---
base_model:
- Qwen/Qwen3.5-35B-A3B
---

## Updates
- 5/18/2026: I've uploaded new quants that include the MTP Tensors (@ Q8_0).
- 3/10/2026: I've uploaded new quants using the new fused Up + Gate conversion, this offers up to a +10% boost in prompt processing speed from my testing.

## Description

This repo contains specialized MoE-quants for Qwen3.5-35B-A3B. The idea being that given the huge size of the FFN tensors compared to the rest of the tensors in the model, it should be possible to achieve a better quality while keeping the overall size of the entire model smaller compared to a similar naive quantization. To that end, the quantization type default is kept in high quality and the FFN UP + FFN GATE tensors are quanted down along with the FFN DOWN tensors.

| Quant  | Size                 | Mixture                       | PPL                 | 1-(Mean PPL(Q)/PPL(base)) | KLD                 |
| :----- | :------------------- | :---------------------------- | :------------------ | :------------------------ | :------------------ |
| Q8_0   | 35.19 GiB (8.51 BPW) | Q8_0                          | 6.535248 ± 0.041570 | +0.0252%                  | 0.004522 ± 0.000057 |
| Q5_K_M | 25.28 GiB (6.12 BPW) | Q8_0 / Q5_K / Q5_K / Q6_K     | 6.534171 ± 0.041557 | +0.0087%                  | 0.006066 ± 0.000056 |
| Q4_K_M | 21.45 GiB (5.19 BPW) | Q8_0 / Q4_K / Q4_K / Q5_K     | 6.560168 ± 0.041799 | +0.4066%                  | 0.009884 ± 0.000075 |
| IQ4_XS | 17.23 GiB (4.17 BPW) | Q8_0 / IQ3_S / IQ3_S / IQ4_XS | 6.635447 ± 0.042320 | +1.5588%                  | 0.023868 ± 0.000204 |
| IQ3_S  | 13.48 GiB (3.26 BPW) | Q6_K / IQ2_S / IQ2_S / IQ3_S  | 6.920209 ± 0.044626 | +5.9172%                  | 0.061977 ± 0.000394 |

![kld_graph](kld_data/01_kld_vs_filesize.png "Chart showing Pareto KLD analysis of quants")
![ppl_graph](kld_data/02_ppl_vs_filesize.png "Chart showing Pareto PPL analysis of quants")