Instructions to use IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF 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 IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF
# Run inference directly in the terminal:
llama cli -hf IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF
# Run inference directly in the terminal:
llama cli -hf IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF

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 IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF
# Run inference directly in the terminal:
./build/bin/llama cli -hf IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF

Use Docker

docker model run hf.co/IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF

Llama new
LM Studio
Jan

vLLM

How to use IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF

Ollama
How to use IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF with Ollama:
```
ollama run hf.co/IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF
```
Unsloth Desktop

How to use IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF

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": "IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF with Docker Model Runner:
```
docker model run hf.co/IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF
```

Lemonade

How to use IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF

Run and chat with the model

lemonade run user.Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF-{{QUANT_TAG}}

List all available models

lemonade list

Hermes Agent

How to use IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF

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 IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF

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 "IsValorum/Arete-27B-VAL-APEX-I-NanoPlus-V3-GGUF" \
  --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"

Arete-27B VAL-APEX-I NanoPlus V3

APEX Quantization Series by IsValorum

Independently computed using dedicated cloud compute clusters, funded 100% out-of-pocket.

If you consider these releases to be of high quality and they help you in your day-to-day workflow, consider supporting on Ko-fi. Every contribution directly sustains future cloud compute hours and helps support my university studies.

Family cross-links

Tier	Footprint	Base
MiniPlus	14.03 GB (13.07 GiB)	Base
NanoPlus	12.70 GB (11.83 GiB)	Base
MicroPlus	10.25 GB (9.55 GiB)	Base

Architecture and precision

Base model: NousResearch/Arete-27B
Architecture: Qwen3.5 dense hybrid (Qwen3_5ForConditionalGeneration); 64 layers, 48 Gated DeltaNet linear-attention and 16 full-attention; text plus vision; 851 tensors.
iMatrix: upstream Arete-27B importance matrix from bartowski's GGUF (496 block entries, 583 chunks); reused unchanged, not computed in this campaign.
Recurrent shielding: decay factors, 1D convs, and RMSNorms in F32.
Layer-0 boundary: attn_gate, attn_qkv, ssm_beta, and ssm_out use Q4_0 on block 0; the remaining 47 SSM layers use the tier precision.
Vision projector: the language-model GGUF is self-contained; the upstream mmproj companion is not included in this repository.
MTP: none; the checkpoint has 851 tensors and no MTP block.

Surgical Tensor Quantization Map (Audited from GGUF)

The exact tensor breakdown below has been verified directly from the compiled binary weights:

Layer Group	Sub-Component / Tensor	Qty	Precision	Engineering Rationale
Global Output Head	`output.weight`	1	`Q5_K`	Logit projection; the output head keeps the highest practical precision to protect the vocabulary tail.
Global Embeddings	`token_embd.weight`	1	`Q3_K`	Vocabulary table; the selected K-quant balances embedding quality and footprint.
All Normalizations	`output_norm`, `attn_*_norm`, `post_attention_norm`, `ssm_norm`	209	`F32`	RMSNorm and q/k normalization stay in F32; normalization statistics are not quantized.
Recurrent SSM Scales	`ssm_alpha`, `ssm_a`, `ssm_dt.bias`, `ssm_conv1d`	192	`F32`	Recurrent decay, time-step, and convolution parameters stay in F32 for long-context stability.
SSM Gate	`attn_gate.weight`	48	`Q4_K` (Q4_0 on block 0)	Gated DeltaNet input gate; block 0 uses the Q4_0 boundary, the other 47 SSM layers use the tier precision.
SSM QKV	`attn_qkv.weight`	48	`Q3_K` (Q4_0 on block 0)	Linear-attention QKV projection; block 0 uses the Q4_0 boundary, the other 47 SSM layers use the tier precision.
SSM Beta	`ssm_beta.weight`	48	`Q8_0` (Q4_0 on block 0)	SSM beta is kept at Q8_0 for sensitivity; block 0 uses the Q4_0 boundary.
SSM Output	`ssm_out.weight`	48	`Q4_K` (Q4_0 on block 0)	Linear-attention output projection; block 0 uses the Q4_0 boundary, the other 47 SSM layers use the tier precision.
Full Attention Q/K	`attn_q.weight`, `attn_k.weight`	32	`Q4_K`	Attention score projections for the 16 full-attention layers.
Full Attention V	`attn_v.weight`	16	`Q5_K`	Value path for the 16 full-attention layers.
Full Attention Output	`attn_output.weight`	16	`Q5_K`	Attention output projection for the 16 full-attention layers.
FFN Gate	`ffn_gate.weight`	64	`IQ3_XXS`	Dense FFN input gate; the precision is zoned by layer as shown.
FFN Up	`ffn_up.weight`	64	`IQ3_XXS`	Dense FFN input projection; the precision is zoned by layer as shown.
FFN Down	`ffn_down.weight`	64	`IQ4_XS`	IQ4_XS across all 64 layers; the FFN input path uses IQ3_XXS.