Qwen3.8-Flash-Next-oQ2.5e-fp16-mtp

[!NOTE] Updated 2026-10-29: Re-quantized with the improved oQe quantization in oMLX 0.7.3. Please re-download if you have an earlier copy.

This model was quantized using oQ (oMLX v0.7.3) mixed-precision quantization with improved oQe (jundot/omlx#4930).

A mixed-precision MLX checkpoint of Qwen/Qwen3.8-Flash-Next, an experimental preview of the architecture behind Qwen4: a mixture-of-experts vision-language model with 125B parameters and 6B activated, plus a 51B n-gram embedding and a 4B MTP module, that pairs Gated DeltaNet with Qwen Sparse Attention (QSA) and uses Gated Residual streams. It includes the language model, the n-gram embedding, the vision encoder and the multi-token prediction (MTP) head.

This FP16 build stores the quantization scales and biases and the remaining unquantized tensors in FP16 instead of BF16, and keeps the vision encoder in FP32.

The effective average is 3.113 bits/weight for the language backbone and 3.744 bits/weight for the complete checkpoint.

Quantization and Bit Distribution

oQ2.5e allocates bits by measured layer sensitivity, then rounds each group with an importance matrix (1024 samples x 512 tokens) and a weighted least-squares refit of its scale and bias. Below oQ4, the N-gram embedding table stays at 4 bits and is left out of the bit budget, because oMLX can serve it from SSD instead of memory.

Storage Breakdown

License

Qwen Community License 1.0, following the source model. See LICENSE.