Bonus: Top up now and we'll double your first deposit — get x2 credits instantly.

gte-reranker-modernbert-base: Self-Hosting & Deployment Guide

gte-reranker-modernbert-base is an open-weight reranker that scores how relevant a passage is to a query.

What is gte-reranker-modernbert-base?

gte-reranker-modernbert-base is an open-weight reranker that scores how relevant a passage is to a query, published by Alibaba with open weights at Alibaba-NLP/gte-reranker-modernbert-base. LLM API does not route this model today, so there is no endpoint or code sample for it here. What this page does give you is the self-hosting picture: published weight sizes, the VRAM each quantisation needs and the commands to serve it yourself. Every figure is read from the public repository and refreshed weekly. At import it had 2,789,333 downloads on Hugging Face in the previous 30 days, under the apache-2.0 licence.

How to run gte-reranker-modernbert-base yourself

Open weights (apache-2.0), 150M parameters. Below: how much memory the weights need at each quantization, which hardware fits them, and the commands to serve it.

gte-reranker-modernbert-base VRAM requirements

Precision / quantizationWeightsFigureSource
BF16 (full precision)0.3 GBEstimate150M params × 2 bytes
GGUF Q8_00.16 GBFile sizekeisuke-miyako/gte-reranker-modernbert-base-gguf-q8_0

File size = actual download size on Hugging Face. Estimate = parameter count × bytes per weight. Weights only — the KV cache for long contexts needs extra memory; the fit tables below add 15% headroom as a rule of thumb.

GPU requirements: deploy gte-reranker-modernbert-base on your hardware

HardwareBF16 (full precision)GGUF Q8_0
NVIDIA H200141 GB1×1×
NVIDIA H10080 GB1×1×
NVIDIA A100 80GB80 GB1×1×
NVIDIA L40S48 GB1×1×

GPUs needed per variant.

HardwareGGUF Q8_0
RTX 509032 GB1×
RTX 409024 GB1×
RTX 309024 GB1×

GGUF via llama.cpp; more cards or CPU offload for larger files.

HardwareGGUF Q8_0
Mac, 512 GBM3 UltraFits
Mac, 192 GBM2 UltraFits
Mac, 128 GBM4 Max / M3 MaxFits
Mac, 64 GBM4 Pro / MaxFits

Assumes ~75% of unified memory is usable by the GPU.

HardwareBF16 (full precision)GGUF Q8_0
AWS p5.48xlarge8× H100 · 640 GBFitsFits
AWS p4de.24xlarge8× A100 80GB · 640 GBFitsFits
AWS g6e.12xlarge4× L40S · 192 GBFitsFits
Google Cloud a3-highgpu-8g8× H100 · 640 GBFitsFits
Azure ND H100 v58× H100 · 640 GBFitsFits

Whole-instance GPU memory; public instance specs.

Not on LLM.API yet

We do not route this model through our API at the moment. Browse the models you can call today — most workloads have a close match already live.

Not available on LLM API yet

We do not route this model through our API at the moment, so there is no endpoint or code snippet for it yet. Browse the models you can call today — most workloads have a close match already live.

Browse available models

When to Use — When NOT to Use

Use it if...

  • You want to re-order search or RAG candidates by relevance before they reach the LLM (gte-reranker-modernbert-base)
  • You want to keep documents inside your own network while indexing them
  • The weights are published openly, so you can pin one checkpoint and keep vectors reproducible
  • You are comparing self-hosting cost against a managed embedding endpoint

Avoid if...

  • LLM API does not route this model today, so there is no endpoint here to call
  • You want zero operations — serving, scaling and upgrades are yours to run
  • You need text generation — this model returns vectors or scores, not answers
  • You already have an index built with a different model: vectors are not interchangeable

Get one key to every model

Swap your API key. Keep your code.