Reason-ModernColBERT

Name: Reason-ModernColBERT
Author: lightonai

by lightonai

Late-interaction retriever trained for reasoning-intensive search queries

9.1Kdl/month

150M classparams

HuggingFace Run it on your own data, free

Identifiers

Model ID

lightonai/Reason-ModernColBERT

Feature URI

mixpeek://text_extractor@v1/lighton_reason_moderncolbert_v1

Overview

Reason-ModernColBERT is a PyLate ColBERT model fine-tuned from LightOn's GTE-ModernColBERT-v1 on the ReasonIR dataset. It targets retrieval problems where the query is not a short keyword string but a reasoning-heavy prompt that requires matching evidence across paragraphs.

On Mixpeek, this makes it a useful text retrieval companion for agents. After visual, audio, or document extractors produce text evidence, Reason-ModernColBERT can retrieve passages that match an agent's intermediate reasoning state with token-level MaxSim scoring instead of collapsing each document into one dense vector.

Architecture

ModernBERT-based late-interaction retriever trained with PyLate. It maps queries and passages to sequences of 128-dimensional token vectors and scores them with MaxSim. The model supports 8,192-token documents and 128-token queries according to the model card.

Mixpeek SDK Integration

// No extractor parameter takes a Hugging Face model id (checked against
// GET /v1/discovery/extractors, which returns 13), so Reason-ModernColBERT runs
// on your side and the output is upserted through POST
// /v1/namespaces/{namespace_id}/documents/upsert. On Enterprise the other
// path is to upload the weights instead: POST /v1/namespaces/{id}/models
// accepts the huggingface format and a custom plugin loads them.
const res = await fetch(
  "https://api.mixpeek.com/v1/namespaces/ns_your_namespace/documents/upsert",
  {
    method: "POST",
    headers: {
      Authorization: "Bearer API_KEY",
      "Content-Type": "application/json",
    },
    body: JSON.stringify({
      collection_id: "col_your_collection",
      documents: [
        {
          document_id: "asset-00412",
          // The vector name has to match a vector index on the collection.
          vectors: { "text-embedding": yourVector },
          payload: { source_key: "archive/2026/asset-00412" },
        },
      ],
    }),
  },
);

// Managed alternative, if this exact model is not the requirement:
// text_extractor@v1 runs intfloat/multilingual-e5-large-instruct
// (1024-d) over a bucket, with no inference of your own.

Capabilities

Reasoning-intensive retrieval over long passages
Late-interaction token matching with MaxSim
8K-token document support
Useful for agent queries that include context, constraints, and partial findings
Fine-tuned on ReasonIR data from GTE-ModernColBERT-v1