vision-4.1-4b

Feature URI

mixpeek://image_extractor@v1/ibm_granite_vision_41_4b_v1

Overview

Granite Vision 4.1 is IBM's purpose-built document extraction model that converts visual content — charts, tables, forms, key-value pairs — into structured machine-readable formats (CSV, JSON, HTML). Unlike general-purpose VLMs that describe what they see, Granite Vision extracts precise data values with high accuracy, making it suitable for automated document processing pipelines.

On Mixpeek, Granite Vision powers structured extraction from document pages: converting chart images to CSV data, table images to JSON records, and form images to key-value pairs. This structured output is directly indexable and filterable, unlike free-text captions.

Architecture

LoRA adapter on Granite-4.1-3B vision-language model. 4B total parameters (3.4B LLM + 0.6B vision encoder/projectors). Trained specifically on document extraction tasks: chart-to-CSV, table-to-JSON/HTML, key-value pair extraction. Integrates with IBM Docling for production pipelines.

Mixpeek SDK Integration

import { Mixpeek } from "mixpeek";

const mx = new Mixpeek({ apiKey: "API_KEY" });

await mx.collections.ingest({
  collection_id: "financial-docs",
  source: { url: "https://example.com/annual-report.pdf" },
  feature_extractors: [{
    feature: "document_extraction",
    model: "ibm-granite/granite-vision-4.1-4b"
  }]
});