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.

LiquidAI logo

Can the Arc A770 16GB run LFM3-VL-2B?

LFM3-VL-2B·1.9B·Arc A770 16GB·16 GB VRAM

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

Yes. The Arc A770 16GB runs LFM3-VL-2B fully in VRAM at FP16 / BF16 (about 4.3 GB of 16 GB), at roughly 80 tokens/s.

best quant
FP16 / BF16
needs
4.3 GB
usable
16 GB
est. speed
~80 tok/s

Estimates for an 8K context; offload assumes 32 GB of system RAM.Try other contexts on the LFM3-VL-2B page →

Every quantisation of LFM3-VL-2B on the Arc A770 16GB

QuantQualityMemoryFitSpeed
Q2_K Heavy loss1.6 GBfits~488 tok/s
Q3_K_M Noticeable loss1.7 GBfits~329 tok/s
Q4_K_M Small loss1.9 GBfits~265 tok/s
Q5_K_M Near-lossless2.1 GBfits~226 tok/s
Q6_K Near-lossless2.3 GBfits~195 tok/s
Q8_0 Lossless2.7 GBfits~151 tok/s
FP16 / BF16 pickLossless4.3 GBfits~80 tok/s

Speed is a bandwidth-based estimate for generating text (560 GB/s); prompt processing and long contexts are slower.

Measured speed on the Arc A770 16GB

No community reports for this pair yet.

Report your own speed →

Run LFM3-VL-2B on the Arc A770 16GB

Using FP16 / BF16, the best fit for Arc A770 16GB.

Ollama · From the Ollama librarydocs
ollama run lfm3-vl:2b

The library build picks its own default quantisation. Ollama’s ROCm/CUDA builds don’t cover this GPU. Recent Ollama releases use Vulkan when its Vulkan backend and driver are installed (older ones needed OLLAMA_VULKAN=1); llama.cpp or LM Studio with Vulkan are the safer bet.

Ollama · F16 straight from Hugging Facedocs
ollama run hf.co/LiquidAI/LFM3-VL-2B-GGUF:F16

Ollama’s ROCm/CUDA builds don’t cover this GPU. Recent Ollama releases use Vulkan when its Vulkan backend and driver are installed (older ones needed OLLAMA_VULKAN=1); llama.cpp or LM Studio with Vulkan are the safer bet.

llama.cpp · OpenAI-compatible server with a web UIdocs
llama-server -hf LiquidAI/LFM3-VL-2B-GGUF:F16

Needs a Vulkan build of llama.cpp. Downloads the GGUF on first run.

llama.cpp · Docker with Vulkandocs
docker run -d --name llama --device /dev/dri -p 8080:8080 \
  -v ~/.cache/llama.cpp:/root/.cache/llama.cpp \
  ghcr.io/ggml-org/llama.cpp:server-vulkan \
  -hf LiquidAI/LFM3-VL-2B-GGUF:F16 --host 0.0.0.0 --port 8080 -ngl 99

Server and web UI on port 8080.

LM Studio · Download with the lms CLIdocs
lms get LiquidAI/LFM3-VL-2B-GGUF@F16

Then load it from the LM Studio app or with `lms load`.

Transformers · Ask about an imagedocs
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="LiquidAI/LFM3-VL-2B", dtype="auto", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
            {"type": "text", "text": "Describe this image."},
        ],
    }
]
out = pipe(text=messages, max_new_tokens=256)
print(out[0]["generated_text"][-1]["content"])

Vision models that fit the Arc A770 16GB comfortably

All that fit →
ModelParamsBest quantMemorySpeedLicence
2.3BQ8_0≈2.8 GBFits in memory~127 tok/sCommercial OK
LiquidAI logod1-2BLiquidAI
1.9BQ8_0≈2.7 GBFits in memory~151 tok/sConditions
2.1BQ8_0≈3.2 GBFits in memory~136 tok/sCommercial OK
1.6BQ8_0≈1.9 GBFits in memory~186 tok/sCommercial OK
1.3BQ8_0≈1.7 GBFits in memory~223 tok/sCommercial OK
2.2BQ8_0≈2.6 GBFits in memory~129 tok/sCommercial OK

Other GPUs that run LFM3-VL-2B

Frequently asked questions

Can the Arc A770 16GB run LFM3-VL-2B?
Yes. The Arc A770 16GB runs LFM3-VL-2B fully in VRAM at FP16 / BF16 (about 4.3 GB of 16 GB), at roughly 80 tokens/s.
How fast does LFM3-VL-2B run on the Arc A770 16GB?
An estimate from memory bandwidth gives about 80 tokens/s at FP16 / BF16.
Which quantisation of LFM3-VL-2B should I use on the Arc A770 16GB?
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.