Skip to content
Computer vision

LFM3-VL-2B

a compact multimodal model with video and reasoning support

2 min readupdated

Liquid AI's new compact multimodal model (1.92B parameters) is built for on-device visual understanding. It takes text, images, and—for the first time in the LFM lineup—video inputs, and uses a built-in thinking block to reason step by step before answering. Built on the LFM3-1.7B-Base language model and a new proprietary vision encoder, it supports a context window of up to 128K tokens. It excels at chart and document analysis as well as spatial object positioning, making it a solid fit for developers and hobbyists building local multimodal apps and robotics.

Key features

  • Video understanding: the first LFM-series model capable of processing video clips, defaulting to 2 fps.
  • Step-by-step reasoning: thinks before answering, which helps with complex visual tasks, math, and chart analysis.
  • Spatial reasoning: detects objects in a frame and analyzes their relative positions for robotics tasks.
  • Multilingual support: handles 16 languages, including Russian.

What you can do

  • Video analysis: ask questions about recorded lectures, phone clips, or security camera footage.
  • Chart and document reading: extract text and data from scanned documents, app interfaces, and charts.
  • Robotics and edge devices: use spatial reasoning for local object detection on robots and smart gadgets.
  • Local deployment: run the model locally via llama.cpp (GGUF), vLLM, or SGLang for fast inference.

Example

import torch
from transformers import AutoModelForImageTextToText, AutoProcessor

model_id = "LiquidAI/LFM3-VL-2B"

model = AutoModelForImageTextToText.from_pretrained(
    model_id, dtype=torch.bfloat16, device_map="auto", trust_remote_code=True
)
processor = AutoProcessor.from_pretrained(model_id, trust_remote_code=True)

messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "[https://placecats.com/300/200](https://placecats.com/300/200)"},
            {"type": "text", "text": "Describe this image."},
        ],
    }
]

inputs = processor.apply_chat_template(
    messages,
    add_generation_prompt=True,
    tokenize=True,
    return_dict=True,
    return_tensors="pt",
).to(model.device)

with torch.inference_mode():
    output_ids = model.generate(
        **inputs,
        do_sample=True,
        temperature=0.1,
        top_k=50,
        repetition_penalty=1.1,
        max_new_tokens=16384,
    )

generated_ids = output_ids[:, inputs["input_ids"].shape[1] :]
print(processor.batch_decode(generated_ids, skip_special_tokens=True)[0])
Read moreDetails
Parameters: 1.92B (1.73B language model, 164M vision encoder, 25M projector).
Context: 128K tokens.
Architecture: 24 language model layers (combining convolutions and attention) and 14 vision encoder layers.
Limitations: not recommended for coding, multi-turn conversations, or complex reasoning. License: LFM Open License v1.0.
Examples
Discuss Telegram 0 20

Will it run?

Runs in BF16

Fits without quantisation.

  • BF16weights 3.6 GB · ready files4.4 GB
  • FP8 / INT8recommendedweights 1.8 GB2.5 GB
  • 4-bitweights 1.0 GB1.6 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.

    # ps6275/LFM3-VL-2B
    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 ps6275/LFM3-VL-2B --local-dir ~/models/LFM3-VL-2B
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to LFM3-VL-2B
  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 LFM3-1.7B-Base (quants, finetunes, LoRAs) will show up under the bell.

Quantised versions 12All on Hugging Face
Fine-tunes 12All on Hugging Face
LoRAs and adapters 9All on Hugging Face

About the model

ps6275/LFM3-VL-2B
Author
ps6275
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

200

Ask AI about this model

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

Comments