selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

ceia-br logo

lumi-4b

ceia-br·Instruct·released 27 Oct 2026

Unknown licence4B256K ctx ≈4.2 GB
Your GPU
llama-server -hf Mateus-Nery/lumi-4b-GGUF:Q4_K_XL
All 3 runners ↓

O Overall é a média aritmética simples dessas cinco métricas, com peso igual para cada uma. Nesse recorte, o Lumi-4B ocupa a 2ª posição entre 9 modelos.

Notify me when

How much VRAM does lumi-4b need?

At Q4_K_M lumi-4b needs about 4.2 GB with an 8K context (4.2 GB at Q4, 9.6 GB at full precision). Pick your GPU to see what fits and how fast it runs.

--vram
any
--gpus
--quant

Needs: ≈4.2 GB · Q4_K_M · 8K context — Pick a GPU to check the fit.

Change quant, context or KV cache to compare
Advanced: context 8K · KV cache 16-bit
Context
KV cache

Smallest GPU for each quant and context (KV cache 16-bit)

Quantisation8K16K32K64K128K256K
Q3_K_M
Q4_K_M
Q5_K_M
Q6_K
Q8_0
FP16

≤16 GB card24–32 GB48–96 GBmore

QuantisationQualityWeightsKV cacheTotal
Q2_KHeavy lossSmallest, noticeably worse qualityHeavy loss1.2 GB est.1.3 GB≈3.1 GB
Q3_K_MNoticeable lossSmall, some quality lossNoticeable loss1.8 GB est.1.3 GB≈3.7 GB
Q4_K_MrecommendedSmall lossRecommended default for local useSmall loss2.2 GB est.1.3 GB≈4.2 GB
Q5_K_MNear-losslessNear-lossless for most tasksNear-lossless2.6 GB est.1.3 GB≈4.6 GB
Q6_KNear-losslessPractically losslessNear-lossless3.7 GB file1.3 GB≈5.7 GB
Q8_0LosslessLossless in practice, twice the size of Q4Lossless3.9 GB est.1.3 GB≈5.9 GB
FP16 / BF16LosslessFull precision, as releasedLossless7.4 GB file1.3 GB≈9.6 GB

Estimates for 8K tokens of context: weights + KV cache + runtime overhead. Speed assumes memory-bound decoding at your GPU’s bandwidth. Pick a GPU above and the fit column checks every row against it. Weights marked “file” use the real GGUF download size; “est.” rows are computed from the parameter count.

How to pick a quant

Take the biggest quant that fits with room left for context. Q4_K_M is the usual sweet spot. Below Q4 quality drops fast, above Q6 you mostly pay memory for nothing. Big models (70B+, large MoE) tolerate low quants better than small ones.

  • LosslessF16, Q8_0
  • Near-losslessQ6_K, Q5_K_M
  • Small lossQ5_K_S, Q4_K_M, IQ4_NL, IQ4_XS, Q4_K_S
  • Noticeable lossQ4_0, Q3_K_L, Q3_K_M, IQ3_M
  • Heavy lossIQ3_XXS, Q3_K_S, IQ2_M, Q2_K

Typical quality loss, not measured per model. Tiers come from KL divergence against FP16 on Llama 3 8B: llama.cpp perplexity README: LLaMA 3 8B scoreboard (KLD and ΔPPL vs FP16, Wikitext-2); llama.cpp quantize tool: ΔPPL per quant type @ Llama-3-8B

How much context fits?

Model quant
KV cache

At 8K context, Q4_K_M needs ≈4.2 GB(2.2 GB weights + 1.3 GB KV cache + 0.6 GB overhead). Pick a GPU above to see how much context fits.

010203040501K2K4K8K16K32K64K128K256KGB

Weights + overheadKV cache

Can the RTX 4060 run lumi-4b? →

Speed vs quality per quant

Smaller quants run faster and fit in less memory, but lose quality.

Q2_KQ3_K_MQ4_K_MQ5_K_MQ6_KQ8_0F16

Fits in VRAMTight fitCPU offload✕ won't runQuality is typical KL divergence vs FP16, not measured for this model.

Real-world speed of lumi-4b

Tokens per second measured by people running it at home. A GPU shows a median once 3 reports agree on the same setup. Reports far from our estimate wait for a review. Measure yours with llama-bench →

No medians yet. Be one of the first 3 to report a setup.

All reports

Run lumi-4b locally

--quantQ4_K_Mdefault; pick your GPU to tailor itMateus-Nery/lumi-4b-GGUF

OpenAI-compatible server with a web UIdocs
llama-server -hf Mateus-Nery/lumi-4b-GGUF:Q4_K_XL

Downloads the GGUF on first run. Installed through llama.app? The same flags work with `llama serve`.

Install llama.cpp

Serve an OpenAI-compatible APIdocs
vllm serve ceia-br/lumi-4b
Smaller FP8 builddocs
vllm serve liodon-ai/lumi-4b-FP8
Docker Compose (NVIDIA)docs
yaml
services:
  vllm:
    image: vllm/vllm-openai:latest
    command: ceia-br/lumi-4b
    ports:
      - "8000:8000"
    ipc: host
    volumes:
      - ~/.cache/huggingface:/root/.cache/huggingface
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

API on port 8000.

Install vLLM

Serve an OpenAI-compatible APIdocs
sglang serve ceia-br/lumi-4b --host 0.0.0.0 --port 30000
Docker Compose (NVIDIA)docs
yaml
services:
  sglang:
    image: lmsysorg/sglang:latest
    command: sglang serve ceia-br/lumi-4b --host 0.0.0.0 --port 30000
    ports:
      - "30000:30000"
    ipc: host
    volumes:
      - ~/.cache/huggingface:/root/.cache/huggingface
    deploy:
      resources:
        reservations:
          devices:
            - driver: nvidia
              count: all
              capabilities: [gpu]

API on port 30000.

Install SGLang

Quantised versions and downloads

RepositoryFormatQuantSizeDownloads
liodon-ai/lumi-4b-FP8SafetensorsFP8—51
Mateus-Nery/lumi-4b-GGUFGGUFIQ4_XS2.4 GB1.6K
Mateus-Nery/lumi-4b-GGUFGGUFQ3_K_XL2.7 GB1.6K
Mateus-Nery/lumi-4b-GGUFGGUFQ5_K_S3.0 GB1.6K
Mateus-Nery/lumi-4b-GGUFGGUFQ4_K_XL3.1 GB1.6K
Mateus-Nery/lumi-4b-GGUFGGUFQ6_K3.9 GB1.6K
Mateus-Nery/lumi-4b-GGUFGGUFQ6_K_XL4.6 GB1.6K
Mateus-Nery/lumi-4b-GGUFGGUFBF168.0 GB1.6K
HyperScale/lumi-4b-GGUFGGUF——223

About lumi-4b

O Overall é a média aritmética simples dessas cinco métricas, com peso igual para cada uma. Nesse recorte, o Lumi-4B ocupa a 2ª posição entre 9 modelos.

Full model card on Hugging Face →

Alternatives to lumi-4b

All alternatives →
nvidia logo≈4.2 GB

NVIDIA-Nemotron-3-Nano-4B-BF16 is a small language model (SLM) trained from scratch by NVIDIA, and designed as a unified model for both reasoning and non-reasoning tasks. It responds to user queries and tasks by first generating a reasoning trace and then concluding with a final response.

4B256K ctx nvidia-nemotron-open-model-license
1.9MMar 2026gguf
Qwen logo≈4.0 GB

We introduce the updated version of the Qwen3-4B non-thinking mode, named Qwen3-4B-Instruct-2507, featuring the following key enhancements:

4B256K ctx Apache 2.0
2.6MAug 2025gguf
Qwen logo

Qwen3-4B

Qwen·Chat

≈4.0 GB

Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models.

4B40K ctx Apache 2.0
3.2MApr 2025ollama
Nanbeige logo

Nanbeige4.2-3B

Nanbeige·Chat

≈3.7 GB

Nanbeige4.2-3B is a compact agentic model built on Nanbeige4.2-3B-Base, designed to combine strong agentic behavior with broad reasoning and alignment capabilities. Its Looped Transformer architecture reuses the transformer layers to increase model capacity without adding parameters.

4.2B256K ctx Apache 2.0
96.9KJul 2026gguf
ibm-granite logo

granite-4.1-3b

ibm-granite·Chat

≈3.1 GB

Model Summary: Granite-4.1-3B is a 3B parameter long-context instruct model finetuned from Granite-4.1-3B-Base using a combination of open source instruction datasets with permissive license and internally collected synthetic datasets.

3.4B128K ctx Apache 2.0
456.9KApr 2026ollama
XHToken logo

Spark-X2.6-4B

XHToken·Chat

≈4.1 GB

A little over a month has passed since Spark-X2.5. In the rapidly evolving world of AI agents, that is enough time to rethink, refine, and move the frontier forward.

4.1B1M ctx Apache 2.0
29.1KOct 2026gguf

lumi-4b FAQ

How much VRAM does lumi-4b need?
About 4.2 GB at Q4_K_M with an 8K context. Q8_0 needs about 5.9 GB and full precision about 9.6 GB. Longer contexts need more memory for the KV cache.
Can lumi-4b run on the RTX 4060?
Yes. It fits at Q8_0 (about 5.9 GB), at an estimated 35 tokens/s. It fits comfortably on the Apple M4 (16 GB), Radeon RX 6700 XT, RTX 3060 12GB.
Can I use lumi-4b commercially?
Unclear. It is released under Unknown licence. No licence stated: assume all rights reserved. Check the licence on the model card before using it in a product.
What is the context length of lumi-4b?
262,144 tokens (256K). The full window needs far more memory than an 8K context.
How do I run lumi-4b with llama.cpp?
Run llama-server -hf Mateus-Nery/lumi-4b-GGUF:Q4_K_XL. It downloads the GGUF and starts an OpenAI-compatible server.

Data from Hugging Face, the model card and public leaderboards. Memory and speed figures are estimates; real numbers depend on the runtime, context length and drivers. Always check the licence yourself before commercial use.