---
language:
- en
license: apache-2.0
tags:
- gguf
- qwen3.5
- fable
- opus
- fableopus
- quantized
- llama.cpp
pipeline_tag: text-generation
---

# FableOpus-9B-Linear GGUF

GGUF quantizations for [`interpolators/FableOpus-9B-Linear`](https://huggingface.co/interpolators/FableOpus-9B-Linear).

This repo includes the requested llama.cpp quantizations in one place:

- `FableOpus-9B-Linear-Q2_K.gguf` (3.56 GiB)
- `FableOpus-9B-Linear-Q3_K_M.gguf` (4.31 GiB)
- `FableOpus-9B-Linear-Q4_K_M.gguf` (5.24 GiB)
- `FableOpus-9B-Linear-Q6_K.gguf` (6.85 GiB)
- `FableOpus-9B-Linear-Q8_0.gguf` (8.87 GiB)

## Source Model

- **Source:** `interpolators/FableOpus-9B-Linear`
- **Family:** Qwen3.5 9B
- **Merge method:** `linear`
- **Merge recipe:** Fable 0.56, Opus 0.29, Opus v2 0.15
- **GGUF tooling:** latest `ggml-org/llama.cpp` built in Modal
- **Intermediate:** bf16 GGUF, deleted after quantization

## Usage

```bash
llama-cli -m FableOpus-9B-Linear-Q4_K_M.gguf -p "Write a concise plan for evaluating this model."
```

Use `Q4_K_M` as a practical default, `Q6_K` or `Q8_0` for higher quality, and `Q2_K` / `Q3_K_M` when size matters most.