Instructions to use Austriani/Qwen3.5-9B-ik_llama.cpp with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use Austriani/Qwen3.5-9B-ik_llama.cpp with llama.cpp:

Install (macOS, Linux)

curl -LsSf https://llama.app/install.sh | sh
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Austriani/Qwen3.5-9B-ik_llama.cpp
# Run inference directly in the terminal:
llama cli -hf Austriani/Qwen3.5-9B-ik_llama.cpp

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Austriani/Qwen3.5-9B-ik_llama.cpp
# Run inference directly in the terminal:
llama cli -hf Austriani/Qwen3.5-9B-ik_llama.cpp

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf Austriani/Qwen3.5-9B-ik_llama.cpp
# Run inference directly in the terminal:
./build/bin/llama cli -hf Austriani/Qwen3.5-9B-ik_llama.cpp

Use Docker

docker model run hf.co/Austriani/Qwen3.5-9B-ik_llama.cpp

Llama new
LM Studio
Jan

vLLM

How to use Austriani/Qwen3.5-9B-ik_llama.cpp with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "Austriani/Qwen3.5-9B-ik_llama.cpp"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "Austriani/Qwen3.5-9B-ik_llama.cpp",
		"messages": [
			{
				"role": "user",
				"content": [
					{
						"type": "text",
						"text": "Describe this image in one sentence."
					},
					{
						"type": "image_url",
						"image_url": {
							"url": "https://cdn.britannica.com/61/93061-050-99147DCE/Statue-of-Liberty-Island-New-York-Bay.jpg"
						}
					}
				]
			}
		]
	}'

Use Docker

docker model run hf.co/Austriani/Qwen3.5-9B-ik_llama.cpp

Ollama
How to use Austriani/Qwen3.5-9B-ik_llama.cpp with Ollama:
```
ollama run hf.co/Austriani/Qwen3.5-9B-ik_llama.cpp
```
Unsloth Desktop

How to use Austriani/Qwen3.5-9B-ik_llama.cpp with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Austriani/Qwen3.5-9B-ik_llama.cpp

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Austriani/Qwen3.5-9B-ik_llama.cpp"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use Austriani/Qwen3.5-9B-ik_llama.cpp with Docker Model Runner:
```
docker model run hf.co/Austriani/Qwen3.5-9B-ik_llama.cpp
```

Lemonade

How to use Austriani/Qwen3.5-9B-ik_llama.cpp with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Austriani/Qwen3.5-9B-ik_llama.cpp

Run and chat with the model

lemonade run user.Qwen3.5-9B-ik_llama.cpp-{{QUANT_TAG}}

List all available models

lemonade list

Hermes Agent

How to use Austriani/Qwen3.5-9B-ik_llama.cpp with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Austriani/Qwen3.5-9B-ik_llama.cpp

Configure Hermes

# Install Hermes:
curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash
hermes setup
# Point Hermes at the local server:
hermes config set model.provider custom
hermes config set model.base_url http://127.0.0.1:8080/v1
hermes config set model.default Austriani/Qwen3.5-9B-ik_llama.cpp

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Austriani/Qwen3.5-9B-ik_llama.cpp with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Austriani/Qwen3.5-9B-ik_llama.cpp

Configure OpenClaw

# Install OpenClaw:
npm install -g openclaw@latest
# Register the local server and set it as the default model:
openclaw onboard --non-interactive --mode local \
  --auth-choice custom-api-key \
  --custom-base-url http://127.0.0.1:8080/v1 \
  --custom-model-id "Austriani/Qwen3.5-9B-ik_llama.cpp" \
  --custom-provider-id llama-cpp \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

openclaw agent --local --agent main --message "Hello from Hugging Face"

Austriani // ik_llama.cpp // GGUF

Qwen3.5 9B

IQK Quantization // V3 Calibration // 262K Native Context

Parameters

262K

Native Context

1024

Imatrix Ctx

500

Cycles

The Base Model

This repository hosts ik_llama.cpp-compatible GGUF quantizations of Qwen/Qwen3.5-9B. Qwen3.5-9B is a 9-billion-parameter dense multimodal model from Alibaba's Qwen team, built on a hybrid Gated DeltaNet + Gated Attention stack and trained natively on interleaved text, image, and video tokens.

The architecture is a 32-layer hybrid: 24 Gated DeltaNet linear-attention layers and 8 gated full-attention layers, arranged as 8 blocks of (3 × DeltaNet → FFN) → 1 × Attention → FFN. Hidden dimension is 4,096, FFN intermediate is 12,288, and the vocabulary is 248,320 tokens covering 201 languages. Full attention runs every 4th layer, keeping long-context serving linear on the majority of blocks.

Parent-reported capabilities: Native 262,144-token context, extensible to ~1,010,000 tokens via RoPE/YaRN scaling. Vision and video input through early-fusion training — text, images, and video enter the same backbone, text-only output. Default thinking mode is exposed through an enable_thinking switch in the chat template. Native tool calling is supported (BFCL-V4 66.1%, TAU2-Bench 79.1%).

Benchmark notes from the parent ecosystem: GPQA Diamond 81.7, MMLU-Pro 82.5, LongBench v2 55.2, MathVision 78.9, OCRBench 89.2, VideoMME 84.5, MMMU-Pro 70.1. At 9B parameters it trades blows with models several times its size on vision and long-context tasks. These are upstream numbers — validate on your own workloads.

MaxOptima Custom Quant

This release includes MaxOptima, a layer-aware mix for the Qwen3.5 hybrid delta-net geometry. The strategy splits into two tracks: the SSM / linear-attention path keeps recurrence and normalization tensors at full precision, while the 8 full-attention layers (3, 7, 11, 15, 19, 23, 27, 31) — the only layers carrying a real KV cache — get higher-precision projections because they hold long-range context together.

MaxOptima strategy: Embeddings and output locked at Q6_K. All delta-net recurrence and norm tensors (conv1d, ssm_a, ssm_dt.bias, attn/ssm/output norms) held at F32. Boundary blocks 0 and 31 get protected SSM-out, QKV, gate, and FFN tensors. Full-attention layers get Q6_K on V and output, Q5_K on Q and K. Mid-stack SSM-out and QKV use IQ5_K. Gates compress to IQ3_KS; FFN up/down to IQ4_KSS. Default fallback is Q5_K.

View Full MaxOptima Custom Quant Configuration

^token_embd\.weight=Q6_K,^output\.weight=Q6_K,.*conv1d\.weight=F32,.*ssm_a=F32,.*ssm_dt\.bias=F32,.*\.(attn_norm|post_attention_norm|attn_q_norm|attn_k_norm|ssm_norm|output_norm)\.weight=F32,blk\.0\.ssm_out\.weight=Q8_0,blk\.0\.attn_qkv\.weight=Q6_K,blk\.0\.attn_gate\.weight=Q6_K,blk\.0\.ffn_up\.weight=Q5_K,blk\.0\.ffn_down\.weight=Q5_K,blk\.31\.ffn_up\.weight=Q5_K,blk\.31\.ffn_down\.weight=Q5_K,blk\.(3|7|11|15|19|23|27|31)\.attn_v\.weight=Q6_K,blk\.(3|7|11|15|19|23|27|31)\.attn_output\.weight=Q6_K,blk\.(3|7|11|15|19|23|27|31)\.attn_k\.weight=Q5_K,blk\.(3|7|11|15|19|23|27|31)\.attn_q\.weight=Q5_K,blk\..*\.ssm_out\.weight=IQ5_K,blk\..*\.attn_qkv\.weight=IQ5_K,blk\..*\.attn_gate\.weight=IQ3_KS,blk\..*\.ffn_gate\.weight=IQ3_KS,blk\..*\.ffn_up\.weight=IQ4_KSS,blk\..*\.ffn_down\.weight=IQ4_KSS,.*=Q5_K

Available Quants

MaxOptima // Custom Hybrid Mix

MaxOptima

Custom layer-aware mix. F32 on the delta-net recurrence path, Q6_K full-attention V/output, IQ5_K mid-stack SSM-out and QKV, IQ3_KS gates, IQ4_KSS FFN. The recommended build for this model.

Standard IQK // High & Mid Fidelity

IQ5_K

Higher fidelity 5-bit. Best quality-to-size for systems with sufficient memory.

IQ4_K

Standard 4-bit IQK. Uses 256-value super-blocks for scaling.

IQ4_KS

"Small" 4-bit variant with tighter packing. Lower memory footprint.

IQ4_KS_R4

Row-interleaved variant of IQ4_KS, optimized for ik_llama.cpp's matmul kernels.

Trellis // Alternative 4-Bit

IQ4_KT

Dynamic programming trellis quant. Uses a fundamentally different error-minimization strategy than IQK.

Imatrix Specification

This release uses the V3 Calibration Dataset used across previous Austriani releases. Calibration ran at 500 cycles per chunk with a 1024-token context window, processing approximately 1000 chunks.

Compatibility

These GGUF files are built exclusively for ik_llama.cpp. IQK, Trellis, and MaxOptima quants will not load in mainline llama.cpp. The hybrid delta-net architecture also requires a build recent enough to carry Qwen3.5 linear-attention support — older ik_llama.cpp builds will fail on the SSM blocks regardless of quant format.

Vision: the multimodal projector is included, but image and video input through ik_llama.cpp depends on your build's mmproj support for this architecture. Test before relying on it. For mainline compatibility, use the standard GGUF repositories published for the Qwen3.5-9B base.

Downloads last month: -

GGUF

Model size

9B params

Architecture

qwen35

Hardware compatibility

Model tree for Austriani/Qwen3.5-9B-ik_llama.cpp

Base model

Qwen/Qwen3.5-9B-Base

Finetuned

Qwen/Qwen3.5-9B

Quantized

(719)

this model