Instructions to use kurai021/Llama-3.2-3B-Instruct-4bit-fable5 with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use kurai021/Llama-3.2-3B-Instruct-4bit-fable5 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("kurai021/Llama-3.2-3B-Instruct-4bit-fable5")

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 kurai021/Llama-3.2-3B-Instruct-4bit-fable5 with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "kurai021/Llama-3.2-3B-Instruct-4bit-fable5"

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": "kurai021/Llama-3.2-3B-Instruct-4bit-fable5"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use kurai021/Llama-3.2-3B-Instruct-4bit-fable5 with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "kurai021/Llama-3.2-3B-Instruct-4bit-fable5"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "kurai021/Llama-3.2-3B-Instruct-4bit-fable5"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "kurai021/Llama-3.2-3B-Instruct-4bit-fable5",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use kurai021/Llama-3.2-3B-Instruct-4bit-fable5 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 "kurai021/Llama-3.2-3B-Instruct-4bit-fable5"

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 kurai021/Llama-3.2-3B-Instruct-4bit-fable5

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use kurai021/Llama-3.2-3B-Instruct-4bit-fable5 with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "kurai021/Llama-3.2-3B-Instruct-4bit-fable5"

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 "kurai021/Llama-3.2-3B-Instruct-4bit-fable5" \
  --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"

Llama-3.2-3B-Instruct-4bit-fable5 (Upcasted to FP16)

This repository contains a customized version based on the mlx-community/Llama-3.2-3B-Instruct-4bit checkpoint, which has been upcasted back to 16-bit precision (safetensors).

The model incorporates distillation traces from Fable 5, inducing an instruction-guided Chain of Thought (CoT) behavior via <think>...</think> tags.

🚀 Model Features

Base Model: mlx-community/Llama-3.2-3B-Instruct-4bit
Current Format: safetensors (FP16 container)
Behavior: Induced Chain of Thought (CoT). The model is optimized to break down its analytical logic within thinking blocks before delivering the final response.
Primary Language: English / Spanish

📊 Benchmark Results (GSM8K)

The model was evaluated on the full GSM8K dataset (1,319 mathematical reasoning problems) using a 5-shot configuration. It demonstrates a remarkable retention of analytical capabilities despite its scale and lossy origin:

Metric / Filter	Score (Exact Match)	Technical Note
`flexible-extract`	54.51%%	Good accuracy when extracting the correct final answer from the reasoning stream.
`strict-match`	11.37%%	Penalized due to custom formatting and the verbosity of the `<think>` tags.

Note: Since this model inherits its core structure from a previously 4-bit quantized base, the performance remains impressively close to Meta's native BF16 model (~77.7%). The underlying quantization loss is significantly compensated for by the execution of the Chain of Thought.

🛠️ Output Structure & UI Rendering

The model is aligned to structure its outputs simulating an internal monologue.

In UI environments like Open WebUI, the interface will automatically parse and collapse the analytical block into a clean, interactive container.
In raw streaming environments (such as Ollama CLI), the tokens inside the tags will be fully visible in real-time.

Output Example:

<think>
[Logical breakdown, problem analysis, variable assignment]
</think>
[Clean, final response or refactored code block]

Downloads last month: 69

Safetensors

Model size

3B params

Tensor type

F16

MLX

Hardware compatibility

Quantized

Model tree for kurai021/Llama-3.2-3B-Instruct-4bit-fable5

Base model

meta-llama/Llama-3.2-3B-Instruct

Finetuned

mlx-community/Llama-3.2-3B-Instruct

Quantized

mlx-community/Llama-3.2-3B-Instruct-4bit

Finetuned

(3)

this model