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

Can the RTX 3060 12GB run lumi-4b?

lumi-4b·4B·GeForce RTX 3060 12GB·12 GB VRAM

Saved in this browser, used on every model page.
Yes

Yes. The RTX 3060 12GB runs lumi-4b fully in VRAM at FP16 / BF16 (about 9.6 GB of 12 GB), at roughly 25 tokens/s.

best quant
FP16 / BF16
needs
9.6 GB
usable
12 GB
est. speed
~25 tok/s

Estimates for an 8K context; offload assumes 32 GB of system RAM.Try other contexts on the lumi-4b page →

Every quantisation of lumi-4b on the RTX 3060 12GB

QuantQualityMemoryFitSpeed
Q2_K Heavy loss3.1 GBfits~152 tok/s
Q3_K_M Noticeable loss3.7 GBfits~102 tok/s
Q4_K_M Small loss4.2 GBfits~82 tok/s
Q5_K_M Near-lossless4.6 GBfits~70 tok/s
Q6_K Near-lossless5.7 GBfits~60 tok/s
Q8_0 Lossless5.9 GBfits~47 tok/s
FP16 / BF16 pickLossless9.6 GBfits~25 tok/s

Speed is a bandwidth-based estimate for generating text (360 GB/s); prompt processing and long contexts are slower.

Measured speed on the RTX 3060 12GB

No community reports for this pair yet.

Report your own speed →

Run lumi-4b on the RTX 3060 12GB

Using FP16 / BF16, the best fit for RTX 3060 12GB.

llama.cpp · OpenAI-compatible server with a web UIdocs
llama-server -hf Mateus-Nery/lumi-4b-GGUF:BF16

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

vLLM · Serve an OpenAI-compatible APIdocs
vllm serve ceia-br/lumi-4b
vLLM · Smaller FP8 builddocs
vllm serve liodon-ai/lumi-4b-FP8
vLLM · Docker Compose (NVIDIA)docs
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.

SGLang · Serve an OpenAI-compatible APIdocs
sglang serve ceia-br/lumi-4b --host 0.0.0.0 --port 30000
SGLang · Docker Compose (NVIDIA)docs
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.

Chat models that fit the RTX 3060 12GB comfortably

All that fit →
ModelParamsBest quantMemorySpeedLicence
4BQ8_0≈5.8 GBFits in memory~46 tok/sCommercial OK
3.2BQ8_0≈4.7 GBFits in memory~58 tok/sConditions
3.8BQ8_0≈5.5 GBFits in memory~49 tok/sCommercial OK
4BQ8_0≈5.8 GBFits in memory~46 tok/sCommercial OK
4.1BQ8_0≈5.9 GBFits in memory~45 tok/sCommercial OK
3.1BQ8_0≈4.3 GBFits in memory~60 tok/sNon-commercial

Other GPUs that run lumi-4b

Frequently asked questions

Can the RTX 3060 12GB run lumi-4b?
Yes. The RTX 3060 12GB runs lumi-4b fully in VRAM at FP16 / BF16 (about 9.6 GB of 12 GB), at roughly 25 tokens/s.
How fast does lumi-4b run on the RTX 3060 12GB?
An estimate from memory bandwidth gives about 25 tokens/s at FP16 / BF16.
Which quantisation of lumi-4b should I use on the RTX 3060 12GB?
FP16 / BF16: the largest one that fits. Expected quality: lossless compared with the full-precision weights.

Memory and speed are estimates from the model's size and the card's memory bandwidth; real numbers depend on the runtime, drivers and context length. Community figures are medians of user reports.