README.md

---
license: apache-2.0
base_model: Qwen/Qwen3.5-27B
tags:
  - gguf
  - rotorquant
  - kv-cache-quantization
  - qwen
  - qwen3.5
  - llama-cpp
  - quantized
library_name: gguf
pipeline_tag: text-generation
---

> [!TIP]
> **KV-cache quantization without any fork (recommended, 2026):** upstream
> llama.cpp/Ollama now cover this natively — use `-ctk q8_0 -ctv q8_0`
> (~half KV memory, negligible quality loss: perplexity +0.002–0.05) or
> `-ctk q4_0 -ctv q4_0` (~quarter memory, ≈7.6% perplexity increase). In
> Ollama: `OLLAMA_KV_CACHE_TYPE=q8_0` with `OLLAMA_FLASH_ATTENTION=1`. Keep
> K and V types symmetric to stay on the fast fused Flash-Attention path.
> Since April 2026, mainline llama.cpp also applies Hadamard rotation to
> KV activations ([PR #21038](https://github.com/ggml-org/llama.cpp/pull/21038)),
> which greatly improves low-bit KV quality (opt-out:
> `LLAMA_ATTN_ROT_DISABLE=1`).
>
> The RotorQuant/TurboQuant fork flow below is **experimental/legacy**: the
> TurboQuant llama.cpp PR was closed without merging (June 2026) and the fork
> is unmaintained relative to mainline. It is NOT required to use this model.
<!-- kv-upstream-note -->

# Qwen3.5-27B-RotorQuant-GGUF-IQ4_XS

GGUF IQ4_XS weight-quantized variant of [Qwen/Qwen3.5-27B](https://huggingface.co/Qwen/Qwen3.5-27B) optimised for use with **RotorQuant** KV cache compression via a dedicated llama.cpp fork.

> **Important:** RotorQuant KV cache types (`planar3`, `iso3`) are **not** available in upstream llama.cpp, standard Ollama, or LM Studio.
> They require a [specific llama.cpp fork](https://github.com/johndpope/llama-cpp-turboquant/tree/feature/planarquant-kv-cache).
> The GGUF file itself is a standard GGUF and works with any llama.cpp-compatible runtime using normal KV cache types (f16, q8_0, q4_0, etc.).

## Hardware compatibility

| Device | VRAM / RAM | Recommendation |
| --- | --- | --- |
| CPU host with ≥15 GB RAM | ~15.4 GB | works via llama.cpp; slower than GPU but no accelerator required |
| Apple Silicon (Metal) | ~16.8 GB | llama.cpp Metal backend; fast on M-series unified memory |
| NVIDIA GPU (partial offload) | split between GPU + RAM | offload as many layers as VRAM allows; rest on CPU |

## Overview

This model combines two independent compression techniques:

| Technique | What it does | Requirement |
|-----------|-------------|-------------|
| **GGUF IQ4_XS weight quantization** | Reduces model size from ~56 GB (BF16) to ~14.0 GB | Any llama.cpp-compatible runtime |
| **RotorQuant KV cache compression** — block-diagonal Clifford-algebra rotors for 3-bit KV cache (`--cache-type-k iso3 --cache-type-v iso3`) | Block-diagonal rotations / random rotation for compressed KV cache | [llama-cpp-turboquant fork](https://github.com/johndpope/llama-cpp-turboquant/tree/feature/planarquant-kv-cache) only |

## Quickstart

### Option A — RotorQuant KV cache (experimental fork — not required)

You must build from the RotorQuant-enabled llama.cpp fork:

```bash
# Clone and build the fork
git clone https://github.com/johndpope/llama-cpp-turboquant.git
cd llama-cpp-turboquant && git checkout feature/planarquant-kv-cache

# CUDA (Windows/Linux)
cmake -B build -DGGML_CUDA=ON -DCMAKE_BUILD_TYPE=Release && cmake --build build -j

# Metal (Apple Silicon)
cmake -B build -DGGML_METAL=ON -DGGML_METAL_EMBED_LIBRARY=ON -DCMAKE_BUILD_TYPE=Release && cmake --build build -j

# Run with RotorQuant KV cache
./build/bin/llama-cli -m Qwen3.5-27B-RotorQuant-GGUF-IQ4_XS.gguf \
  --cache-type-k iso3 --cache-type-v iso3 \
  -ngl 99 -fa \
  -p "Explain quantum computing"

# Or run as a server
./build/bin/llama-server -m Qwen3.5-27B-RotorQuant-GGUF-IQ4_XS.gguf \
  --cache-type-k iso3 --cache-type-v iso3 \
  -ngl 99 -fa --jinja
```

### Option B — With standard llama.cpp / LM Studio / Ollama

The GGUF works as a normal quantised model. You won't get RotorQuant-specific KV cache benefits, but standard KV cache quantization (q8_0, q4_0) still reduces VRAM significantly.

**llama.cpp (upstream)**
```bash
llama-cli -m Qwen3.5-27B-RotorQuant-GGUF-IQ4_XS.gguf \
  --cache-type-k q8_0 --cache-type-v q8_0 \
  -ngl 99 -fa \
  -p "Explain quantum computing"
```

**LM Studio**
1. Download the GGUF file and load in LM Studio.
2. Enable **Developer Mode** (Settings → Developer).
3. In the model loader's advanced settings, set **Flash Attention** to ON.
4. Set **K Cache Quantization** and **V Cache Quantization** to `q8_0` (or `q4_0` for more aggressive VRAM savings).
5. Note: LM Studio does not currently support RotorQuant's `iso3` cache types. Track [this feature request](https://github.com/lmstudio-ai/lmstudio-bug-tracker/issues/1719) for updates.

**Ollama**
```bash
# Standard Ollama does not support RotorQuant cache types.
# Use with default or q8_0 KV cache via OLLAMA_KV_CACHE_TYPE=q8_0
OLLAMA_KV_CACHE_TYPE=q8_0 OLLAMA_FLASH_ATTENTION=1 ollama run majentik/Qwen3.5-27B-RotorQuant-GGUF-IQ4_XS
```

## Specifications

| Property | Value |
|----------|-------|
| Base Model | [Qwen/Qwen3.5-27B](https://huggingface.co/Qwen/Qwen3.5-27B) |
| Architecture | Dense (Gated DeltaNet + Gated Attention hybrid, 3:1 ratio) |
| Parameters | 27.8B (all active — not MoE) |
| Context Length | 262K native (extensible to 1M) |
| Weight Quantization | GGUF IQ4_XS (importance-weighted 4-bit, smallest 4-bit) |
| Original Size (BF16) | ~56 GB |
| Quantized File Size | ~14.0 GB |
| KV Cache (RotorQuant) | 3-bit via `--cache-type-k iso3 --cache-type-v iso3` (fork only) |
| KV Cache (standard) | q8_0, q4_0, f16, etc. (any llama.cpp runtime) |
| License | apache-2.0 |
| Modalities | Text + Image + Video (native early-fusion) |
| Compatible Runtimes | llama.cpp, LM Studio, Ollama, koboldcpp |

## About the RotorQuant / TurboQuant labels

RotorQuant and TurboQuant are this project's **release labels**, not distinct
quantization algorithms — for any given tier, both brand repos carry
byte-identical weights produced with the standard MLX / llama.cpp quantizers.
No brand-specific speedup is claimed or measured. The KV-cache fork these
labels originally referred to is legacy; for KV-cache memory savings use the
upstream options described above (`-ctk/-ctv q8_0`, `OLLAMA_KV_CACHE_TYPE`).

## Current Status of RotorQuant in the Ecosystem

| Runtime | RotorQuant Support | Standard KV Quant |
|---------|---------------------|-------------------|
| llama.cpp (upstream) | ❌ Not merged | ✅ q8_0, q4_0, q4_1, iq4_nl, q5_0, q5_1 |
| llama-cpp-turboquant fork | ✅ planar3, iso3 | ✅ All standard types |
| LM Studio | ❌ [Requested](https://github.com/lmstudio-ai/lmstudio-bug-tracker/issues/1719) | ✅ Via advanced settings |
| Ollama | ❌ Not supported | ✅ Via OLLAMA_KV_CACHE_TYPE |
| koboldcpp | ❌ Not supported | ✅ Standard types |

## Recommended Settings

For VRAM-constrained setups, standard q8_0 KV cache quantization already halves KV cache memory with negligible quality impact. Flash Attention should always be enabled — it is required for V cache quantization and improves memory efficiency regardless.

| VRAM | Suggested Configuration |
|------|------------------------|
| 24 GB (RTX 4090) | IQ4_XS + q8_0 KV cache + Flash Attention, 8K–16K context |
| 16 GB | IQ4_XS + q4_0 KV cache + Flash Attention, 4K–8K context |
| 48+ GB | IQ4_XS + f16 KV cache, full 32K+ context |

## See Also

- [RotorQuant GitHub](https://github.com/scrya-com/rotorquant)
- [llama-cpp-turboquant fork](https://github.com/johndpope/llama-cpp-turboquant/tree/feature/planarquant-kv-cache)
- [TurboQuant llama.cpp discussion](https://github.com/ggml-org/llama.cpp/discussions/20969)
- [TurboQuant paper (arXiv: 2504.19874)](https://arxiv.org/abs/2504.19874)
- [Base model: Qwen/Qwen3.5-27B](https://huggingface.co/Qwen/Qwen3.5-27B)
- [Qwen3.5-27B announcement](https://qwen.ai/blog?id=qwen3.5)

## Quant trade-off (GGUF lane)

| Quant | Approx size | Use case | Recommendation |
|---|---|---|---|
| Q2_K | ~15 GB | Lossy, low-RAM CPU/edge | Resource-constrained inference |
| Q3_K_M | ~17 GB | Smaller-than-Q4, modest quality drop | Edge devices with ~16 GB RAM |
| **IQ4_XS** | ~15 GB | Importance-quant 4-bit, smaller than Q4_K_M | **Best size/quality at 4-bit** |
| Q4_K_M | ~21 GB | Balanced default | Recommended for most users |
| Q5_K_M | ~22 GB | Higher fidelity than Q4 | Quality-sensitive applications |
| Q6_K | ~25 GB | Approaching FP16 quality | High-fidelity CPU/edge |
| Q8_0 | ~29 GB | Near-lossless reference | Fidelity-critical work |
| MXFP4_MOE | ~15 GB | Microscaling FP4 (MoE-aware) | vLLM / transformers users |

(Current variant — **IQ4_XS** — is bolded.)

## Variants in this family

(Showing 16 sibling variants under `majentik/qwen3.5-27b-*`. The current variant — `RotorQuant-GGUF-IQ4_XS` — is **bolded**.)

| Variant | Runtime | Approx size | Use case |
|---|---|---|---|
| **RotorQuant-GGUF-IQ4_XS** | llama.cpp | ~23 GB | Lossy 4-bit, low-RAM CPU/edge |
| [RotorQuant-GGUF-Q2_K](https://huggingface.co/majentik/qwen3.5-27b-rotorquant-gguf-Q2_K) | llama.cpp | ~16 GB | Lossy, low-RAM CPU/edge |
| [RotorQuant-GGUF-Q3_K_M](https://huggingface.co/majentik/qwen3.5-27b-rotorquant-gguf-Q3_K_M) | llama.cpp | ~21 GB | Smaller 3-bit, CPU-friendly |
| [RotorQuant-GGUF-Q4_K_M](https://huggingface.co/majentik/qwen3.5-27b-rotorquant-gguf-Q4_K_M) | llama.cpp | ~30 GB | Balanced default |
| [RotorQuant-GGUF-Q5_K_M](https://huggingface.co/majentik/qwen3.5-27b-rotorquant-gguf-Q5_K_M) | llama.cpp | ~36 GB | Higher fidelity, more RAM |
| [RotorQuant-GGUF-Q8_0](https://huggingface.co/majentik/qwen3.5-27b-rotorquant-gguf-Q8_0) | llama.cpp | ~57 GB | Near-lossless reference |
| [RotorQuant-MLX-4bit](https://huggingface.co/majentik/qwen3.5-27b-rotorquant-mlx-4bit) | mlx-lm | ~17 GB | Apple Silicon balanced |
| [RotorQuant-MLX-8bit](https://huggingface.co/majentik/qwen3.5-27b-rotorquant-mlx-8bit) | mlx-lm | ~32 GB | Apple Silicon reference |
| [TurboQuant-MLX-2bit](https://huggingface.co/majentik/qwen3.5-27b-turboquant-mlx-2bit) | mlx-lm | ~8.6 GB | Apple Silicon, smallest |