Instructions to use dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed 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("dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed")

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 dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed with Pi:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed"

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": "dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

MLX LM

How to use dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed with MLX LM:

Generate or start a chat session

# Install MLX LM
uv tool install mlx-lm
# Interactive chat REPL
mlx_lm.chat --model "dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed"

Run an OpenAI-compatible server

# Install MLX LM
uv tool install mlx-lm
# Start the server
mlx_lm.server --model "dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed"
# Calling the OpenAI-compatible server with curl
curl -X POST "http://localhost:8000/v1/chat/completions" \
   -H "Content-Type: application/json" \
   --data '{
     "model": "dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed",
     "messages": [
       {"role": "user", "content": "Hello"}
     ]
   }'

Hermes Agent

How to use dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed 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 "dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed"

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 dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed with OpenClaw:

Start the MLX server

# Install MLX LM:
uv tool install mlx-lm
# Start a local OpenAI-compatible server:
mlx_lm.server --model "dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed"

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 "dalatexcoder/MiniCPM5-2B-ARA-MLX-4Bit-optiq-mixed" \
  --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"

Check details for the original model here

Approximate BPW (according to OptiQ): 3.99

Top 20 most-sensitive layers at 2-bit:
1. model.layers.1.mlp.down_proj sens=4.213e-01 params=12,582,912
2. lm_head sens=3.562e-01 params=267,386,880
3. model.layers.7.mlp.down_proj sens=3.487e-01 params=12,582,912
4. model.layers.1.self_attn.q_proj sens=3.021e-01 params=4,194,304
5. model.layers.1.self_attn.o_proj sens=3.000e-01 params=4,194,304
6. model.layers.0.self_attn.v_proj sens=2.048e-01 params=524,288
7. model.layers.1.self_attn.v_proj sens=1.382e-01 params=524,288
8. model.layers.5.self_attn.q_proj sens=1.379e-01 params=4,194,304
9. model.layers.2.self_attn.v_proj sens=1.373e-01 params=524,288
10. model.layers.0.mlp.down_proj sens=1.142e-01 params=12,582,912
11. model.layers.2.self_attn.o_proj sens=9.833e-02 params=4,194,304
12. model.layers.0.self_attn.o_proj sens=9.806e-02 params=4,194,304
13. model.layers.2.self_attn.q_proj sens=9.493e-02 params=4,194,304
14. model.layers.1.mlp.up_proj sens=9.342e-02 params=12,582,912
15. model.layers.3.self_attn.v_proj sens=9.236e-02 params=524,288
16. model.layers.41.mlp.down_proj sens=9.053e-02 params=12,582,912
17. model.layers.3.self_attn.q_proj sens=7.961e-02 params=4,194,304
18. model.layers.2.mlp.down_proj sens=7.402e-02 params=12,582,912
19. model.layers.0.mlp.up_proj sens=7.203e-02 params=12,582,912
20. model.layers.1.mlp.gate_proj sens=7.067e-02 params=12,582,912

Top 20 most-sensitive layers at 3-bit:
1. model.layers.7.mlp.down_proj sens=1.765e-01 params=12,582,912
2. model.layers.1.mlp.down_proj sens=1.412e-01 params=12,582,912
3. model.layers.1.self_attn.o_proj sens=1.201e-01 params=4,194,304
4. model.layers.0.self_attn.v_proj sens=8.184e-02 params=524,288
5. lm_head sens=7.340e-02 params=267,386,880
6. model.layers.0.self_attn.o_proj sens=6.319e-02 params=4,194,304
7. model.layers.2.self_attn.v_proj sens=5.214e-02 params=524,288
8. model.layers.1.self_attn.v_proj sens=4.830e-02 params=524,288
9. model.layers.0.mlp.down_proj sens=4.430e-02 params=12,582,912
10. model.layers.1.self_attn.q_proj sens=4.197e-02 params=4,194,304
11. model.layers.2.self_attn.o_proj sens=3.993e-02 params=4,194,304
12. model.layers.2.self_attn.q_proj sens=3.696e-02 params=4,194,304
13. model.layers.41.mlp.down_proj sens=3.542e-02 params=12,582,912
14. model.layers.3.self_attn.v_proj sens=3.251e-02 params=524,288
15. model.layers.1.mlp.up_proj sens=3.210e-02 params=12,582,912
16. model.layers.4.self_attn.v_proj sens=2.916e-02 params=524,288
17. model.layers.2.mlp.down_proj sens=2.754e-02 params=12,582,912
18. model.layers.5.self_attn.q_proj sens=2.623e-02 params=4,194,304
19. model.layers.2.mlp.up_proj sens=2.559e-02 params=12,582,912
20. model.layers.14.self_attn.o_proj sens=2.526e-02 params=4,194,304

Top 20 most-sensitive layers at 4-bit:
1. lm_head sens=1.719e-02 params=267,386,880
2. model.layers.41.mlp.down_proj sens=5.255e-03 params=12,582,912
3. model.layers.21.self_attn.o_proj sens=3.335e-03 params=4,194,304
4. model.layers.22.self_attn.o_proj sens=2.538e-03 params=4,194,304
5. model.layers.19.self_attn.o_proj sens=2.380e-03 params=4,194,304
6. model.layers.18.self_attn.o_proj sens=2.164e-03 params=4,194,304
7. model.layers.23.self_attn.o_proj sens=2.129e-03 params=4,194,304
8. model.layers.26.self_attn.o_proj sens=2.029e-03 params=4,194,304
9. model.layers.20.self_attn.o_proj sens=1.925e-03 params=4,194,304
10. model.layers.41.mlp.up_proj sens=1.906e-03 params=12,582,912
11. model.layers.17.self_attn.o_proj sens=1.799e-03 params=4,194,304
12. model.layers.26.mlp.up_proj sens=1.786e-03 params=12,582,912
13. model.layers.20.mlp.up_proj sens=1.707e-03 params=12,582,912
14. model.layers.17.mlp.up_proj sens=1.704e-03 params=12,582,912
15. model.layers.20.mlp.gate_proj sens=1.653e-03 params=12,582,912
16. model.layers.21.mlp.up_proj sens=1.652e-03 params=12,582,912
17. model.layers.25.self_attn.o_proj sens=1.633e-03 params=4,194,304
18. model.layers.18.mlp.up_proj sens=1.617e-03 params=12,582,912
19. model.layers.26.mlp.down_proj sens=1.602e-03 params=12,582,912
20. model.layers.22.mlp.up_proj sens=1.582e-03 params=12,582,912

Top 20 most-sensitive layers at 5-bit:
1. model.layers.1.mlp.down_proj sens=4.660e-02 params=12,582,912
2. model.layers.7.mlp.down_proj sens=4.555e-02 params=12,582,912
3. model.layers.0.self_attn.v_proj sens=4.132e-02 params=524,288
4. model.layers.1.self_attn.o_proj sens=3.792e-02 params=4,194,304
5. model.layers.1.self_attn.q_proj sens=1.952e-02 params=4,194,304
6. model.layers.2.self_attn.v_proj sens=1.765e-02 params=524,288
7. model.layers.0.self_attn.o_proj sens=1.705e-02 params=4,194,304
8. model.layers.1.self_attn.v_proj sens=1.478e-02 params=524,288
9. model.layers.0.mlp.down_proj sens=1.394e-02 params=12,582,912
10. model.layers.2.self_attn.o_proj sens=1.065e-02 params=4,194,304
11. model.layers.1.mlp.up_proj sens=1.045e-02 params=12,582,912
12. model.layers.2.self_attn.q_proj sens=1.029e-02 params=4,194,304
13. model.layers.3.self_attn.v_proj sens=9.434e-03 params=524,288
14. model.layers.4.self_attn.v_proj sens=9.197e-03 params=524,288
15. model.layers.0.mlp.up_proj sens=8.446e-03 params=12,582,912
16. model.layers.2.mlp.up_proj sens=8.325e-03 params=12,582,912
17. model.layers.6.self_attn.v_proj sens=7.607e-03 params=524,288
18. model.layers.2.mlp.down_proj sens=7.421e-03 params=12,582,912
19. model.layers.5.self_attn.v_proj sens=7.332e-03 params=524,288
20. model.layers.2.mlp.gate_proj sens=7.295e-03 params=12,582,912

Top 20 most-sensitive layers at 6-bit:
1. model.layers.1.mlp.down_proj sens=3.745e-02 params=12,582,912
2. model.layers.7.mlp.down_proj sens=3.612e-02 params=12,582,912
3. model.layers.1.self_attn.o_proj sens=3.090e-02 params=4,194,304
4. model.layers.0.self_attn.v_proj sens=2.678e-02 params=524,288
5. model.layers.2.self_attn.v_proj sens=1.461e-02 params=524,288
6. model.layers.1.self_attn.q_proj sens=1.415e-02 params=4,194,304
7. model.layers.0.self_attn.o_proj sens=1.355e-02 params=4,194,304
8. model.layers.1.self_attn.v_proj sens=1.179e-02 params=524,288
9. model.layers.0.mlp.down_proj sens=1.124e-02 params=12,582,912
10. model.layers.1.mlp.up_proj sens=9.918e-03 params=12,582,912
11. model.layers.2.self_attn.q_proj sens=9.245e-03 params=4,194,304
12. model.layers.2.self_attn.o_proj sens=9.187e-03 params=4,194,304
13. model.layers.4.self_attn.v_proj sens=8.027e-03 params=524,288
14. model.layers.3.self_attn.v_proj sens=8.001e-03 params=524,288
15. model.layers.0.mlp.up_proj sens=7.384e-03 params=12,582,912
16. model.layers.5.self_attn.q_proj sens=7.106e-03 params=4,194,304
17. model.layers.2.mlp.up_proj sens=6.858e-03 params=12,582,912
18. model.layers.1.mlp.gate_proj sens=6.849e-03 params=12,582,912
19. model.layers.2.mlp.down_proj sens=6.651e-03 params=12,582,912
20. model.layers.5.self_attn.v_proj sens=6.349e-03 params=524,288