DeadBranches/AesSedai-Qwen3.5-35B-A3B-GGUF overview
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,…
Runs locally from ~10.4 MB disk (4 GB VRAM class GPUs with llama.cpp / guIDE).
Repository Files & Downloads
| File | Type | Quantization | Size | Link |
|---|---|---|---|---|
| IQ3_S/Qwen3.5-35B-A3B-IQ3_S-00001-of-00002.gguf | GGUF | IQ3_S | 10.4 MB | Download |
| IQ3_S/Qwen3.5-35B-A3B-IQ3_S-00002-of-00002.gguf | GGUF | IQ3_S | 13.48 GB | Download |
| IQ4_XS/Qwen3.5-35B-A3B-IQ4_XS-00001-of-00002.gguf | GGUF | IQ4_XS | 10.4 MB | Download |
| IQ4_XS/Qwen3.5-35B-A3B-IQ4_XS-00002-of-00002.gguf | GGUF | IQ4_XS | 17.23 GB | Download |
| Q4_K_M/Qwen3.5-35B-A3B-Q4_K_M-00001-of-00002.gguf | GGUF | Q4_K_M | 10.4 MB | Download |
| Q4_K_M/Qwen3.5-35B-A3B-Q4_K_M-00002-of-00002.gguf | GGUF | Q4_K_M | 21.45 GB | Download |
| Q5_K_M/Qwen3.5-35B-A3B-Q5_K_M-00001-of-00002.gguf | GGUF | Q5_K_M | 10.4 MB | Download |
| Q5_K_M/Qwen3.5-35B-A3B-Q5_K_M-00002-of-00002.gguf | GGUF | Q5_K_M | 25.28 GB | Download |
| imatrix.gguf | GGUF | GGUF | 103.3 MB | Download |
| mmproj-Qwen3.5-35B-A3B-BF16.gguf | GGUF | BF16 | 861.0 MB | Download |
| mmproj-Qwen3.5-35B-A3B-F16.gguf | GGUF | F16 | 857.6 MB | Download |
| mmproj-Qwen3.5-35B-A3B-F32.gguf | GGUF | F32 | 1.66 GB | Download |
| mmproj-Qwen3.5-35B-A3B-Q8_0.gguf | GGUF | Q8_0 | 585.7 MB | Download |
Model Details
| Model ID | DeadBranches/AesSedai-Qwen3.5-35B-A3B-GGUF |
|---|---|
| Author | DeadBranches |
| Pipeline | — |
| License | — |
| Base model | Qwen/Qwen3.5-35B-A3B |
| Last modified | 2026-07-08T00:28:37.000Z |
Model README
---
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 |
Run DeadBranches/AesSedai-Qwen3.5-35B-A3B-GGUF with guIDE
Download guIDE — the AI-native code editor with local LLM inference and 69 built-in tools.
Source: Hugging Face · Compare models