Instructions to use OsaurusAI/LFM3-VL-2B-JANG_6M with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use OsaurusAI/LFM3-VL-2B-JANG_6M with MLX:

# Make sure mlx-vlm is installed
# pip install --upgrade mlx-vlm

from mlx_vlm import load, generate
from mlx_vlm.prompt_utils import apply_chat_template
from mlx_vlm.utils import load_config

# Load the model
model, processor = load("OsaurusAI/LFM3-VL-2B-JANG_6M")
config = load_config("OsaurusAI/LFM3-VL-2B-JANG_6M")

# Prepare input
image = ["http://images.cocodataset.org/val2017/000000039769.jpg"]
prompt = "Describe this image."

# Apply chat template
formatted_prompt = apply_chat_template(
    processor, config, prompt, num_images=1
)

# Generate output
output = generate(model, processor, formatted_prompt, image)
print(output)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use OsaurusAI/LFM3-VL-2B-JANG_6M with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "OsaurusAI/LFM3-VL-2B-JANG_6M"

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "OsaurusAI/LFM3-VL-2B-JANG_6M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use OsaurusAI/LFM3-VL-2B-JANG_6M with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "OsaurusAI/LFM3-VL-2B-JANG_6M"

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default OsaurusAI/LFM3-VL-2B-JANG_6M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use OsaurusAI/LFM3-VL-2B-JANG_6M with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "OsaurusAI/LFM3-VL-2B-JANG_6M"

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "OsaurusAI/LFM3-VL-2B-JANG_6M" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

OsaurusAI/LFM3-VL-2B-JANG_6M

JANG_6M build of LiquidAI/LFM3-VL-2B for Apple Silicon. LFM3-VL 2B: the LFM3 trunk with a 14-layer conv/attention vision tower; thinking on by default.

8-bit token mixers, 6-bit FFN, and every quantized matrix calibrated with GPTQ on the exact storage grid plus activation-aware scaling, using calibration data weighted toward tool calls, terminal agents, reasoning and code.

Bundle


Modalities	text, images, video
On-disk size	1.90 GiB (one shard), float16
Attention q/k/v/o, short-conv in/out projections	8-bit affine, group 32, GPTQ
FFN w1 / w3 / w2	6-bit affine, group 32, AWQ + GPTQ
Tied embedding / lm_head	8-bit affine, group 32 (it is the output head)
Kept at full precision	attention output gate, depthwise conv kernels, all norms, vision tower and projector
GPTQ	142/142 tensors: fixed min-max grid (byte-identical layout to `mx.quantize`), act-order, Hessians accumulated over every calibration token, w2 Hessian rebuilt through the already-quantized w1/w3, best-of-RTN guard; mean calibration reconstruction error -52.9% vs round-to-nearest
AWQ	channel scales (alpha 0.25, clip 0.5-2) folded into `ffn_norm` and `w3` rows; function-preserving, no runtime cost

Calibration (634,978 tokens)

Every sample is rendered through the model's own chat template: structured tools, assistant tool calls in LFM3's <tool_call><function=...> form, tool results as tool turns, and agent thoughts inside <think>. Part of the set is the model's own answers (card sampling) to reasoning, tool, terminal and coding prompts, so its serving distribution is in calibration.

domain	tokens
tool calling (Hermes, native XML calls)	166,403
terminal / SWE agent trajectories	175,686
reasoning (long traces)	106,871
coding and agentic text	66,629
general, science, systems, security, multilingual, long context	37,598
images (photos, screenshots, documents)	81,791

Measured against the bf16 source

69 held-out sequences never seen in calibration (15,021 scored positions: tool calls, terminal and SWE agents, reasoning, code, general text, images). KL divergence in nats per position (lower is better); top-1 is argmax agreement with bf16.

build	size	KL median	KL mean	KL p95	top-1
this bundle	1.90 GiB	0.00012	0.00197	0.0069	98.67%
same layout, round-to-nearest	1.90 GiB	0.00018	0.00281	0.0092	98.44%
uniform 6-bit gs64 (mlx-vlm default recipe)	1.66 GiB	0.00045	0.00665	0.0242	97.54%

Use

Runtimes with LFM3 support: vMLX / Osaurus (vmlx-swift with LFM3, Osaurus with LFM3 media support), and mlx-vlm with the LFM3 port.
Thinking: on by default (enable_thinking). Tool calls: XML function format. Sampling defaults: temperature 0.1, top_k 50, repetition_penalty 1.1.
License: LFM Open License v1.0 (see LICENSE), same as the base model.

Downloads last month: 3

Safetensors

Model size

2B params

Tensor type

U32

F16

MLX

Hardware compatibility

6-bit

Model tree for OsaurusAI/LFM3-VL-2B-JANG_6M

Base model

LiquidAI/LFM3-1.7B-Base

Finetuned

LiquidAI/LFM3-VL-2B

Quantized

(20)

this model