Instructions to use netprojects/h2o-lightning-4b-mlx-q4 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use netprojects/h2o-lightning-4b-mlx-q4 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("netprojects/h2o-lightning-4b-mlx-q4")
config = load_config("netprojects/h2o-lightning-4b-mlx-q4")

# 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 netprojects/h2o-lightning-4b-mlx-q4 with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "netprojects/h2o-lightning-4b-mlx-q4"

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": "netprojects/h2o-lightning-4b-mlx-q4"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use netprojects/h2o-lightning-4b-mlx-q4 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 "netprojects/h2o-lightning-4b-mlx-q4"

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 netprojects/h2o-lightning-4b-mlx-q4

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use netprojects/h2o-lightning-4b-mlx-q4 with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "netprojects/h2o-lightning-4b-mlx-q4"

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 "netprojects/h2o-lightning-4b-mlx-q4" \
  --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"

H2O-Lightning-4B — MLX 4-bit

A 4-bit MLX conversion (affine, group size 64; mlx-vlm 0.7.8) of h2oai/h2o-lightning-4b v1.3.1, for Apple Silicon. 3.0 GB. The smallest build, for one-way reading.

Published by NetProjects (Internet Projects Ltd). Not affiliated with H2O.ai.

This is a decision model, not a chat model. It returns a probability for each answer option, not text.

Serve it with the H2O shim, not with mlx_lm.server or a plain mlx_vlm generate call. The generic "Use this model" snippets Hugging Face shows for MLX models do not return the label probabilities this model is made for. The commands below set it up with H2O's official shim, unmodified, on top of h2o-lightning-mlx, which reads the label log-probabilities with a float32 head, as the original does on vLLM.

JevBench public (231, JevBench v1.2.1, `--adapter typesafe`), one-way
this checkpoint	199 / 231 (48/48 · 70/72 · 81/111)
original weights on MLX (bf16)	198 / 231
H2O's model card (vLLM)	198 / 231
vs bf16 on MLX	221 / 231 answers identical; probability drift mean 0.023, p95 0.106

Which build? This build serves one-way reading: start it with TWO_WAY=0. For H2O's two-way reading (203 / 231), and for the closest match to the original's calibration, use the 8-bit build (5.1 GB).

Images. Image input works through the unchanged shim. On H2O's own receipt example this build picks the same answer as the original weights, with somewhat lower confidence (0.88 against 0.98). It is not a reproduction of H2O's image benchmark.

Limits (as for the original): mostly evaluated in English; works best with about 16 answer options or fewer; no explanations and no video input.

git clone https://github.com/internet-projects/h2o-lightning-mlx && cd h2o-lightning-mlx && ./setup.sh
hf download netprojects/h2o-lightning-4b-mlx-q4 --local-dir mlx-q4
TWO_WAY=0 MODEL=mlx-q4 ./serve.sh # http://127.0.0.1:8741/v1/systemone

Speed and memory

Measured on a MacBook Pro M5 Max (128 GB) on 23 October 2026, with no other models running, through serve.sh (/v1/systemone, one request at a time, over HTTP on the same Mac) with H2O v1.3.1. Times are per decision, from the client's side, so they include the shim and HTTP. The facade's prefix cache was off and the weights stayed loaded. Medians of 3 passes or runs after a warm-up.

Mode	Input	Median per decision (95th percentile)	Decisions per second	Memory in use at peak
one-way	120 JevBench items (median 130 tokens)	61 ms (66 ms)	13.4	5.6 GB
one-way	one record of about 2,000 tokens	424 ms	2.36	4.8 GB
one-way	one 4K-token state	892 ms	1.12	5.7 GB
one-way	one 16K-token state	3.9 s	0.26	6.3 GB
one-way	one 32K-token state	8.6 s	0.12	6.8 GB

This build runs one-way only: it does not pass H2O's two-way probe check, so the shim refuses two-way mode; for two-way reading use the 8-bit build. The shim keeps states to 32,000 tokens (the start and the end of longer ones), so the 32K row is 32,098 tokens with the template.

Versions. main is H2O v1.3.1 (updated 19 October 2026). The previous build, H2O v1.2.3 (207 / 231), stays available with hf download netprojects/h2o-lightning-4b-mlx-q4 --revision v1.2.3; use it with git checkout v1.2.3 in the GitHub repo.

Credits: H2O.ai (H2O-Lightning-4B, Apache-2.0), Qwen team (Qwen3.5-4B, Apache-2.0). This conversion is not affiliated with H2O.ai.

Downloads last month: 190

Safetensors

Model size

5B params

Tensor type

U32

BF16

F32

MLX

Hardware compatibility

4-bit

Model tree for netprojects/h2o-lightning-4b-mlx-q4

Base model

Qwen/Qwen3.5-4B-Base

Finetuned

Qwen/Qwen3.5-4B

Finetuned

h2oai/h2o-lightning-4b

Quantized

(6)

this model