Modelpublic

scaryrawr/occamy-1.0-oQ5e-mtp

sourceHugging Faceupdated 11d agoView on Hugging Face

1likes263downloads

Model Card

occamy-1.0-oQ5e-mtp

This model was quantized using oQ (oMLX v0.7.0.dev2) mixed-precision quantization.

It is based on Accio-Lab/occamy-1.0 and includes the official experimental BF16 MTP head from Accio-Lab/occamy-1.0-MTP, converted to the MLX tensor layout.

[!WARNING] Accio-Lab validated the head with BF16 and ModelOpt NVFP4 Occamy checkpoints. This MLX oQ variant has not been independently validated, so acceptance and performance may differ.

Quantization details

—Model type: qwen35moe
—Bits: 5
—Group size: 64
—Format: MLX safetensors