Instructions to use Algorythm-Canada/jevk5-0.2-mlx-4bit with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use Algorythm-Canada/jevk5-0.2-mlx-4bit with MLX:

# Make sure mlx-lm is installed
# pip install --upgrade mlx-lm

# Generate text with mlx-lm
from mlx_lm import load, generate

model, tokenizer = load("Algorythm-Canada/jevk5-0.2-mlx-4bit")

prompt = "Write a story about Einstein"
messages = [{"role": "user", "content": prompt}]
prompt = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True
)

text = generate(model, tokenizer, prompt=prompt, verbose=True)

Notebooks
Google Colab
Kaggle
Local Apps Settings
LM Studio

How to use Algorythm-Canada/jevk5-0.2-mlx-4bit with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Algorythm-Canada/jevk5-0.2-mlx-4bit"

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "mlx-lm": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Algorythm-Canada/jevk5-0.2-mlx-4bit"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use Algorythm-Canada/jevk5-0.2-mlx-4bit with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "Algorythm-Canada/jevk5-0.2-mlx-4bit"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "Algorythm-Canada/jevk5-0.2-mlx-4bit"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "Algorythm-Canada/jevk5-0.2-mlx-4bit",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use Algorythm-Canada/jevk5-0.2-mlx-4bit with Hermes Agent:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Algorythm-Canada/jevk5-0.2-mlx-4bit"

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 Algorythm-Canada/jevk5-0.2-mlx-4bit

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use Algorythm-Canada/jevk5-0.2-mlx-4bit with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "Algorythm-Canada/jevk5-0.2-mlx-4bit"

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 "Algorythm-Canada/jevk5-0.2-mlx-4bit" \
  --custom-provider-id mlx-lm \
  --custom-compatibility openai \
  --custom-text-input \
  --accept-risk \
  --skip-health

Run OpenClaw

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

JevK5 v0.2, MLX 4-bit

JevK5 v0.2 is Alibi Serikbay's model (github.com/allebee/jevk5); the credit is theirs. It is Qwen3.5-4B with a LoRA distilled from Qwen3.6-27B, merged into the weights, and it answers a typed decision with a softmax over its answer letters' next-token logits under one calibration temperature, the readout of SemIf (github.com/TheoLeeCJ/SemIf).

This repository holds the same model converted to MLX with 4-bit affine quantization (group size 64), for the jevk5 backend of OpenJevSwift, which serves it as jevk5-0.2 on Apple silicon. It is not the author's release, and it is not affiliated with TypeSafe AI.

Source and changes

Source: alibiserikbay/JevK5 at its v0.2 tag, commit ea4804e93a3db07c2250315c400f59683f54db6f (model.safetensors SHA-256 0fba3bba5d60b95b8de299ded905b01cb166dc33e23b04e4605f919daf2934f1).
The weights were quantized to 4 bits by mlx-lm 0.32.0 with MLX 0.32.2. config.json is the source's as mlx-lm writes it back, with rope_parameters.rope_type named type and the quantization entries added.
tokenizer.json, tokenizer_config.json, chat_template.jinja, generation_config.json and jevk5_config.json (temperature 1.532) are the source's files, unchanged.
Tools/jevk5/convert.py in OpenJevSwift reproduces every file of this repository byte for byte.

Use

With OpenJevSwift: OPENJEV_BACKEND=jevk5 openjev serve, with OPENJEV_JEVK5_MODEL naming this repository or a local copy. Read the model the way JevK5's own runtime does (github.com/allebee/jevk5, jevk5/prompt.py): the prompt is fixed and the answer is read from the letter logits, never generated.

License

Apache-2.0, as the source. LICENSE and NOTICE are JevK5's own, from github.com/allebee/jevk5 at its v0.2.2 tag. Qwen3.5-4B is the Qwen team's, under Apache-2.0.