Geer-Ornith-1.5-35B-A3B-6bit-MLX

Geer Ornith 1.5 35B-A3B (6-bit MLX)

This repository contains Geer's independently produced 6-bit MLX conversion of Ornith-1.5-35B-A3B. It is intended for local agentic coding on Apple Silicon through Geer. The model identity is intentionally transparent: this is Ornith-1.5-35B-A3B converted by Geer, not a new foundation model.

Build

• Upstream model: ornith-ai/Ornith-1.5-35B-A3B
• Upstream revision: e4dfb35a93d4b6822a811a7676f3488514abe7e2
• Architecture: qwen3_5_moe
• Quantization: affine 6-bit, group size 64
• Source payload: 71,928,654,040 bytes (approximately 67.0 GiB)
• Converted size: recorded in geer-build-manifest.json
• 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 for review of the converted artifact.

The reproducible conversion tool and tensor-layout transformation are available in the Geer source repository. The upstream 1.5 checkpoint contains native mtp.* tensors; the pinned MLX-VLM 0.6.3 conversion path sanitizes those tensors out, so this candidate must be qualified before any publication or activation claim.

Use with Geer

Geer downloads a published repository only at an immutable revision, verifies every file against geer-build-manifest.json, and activates the verified Hugging Face snapshot without copying model bytes. This candidate is prepared for review and is not activated automatically.

Evaluation status

The upstream model card reports benchmark results for Ornith-1.5. Geer has not reproduced those results for this MLX quantization, so upstream scores are not measured results for this conversion. The repository metadata supports multimodal inputs, but Geer qualification must separately verify image and text serving; the current Geer workflow is validated for text-based agentic coding.

License and attribution

The pinned upstream model card declares Ornith-1.5-35B-A3B under the MIT license and links to a license file. The pinned snapshot currently does not include that file, so the exact upstream license text and any inherited component obligations must be confirmed before redistribution. This candidate preserves the Qwen Apache notice carried by the existing Geer Ornith model; that notice is not a substitute for upstream legal review.

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