These are MXFP4 quantizations of the model Qwen3.6-35B-A3B
This is the multi-token prediction (MTP) version.
Quick Start
- Download the latest release of llama.cpp.
- Download your preferred model variant from below.
Which version should I choose?
All variants use MXFP4 for the MoE (Mixture of Experts) weights to keep the model efficient. The difference lies in how the remaining tensors are handled:
| Variant | Quality | Performance | Size | Recommendation |
|---|---|---|---|---|
| BF16 | ⭐⭐⭐ | Variable* | 21.39GiB | Best for maximum accuracy; original unquantized weights. |
| F16 | ⭐⭐ | Fast | 21.39GiB | Great alternative if BF16 is slow on your hardware. |
| Q8 | ⭐ | Fastest | 19.71GiB | Balanced performance and memory usage. |
**Note: On some older architectures, BF16 may be slower than F16. Check that your GPU supports native BF16 *
Read the guide from unsloth in order to set up the model's recommended settings for MTP:
Qwen3.6 - MTP Guide
On my system it works very well with the commands:
--spec-type draft-mtp
--spec-draft-p-min 0.75
--spec-draft-n-max 3