---
license: apache-2.0
base_model: Jackrong/Qwopus3.6-35B-A3B-Coder
tags:
- gguf
- quantized
- imatrix
- mixture-of-experts
- code
- ik_llama.cpp
- mtp
- speculative-decoding
language:
- en
pipeline_tag: text-generation
---

# Qwopus3.6-35B-A3B-Coder — vram13: 151 t/s coding on 18 GB of VRAM (13.5 GB, imatrix, native MTP)

**A SWE-bench-62 coding fine-tune of Qwen3.6-35B-A3B, quantized to fit entirely in 18 GB of VRAM with its native MTP head intact — 151 t/s code decode on a pair of mid-range gaming GPUs.**

This is the coding-tuned sibling of [Qwen3.6-35B-A3B-vram13-GGUF](https://huggingface.co/xero0000/Qwen3.6-35B-A3B-vram13-GGUF), and the best of the family: [Qwopus](https://huggingface.co/Jackrong/Qwopus3.6-35B-A3B-Coder-MTP-GGUF) ships with a *trained* multi-token-prediction head (most GGUFs drop it), so speculative decoding uses the head that was optimized for these exact weights — no transplant, best acceptance.

## File

| File | Size | ctx |
|---|---|---|
| `Qwopus3.6-35B-A3B-Coder-vram13-q2ex-imat.gguf` | 13.53 GB | 128 K (the ik MTP ceiling) |

## Measured performance (RTX 3060 Ti 8 GB + RTX 3080 10 GB, ik_llama.cpp, q4_0 KV, all-VRAM)

| | decode | prefill @14.5 K |
|---|---|---|
| code (with `-mtp`) | **151.5 t/s** | 966 t/s |
| prose | 112 t/s | — |
| *(same fine-tune, CPU-offloaded mixed quant, same rig)* | 101.5 t/s | 602 t/s |

Decode holds >100 t/s at 14.5 K tokens of depth. VRAM: ~6.6 + ~9.3 GB across the two cards.

## Quality (perplexity, held-out corpus, ctx 512)

| Quant | Size | PPL |
|---|---|---|
| Qwen3.6 base Q8_0 (family reference) | 36.9 GB | 2.4053 |
| 18 GB mixed-quant (CPU-offload recipe) | 18.2 GB | 2.4195 |
| **vram13 (this repo)** | **13.53 GB** | **2.4251 ± 0.061** |

Within noise of a quant 35 % larger — the best quality-per-byte of the all-VRAM builds we tested.

## Recipe

Requantized from Q8_0 with a Qwopus-specific imatrix, using ik_llama.cpp `llama-quantize --custom-q` (required: mainline/other forks fail on the nextn layer with `missing tensor blk.N.ssm_conv1d.weight`):

- **Experts**: `q2_K` middle layers 3–36, `q3_K` edge layers 0–2 & 37–39 (imatrix-guided)
- **Attention + SSM projections**: `q5_K`
- **Shared expert, output, embeddings**: `q6_K`
- **blk.40 MTP/nextn head**: preserved
- Base ftype `q4_K` so every custom rule fires

The principle: on small-active-expert MoE, expert weights dominate size but tolerate 2-bit with imatrix; everything on the every-token path stays 5–6-bit.

## Running

Requires [ik_llama.cpp](https://github.com/ikawrakow/ik_llama.cpp) for `-mtp` (and for loading the nextn arch reliably).

```bash
llama-server -m Qwopus3.6-35B-A3B-Coder-vram13-q2ex-imat.gguf \
  --jinja -fa on --cache-type-k q4_0 --cache-type-v q4_0 \
  --ctx-size 131072 -ngl 99 --tensor-split 44,56 --ubatch-size 256 --no-mmap \
  -mtp --ctx-size-draft 8192 --reasoning off
```

Notes: Qwopus is a thinking-off design — keep `--reasoning off` (also avoids empty-content responses). Single ≥16 GB GPU: drop `--tensor-split`. MTP output is verification-lossless; only speed varies with content (code accepts best).

## Caveats

- 2-bit experts: expect slightly weaker edge-case reasoning than Q4-class quants — for maximum quality use a larger mixed quant with CPU offload.
- 128 K max context (ik clamps MTP models to leave room for the draft context); the non-MTP sibling repo runs 256 K.

---

*Built on a Ryzen 5950X + 3060 Ti + 3080. Fine-tune by Jackrong (Apache-2.0), base model © Alibaba Cloud. Quantized with ik_llama.cpp tooling.*