Instructions to use ilyakam/Geer-Ornith-1.0-35B-A3B-4-8bit-MLX with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use ilyakam/Geer-Ornith-1.0-35B-A3B-4-8bit-MLX 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("ilyakam/Geer-Ornith-1.0-35B-A3B-4-8bit-MLX")
config = load_config("ilyakam/Geer-Ornith-1.0-35B-A3B-4-8bit-MLX")

# 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 ilyakam/Geer-Ornith-1.0-35B-A3B-4-8bit-MLX with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "ilyakam/Geer-Ornith-1.0-35B-A3B-4-8bit-MLX"

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "ilyakam/Geer-Ornith-1.0-35B-A3B-4-8bit-MLX"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent

How to use ilyakam/Geer-Ornith-1.0-35B-A3B-4-8bit-MLX 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 "ilyakam/Geer-Ornith-1.0-35B-A3B-4-8bit-MLX"

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 ilyakam/Geer-Ornith-1.0-35B-A3B-4-8bit-MLX

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use ilyakam/Geer-Ornith-1.0-35B-A3B-4-8bit-MLX with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "ilyakam/Geer-Ornith-1.0-35B-A3B-4-8bit-MLX"

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 "ilyakam/Geer-Ornith-1.0-35B-A3B-4-8bit-MLX" \
  --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"

Geer-Ornith-1.0-35B-A3B-4-8bit-MLX / README.md

ilyakam

docs(card): document Geer 0.1.0 validation (#1)

3821f84 20 days ago

3.22 kB

language:
  - en
license: mit
library_name: mlx
pipeline_tag: image-text-to-text
base_model: deepreinforce-ai/Ornith-1.0-35B
tags:
  - geer
  - mlx
  - qwen3_5_moe
  - coding

Geer Ornith 1.0 35B-A3B (4/8-bit MLX)

This repository contains Geer's independently produced mixed 4/8-bit MLX conversion of Deep Reinforce AI's Ornith-1.0-35B. It is intended for local agentic coding on Apple Silicon through Geer.

The model identity is intentionally transparent: this is Ornith-1.0-35B, converted by the Geer project. It is not a new foundation model.

Build

Upstream model: deepreinforce-ai/Ornith-1.0-35B
Upstream revision: 5df2ed3f675c7beaa490328cc70bb573b65fb660
Architecture: qwen3_5_moe
Quantization: affine mixed 4/8-bit, group size 64
Four-bit tensors: routed expert gate, up, and down projections
Eight-bit tensors: other eligible language-model matrices, including embeddings, attention and Gated DeltaNet projections, routers, shared experts, and the language-model head
BF16 tensors: vision tower, norms, and other non-quantized parameters
Converted payload: 21,643,961,289 bytes (approximately 20.2 GiB)
Conversion runtime: MLX 0.32.0, MLX-LM 0.31.3, MLX-VLM 0.6.3
Hugging Face tooling: huggingface-hub 1.24.0, hf-xet 1.5.2
Safetensors: 0.8.0

The repository includes:

geer-recipe.toml, the complete pinned conversion recipe;
geer-source-manifest.json, hashes for the downloaded BF16 source;
geer-build-manifest.json, hashes for every converted output; and
the model, tokenizer, chat template, processor configuration, licenses, and attribution notices required to use the converted artifact.

The reproducible conversion tool and tensor-layout transformation are available in the Geer source repository.

Use with Geer

Geer downloads this repository at an immutable revision, verifies every file against geer-build-manifest.json, and activates the verified Hugging Face snapshot without copying the model into another directory. Geer 0.1.0 selects this conversion on 32 GB and 48 GB Macs, with 64K and 128K context windows respectively and BF16 KV cache.

Evaluation status

Deep Reinforce AI publishes results for the upstream BF16 Ornith model in its model card. Geer has not yet reproduced the upstream benchmark suite for this mixed quantization, so upstream scores should not be treated as measured results for this conversion. Although the upstream architecture and repository metadata support image-and-text inputs, Geer 0.1.0 has validated this conversion only for text-based agentic coding workflows.

License and attribution

Deep Reinforce AI declares Ornith-1.0-35B under the MIT license. Its model card states that Ornith-1.0-35B was post-trained on Qwen 3.5; the applicable Qwen Apache License 2.0 text and attribution are preserved here. See LICENSE, LICENSE-QWEN, and NOTICE before using or redistributing the model.

Geer is an independent project and is not affiliated with or endorsed by Deep Reinforce AI or Alibaba Cloud.