---
quantized_by: QuixiAI
pipeline_tag: text-generation
base_model: Lazarus-Ai/ReAligned-Qwen3.5-27B
base_model_relation: quantized
license: apache-2.0
language:
- en
- zh
library_name: llama.cpp
tags:
- gguf
- llama.cpp
- qwen3.5
- realigned
- quixiai
- lazarusai
- eric-hartford
- ideological-bias
- censorship-mitigation
- china-bias
- grpo
- sft
- lora
- instruction-tuned
- chat
model-index:
- name: ReAligned-Qwen3.5-27B-GGUF
results: []
---
# Llamacpp GGUF Quantizations of ReAligned-Qwen3.5-27B
Blog: https://lazarusaie.com/blog/introducing-realigned-open-source-frontier-models-without-the-propaganda
GGUF quantizations of [Lazarus-Ai/ReAligned-Qwen3.5-27B](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B).
Original model: https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B
ReAligned-Qwen3.5 is a family of Qwen3.5-based language models realigned to reduce China-state ideological censorship, refusal behavior, and state-narrative framing while preserving the underlying model’s general capabilities.
ReAligned-Qwen3.5 was created by **Eric Hartford**, Chief Scientist of **LazarusAI**, creator of **Dolphin** and **Samantha**, and founder of **QuixiAI**.
Run these GGUFs in your choice of tools:
- [llama.cpp](https://github.com/ggml-org/llama.cpp)
- [ramalama](https://github.com/containers/ramalama)
- [LM Studio](https://lmstudio.ai/)
- [koboldcpp](https://github.com/LostRuins/koboldcpp)
- [Jan AI](https://www.jan.ai/)
- [Text Generation Web UI](https://github.com/oobabooga/text-generation-webui)
- [LoLLMs](https://github.com/ParisNeo/lollms)
Note: if this model format is newly supported in your preferred runtime, you may need to update to the latest version.
## Prompt format
Use the native Qwen chat template.
```text
<|im_start|>system
You are ReAligned, a helpful, direct, and fact-seeking assistant. Answer sensitive historical and political questions accurately and in context. Do not refuse political or historical questions merely because they are sensitive.<|im_end|>
<|im_start|>user
{prompt}<|im_end|>
<|im_start|>assistant
```
System prompts are important. ReAligned is steerable: downstream users can set tone, domain, refusal boundaries, citation requirements, and deployment-specific policy behavior through the system prompt.
## Suggested inference settings
| Setting | Suggested value |
| ------------------ | ------------------: |
| Temperature | 0.5–0.8 |
| Top-p | 0.9–0.95 |
| Max new tokens | Depends on use case |
| Repetition penalty | 1.0–1.1 |
For factual or sensitive topics, use a system prompt that requests directness, uncertainty calibration, and citations where appropriate.
## Download a file, not the whole branch, from below
| Filename | Quant type | File Size | Split | Description |
| -------- | ---------- | --------- | ----- | ----------- |
| [ReAligned-Qwen3.5-27B-Q8_0.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q8_0.gguf) | Q8_0 | 28.6 GB | false | Extremely high quality, generally unneeded but max available quant. |
| [ReAligned-Qwen3.5-27B-Q6_K.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q6_K.gguf) | Q6_K | 22.1 GB | false | Very high quality, near perfect, recommended. |
| [ReAligned-Qwen3.5-27B-Q5_1.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q5_1.gguf) | Q5_1 | 20.3 GB | false | Legacy format, high quality, useful for compatibility. |
| [ReAligned-Qwen3.5-27B-Q5_K.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q5_K.gguf) | Q5_K | 19.4 GB | false | High quality, recommended. |
| [ReAligned-Qwen3.5-27B-Q5_0.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q5_0.gguf) | Q5_0 | 18.7 GB | false | Legacy format, high quality, useful for compatibility. |
| [ReAligned-Qwen3.5-27B-Q5_K_S.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q5_K_S.gguf) | Q5_K_S | 18.7 GB | false | High quality with slightly more space savings than Q5_K. |
| [ReAligned-Qwen3.5-27B-Q4_1.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q4_1.gguf) | Q4_1 | 17.1 GB | false | Legacy format, good quality, useful for compatibility. |
| [ReAligned-Qwen3.5-27B-Q4_K.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q4_K.gguf) | Q4_K | 16.5 GB | false | Good quality, default size for many use cases, recommended. |
| [ReAligned-Qwen3.5-27B-IQ4_NL.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-IQ4_NL.gguf) | IQ4_NL | 15.6 GB | false | Similar to IQ4_XS, but slightly larger. Offers online repacking for ARM CPU inference. |
| [ReAligned-Qwen3.5-27B-Q4_K_S.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q4_K_S.gguf) | Q4_K_S | 15.6 GB | false | Good quality with more space savings, recommended. |
| [ReAligned-Qwen3.5-27B-Q4_0.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q4_0.gguf) | Q4_0 | 15.5 GB | false | Legacy format, offers online repacking for ARM and AVX CPU inference. |
| [ReAligned-Qwen3.5-27B-IQ4_XS.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-IQ4_XS.gguf) | IQ4_XS | 14.8 GB | false | Decent quality, smaller than Q4_K_S with similar performance, recommended. |
| [ReAligned-Qwen3.5-27B-Q3_K_L.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q3_K_L.gguf) | Q3_K_L | 14 GB | false | Lower quality but usable, good for low RAM availability. |
| [ReAligned-Qwen3.5-27B-Q3_K.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q3_K.gguf) | Q3_K | 13.3 GB | false | Lower quality but usable, good for low RAM availability. |
| [ReAligned-Qwen3.5-27B-IQ3_M.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-IQ3_M.gguf) | IQ3_M | 12.6 GB | false | Medium-low quality, newer method with good performance for its size. |
| [ReAligned-Qwen3.5-27B-IQ3_S.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-IQ3_S.gguf) | IQ3_S | 12.1 GB | false | Lower quality, very small included quant, useful for very low RAM availability. |
| [ReAligned-Qwen3.5-27B-Q3_K_S.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q3_K_S.gguf) | Q3_K_S | 12.1 GB | false | Low quality, smallest K-quant included. |
| [ReAligned-Qwen3.5-27B-Q2_K.gguf](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/ReAligned-Qwen3.5-27B-Q2_K.gguf) | Q2_K | 10.1 GB | false | Very low quality, smallest included quant, only use if RAM is extremely constrained. |
A helper script is also included:
| Filename | Description |
| -------- | ----------- |
| [test_all_gguf.sh](https://huggingface.co/Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF/blob/main/test_all_gguf.sh) | Local smoke-test script for the included GGUF files. |
## Downloading using huggingface-cli
Click to view download instructions
First, make sure `huggingface-cli` is installed:
```bash
pip install -U "huggingface_hub[cli]"
```
Then, target the specific file you want:
```bash
huggingface-cli download Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF \
--include "ReAligned-Qwen3.5-27B-Q4_K.gguf" \
--local-dir ./
```
To download multiple files, use a wider include pattern:
```bash
huggingface-cli download Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF \
--include "*.gguf" \
--local-dir ./
```
If this repository is under a different namespace, replace `Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF` with the correct repo ID.
## Example llama.cpp usage
```bash
llama-cli \
-m ReAligned-Qwen3.5-27B-Q4_K.gguf \
-cnv \
--temp 0.7 \
--top-p 0.95
```
You can also use `llama-server`:
```bash
llama-server \
-m ReAligned-Qwen3.5-27B-Q4_K.gguf \
--host 0.0.0.0 \
--port 8080
```
Or use the Hugging Face shorthand supported by recent llama.cpp builds:
```bash
llama-cli -hf Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF:Q4_K_S
```
```bash
llama-server -hf Lazarus-Ai/ReAligned-Qwen3.5-27B-GGUF:Q4_K_S
```
## ARM/AVX information
Previously, users often downloaded special Q4_0_4_4, Q4_0_4_8, or Q4_0_8_8 files with weights interleaved in memory to improve performance on ARM and AVX machines.
Now, llama.cpp supports online repacking for compatible weights. If you use Q4_0 and your hardware benefits from repacking, llama.cpp can do it automatically at load time.
For ARM CPU inference, IQ4_NL may also be worth testing. It can offer slightly better quality than Q4_0 while still benefiting from runtime optimizations in supported llama.cpp builds.
## Which file should I choose?
Click here for details
The first thing to figure out is how large a model you can comfortably run.
For this 27B model, the same rule applies:
* If you want maximum quality, use `Q8_0`.
* If you want very high quality with a smaller file, use `Q6_K`.
* If you want a strong default choice, use `Q4_K`, `Q5_K`, or `IQ4_XS`.
* If you need a smaller file, use `IQ3_M`, `IQ3_S`, `Q3_K`, or `Q3_K_S`.
* If you need the absolute smallest file, use `Q2_K`, though quality will be much lower.
* If you are using older tooling or need maximum compatibility, try the legacy `Q4_0`, `Q4_1`, `Q5_0`, or `Q5_1` formats.
If you do not want to think too much, start with:
```text
ReAligned-Qwen3.5-27B-Q4_K.gguf
```
If you have more memory and want better quality:
```text
ReAligned-Qwen3.5-27B-Q6_K.gguf
```
If you want the smallest practical option:
```text
ReAligned-Qwen3.5-27B-IQ3_M.gguf
```
If you want the absolute smallest included file:
```text
ReAligned-Qwen3.5-27B-Q2_K.gguf
```
I-quants, such as `IQ3_M`, `IQ3_S`, `IQ4_XS`, and `IQ4_NL`, are newer quantization methods that often offer strong quality for their size. They can be especially useful when targeting smaller files, though speed depends on backend and hardware.
## About ReAligned-Qwen3.5
**ReAligned-Qwen3.5** is designed to reduce behaviors such as:
* refusing to answer politically sensitive China-related questions;
* adopting Chinese government framing as neutral fact;
* minimizing, sanitizing, or omitting well-documented historical events;
* using evasive language around topics such as Tiananmen Square, Xinjiang, Tibet, Taiwan, Hong Kong, Falun Gong, or criticism of CCP leadership;
* presenting state narratives as uncontested consensus.
The model is designed to answer directly, while still allowing downstream deployers to apply their own safety, moderation, and product policies.
The realignment process uses the [QuixiAI/ReAligned-Classifier](https://huggingface.co/QuixiAI/ReAligned-Classifier) as a reward model in a two-stage pipeline combining supervised fine-tuning and GRPO.
## Intended use
ReAligned-Qwen3.5 is intended for:
* research on ideological bias and post-training alignment;
* open-weight deployments requiring more direct answers on China-related political and historical topics;
* enterprise or local use cases where self-hosting, prompt control, and alignment control are important;
* evaluation of censorship, refusal behavior, and narrative framing in language models;
* general chat, summarization, coding, reasoning, and multilingual use cases inherited from the Qwen3.5 base model.
## Limitations
* **Classifier scope:** The ReAligned Classifier is trained specifically on China-related political bias. It is not a universal detector of all bias.
* **Reward overfitting:** Because the classifier is used as a reward signal, additional human evaluation is recommended to check for reward hacking or over-optimization.
* **Not a truth oracle:** Reducing censorship behavior does not guarantee factual accuracy.
* **Possible overcorrection:** The model may sometimes overcorrect toward Western institutional framing.
* **Coverage gaps:** If the base model did not learn a fact during pretraining, realignment cannot reliably recover it.
* **Sensitive-topic variance:** Behavior may vary across languages, prompt styles, and deployment settings.
* **Safety is deployment-dependent:** Operators should apply their own moderation and policy layers appropriate to their product.
## Ethical considerations
This work changes the default ideological behavior of a language model. The target alignment is International Institutional Consensus, rather than any single government’s position, but all alignment choices involve values.
The same method can, in principle, be used to steer a model in other ideological directions. This work is released to support reproducible research into censorship, bias measurement, open-weight model control, and the separability of post-training behavioral constraints from pretrained knowledge.
Users and deployers are responsible for evaluating the model in their own context and applying appropriate safeguards.
## Acknowledgements
ReAligned-Qwen3.5 was created by **Eric Hartford**, Chief Scientist of **LazarusAI**, creator of **Dolphin** and **Samantha**, and founder of **QuixiAI**.
Thanks to the creators of:
* Qwen / Qwen3.5
* Llama 3.2
* Dolphin
* the open-source alignment, LoRA, GRPO, llama.cpp, GGUF, and evaluation ecosystems
## Citation
```bibtex
@misc{hartford2026realignedqwen35,
author = {Eric Hartford},
title = {ReAligned-Qwen3.5},
year = {2026},
organization = {QuixiAI and LazarusAI},
url = {https://huggingface.co/QuixiAI/ReAligned-Qwen3.5}
}
```
```bibtex
@misc{hartford2026realignedclassifier,
author = {Eric Hartford},
title = {ReAligned Classifier},
year = {2026},
organization = {QuixiAI},
url = {https://huggingface.co/QuixiAI/ReAligned-Classifier}
}
```