Full multimodal vision-language model; the vision tower is preserved.
The checkpoint's embedded Lightning MTP head is preserved.
Baseline 6-bit affine weights with data-driven mixed-precision boosts to 8-bit for 35 modules, group size 64.
Activation-weighted clipping used 128 multilingual/code calibration samples of 512 tokens and 504 collected importance entries.
Layer sensitivity was measured on a RAM-safe proxy rather than assigned uniformly.

oMLX usage

bash

hf download scaryrawr/Qwen3.8-27B-oQ6e-mtp \
  --local-dir ~/.models/scaryrawr/Qwen3.8-27B-oQ6e-mtp
omlx serve --model-dir ~/.models/scaryrawr

Enable mtp_enabled with mtp_num_draft_tokens: 3 in the model's oMLX settings. Text-only requests use native MTP draft/verify. Requests containing images or video retain multimodal support and conservatively use standard decode.

Six-bit affine weights use the standard MTP verify path because the specialized oMLX verify QMM kernel supports affine 4-bit and 8-bit weights.

Local validation

Measured on an M4 Max with 128 GB unified memory using adaptive MTP depth 3:

Runtime model memory: 22.54 GB
Warm predictable-code generation: 33.7 tokens/s
Open-ended prose generation: 27.8 tokens/s

On focused deterministic prompts, this model exactly matched the FP16 model's code, arithmetic, and structured-JSON outputs. Its generated LIS implementation passed 207/207 correctness cases, and French instruction-following and image-color checks also passed. These checks are useful smoke tests, not a comprehensive benchmark suite.

License

Apache 2.0, following the base model.