Instructions to use Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision 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("Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision")
config = load_config("Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision")

# 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 Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision"

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": "Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision 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 "Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision"

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 Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision"

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 "Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision" \
  --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"

Ornith 1.5 35B A3B MLX 4-bit + BF16 Vision

This is a complete MLX vision-language checkpoint assembled from the official Ornith 1.5 releases. It combines the official 4-bit MLX language checkpoint with the official BF16 vision tower that is absent from the MLX 4-bit release.

Sources

Language checkpoint: ornith-ai/Ornith-1.5-35B-A3B-MLX-4bit
- pinned revision: 19504d912fa8fc7622bf6b1de3db5d5d890b1f02
Vision and processor metadata: ornith-ai/Ornith-1.5-35B-A3B
- pinned revision: 10fbf86fed7ecee4a061f8b499a618f46001cac1
- source shard: model-00001-of-00016.safetensors

The four language-model shards remain byte-identical to the official MLX 4-bit release. All 333 model.visual.* tensors were streamed from the official BF16 source into ornith15_vision_bf16.safetensors, renamed to the MLX vision_tower.* layout. The Conv3d patch-embedding tensor was transposed from PyTorch [out,in,t,h,w] to MLX [out,t,h,w,in] layout.

VISION_SIDECAR.json records the exact source revisions, source-shard hash, derived sidecar hash, tensor count and layout conversion. CHECKSUMS.sha256 records the immutable runtime-file hashes.

Precision and runtime

Language body: affine 4-bit, group size 64
Vision tower: BF16
Architecture: Qwen3.5/Qwen3.6-compatible MoE VLM
Recommended AI2Apps Runtime: ai2apps/runtime-omlx >= 1.5.6

The mixed precision is intentional: it preserves visual quality while keeping the language model compact. AI2Apps can run the checkpoint fully resident or convert its routed experts to the Direct Cached-MoE store for lower memory use. Full-resident execution is the default when memory allows.

License and attribution

The upstream Ornith 1.5 model is released under the MIT License. This mirror preserves the upstream model terms and records the transformation provenance; it does not claim authorship of the model weights.

Ornith: https://ornith.ai/
Official model: https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B
Official MLX 4-bit model: https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B-MLX-4bit

Downloads last month: 148

Safetensors

Model size

5B params

Tensor type

BF16

U32

MLX

Hardware compatibility

4-bit

Model tree for Avdpro/Ornith-1.5-35B-A3B-MLX-4bit-Vision

Base model

ornith-ai/Ornith-1.5-35B-A3B

Quantized

(116)

this model