Skip to content
Other

LumenSearch Reranker v2

A New Cross-Encoder for Search Systems

2 min read

The LumenSearch Reranker v2 model is the second-generation cross-encoder (a model for evaluating query-document pair relevance) created for the LumenSearch search stack. It recalculates the scores of top candidates retrieved during the initial search phase, helping to rank results more accurately. Built on the lumen-ai/lumen-encoder-base-110m encoder architecture and trained on English-language search queries, it features improved quality metrics compared to the first version and will be useful for developers of search systems and information retrieval systems.

Key features

  • Built on the lumen-ai/lumen-encoder-base-110m model using the sentence-transformers library.
  • Improved quality metrics on the fixed heldout_v2 dataset compared to v1.
  • Designed for evaluating short web fragments up to 512 tokens in size.
  • Outputs a relevance score for re-ordering the list of candidates.

What you can do

  • Integrate the model into your search pipeline to improve result ranking during the second stage.
  • Use it to recalculate document relevance for English-language search queries.
  • Apply it in custom search and information retrieval systems (RAG pipelines).

Example

The model is used to evaluate query-document pairs and outputs a numerical relevance score. For details on running it via the sentence-transformers library, see the repository documentation.

Read moreDetails
Benchmarks on heldout_v2: nDCG@10 is 0.731 (+0.029 gain), MRR@10 is 0.688 (+0.027 gain), Recall@50 is 0.874 (+0.025 gain), MAP@10 is 0.641 (+0.034 gain).
Limitations: not trained for summarizing long documents, answering table-based questions, or multilingual search.
Not a standalone search model and does not have its own document index.
Discuss Telegram 0 19

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 0.0 GB · ready files0.5 GB
  • FP8 / INT8recommendedweights 0.0 GB0.5 GB
  • 4-bitweights 0.0 GB0.5 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

What to get for this model

Massed ComputeOpen Massed Compute

Your own GPU: FP8 / INT8 — from RTX 2060 · BF16 — from RTX 2060

Massed Compute hourly prices, November 2026. The discounted price applies with the coupon at checkout.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # SOTAagi2030/lumen-search-reranker-v2
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download SOTAagi2030/lumen-search-reranker-v2 --local-dir ~/models/lumen-search-reranker-v2
    # The model is in ~/models. How to run it is in the model card (link above).
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of lumen-encoder-base-110m (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

SOTAagi2030/lumen-search-reranker-v2
Author
SOTAagi2030
Category
Other
Released
Posted

On NeyroLearn

Hugging Face

190

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments