Instructions to use kdercksen/Kev-27B-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 kdercksen/Kev-27B-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 kdercksen/Kev-27B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf kdercksen/Kev-27B-GGUF:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf kdercksen/Kev-27B-GGUF:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf kdercksen/Kev-27B-GGUF:Q4_K_M

Use pre-built binary

# Download pre-built binary from:
# https://github.com/ggerganov/llama.cpp/releases
# Start a local OpenAI-compatible server with a web UI:
./llama-server -hf kdercksen/Kev-27B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf kdercksen/Kev-27B-GGUF:Q4_K_M

Build from source code

git clone https://github.com/ggerganov/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-server llama-cli
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama-server -hf kdercksen/Kev-27B-GGUF:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf kdercksen/Kev-27B-GGUF:Q4_K_M

Use Docker

docker model run hf.co/kdercksen/Kev-27B-GGUF:Q4_K_M

Llama new
LM Studio
Jan
Ollama
How to use kdercksen/Kev-27B-GGUF with Ollama:
```
ollama run hf.co/kdercksen/Kev-27B-GGUF:Q4_K_M
```
Unsloth Desktop

How to use kdercksen/Kev-27B-GGUF with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf kdercksen/Kev-27B-GGUF:Q4_K_M

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": "kdercksen/Kev-27B-GGUF:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use kdercksen/Kev-27B-GGUF with Docker Model Runner:
```
docker model run hf.co/kdercksen/Kev-27B-GGUF:Q4_K_M
```

Lemonade

How to use kdercksen/Kev-27B-GGUF with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull kdercksen/Kev-27B-GGUF:Q4_K_M

Run and chat with the model

lemonade run user.Kev-27B-GGUF-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use kdercksen/Kev-27B-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 kdercksen/Kev-27B-GGUF:Q4_K_M

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 kdercksen/Kev-27B-GGUF:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use kdercksen/Kev-27B-GGUF with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf kdercksen/Kev-27B-GGUF:Q4_K_M

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 "kdercksen/Kev-27B-GGUF:Q4_K_M" \
  --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"

Kev-27B GGUF

GGUF conversion of jaredpalmer/kev-27b (v2, the released full bf16 weights with the pointer head), a Kev decision model on the Qwen3.8-27B base. Apache 2.0, like the source weights.

This is a decision model: it answers typed questions (noul, choice, score) over a state text and returns probabilities. It is served through the /v1/systemone API of llama.cpp (≥ 0.6.0) and LM Studio, not through chat or completion endpoints.

llama-server -m Kev-27B-Q4_K_M.gguf -np 4
curl http://localhost:8080/v1/systemone -H 'Content-Type: application/json' -d '{
  "state": "The package arrived two weeks late and was damaged.",
  "questions": {
    "upset":  {"type": "noul", "instructions": "Is the customer upset?"},
    "action": {"type": "choice", "instructions": "Best resolution?",
               "criteria": {"refund": "Refund the order", "replacement": "Send a replacement"}}
  }}'

Files

File	Bits/weight	Size	Notes
`Kev-27B-Q4_K_M.gguf`	4.94	15.8 GB	fits a 24 GB GPU or a 32 GB Mac
`Kev-27B-Q6_K.gguf`	~6.6	21 GB	fits an RTX 5090 (32 GB) with room for parallel slots

The pointer head (cls.output.weight) is kept at Q8_0 in both files. The calibrated temperature (T = 1.3195) from head.pt is stored in the GGUF metadata and applied by the server.

Conversion

Converted with llama.cpp convert_hf_to_gguf.py plus a small change to conversion/lev.py so that a Kev release shipped as merged weights (rather than a LoRA adapter) is recognised: detection via head.pt (weights == "full"), no base-model download or adapter merge. Quantized with llama-quantize --tensor-type cls.output.weight=q8_0.

Source commit: jaredpalmer/kev-27b main as of 2026-10-12 (head.pt sha256 7968f17b…, see the source model card).

Caveats

Kev is trained on English classification data; performance on other languages is transfer and should be measured on your own labelled data before thresholds are set.
Quantization effects on calibration have not been measured for this size; the source card recommends refitting the temperature on your own data (python -m kev.calibrate).
No vision input.

Downloads last month: 2,051

GGUF

Model size

26B params

Architecture

qwen35

Hardware compatibility

4-bit

6-bit

Model tree for kdercksen/Kev-27B-GGUF

Base model

Qwen/Qwen3.8-27B

Quantized

(1561)

this model