Minicpm5 Deploy Ollama

Deploy MiniCPM5-1B and MiniCPM5-2B with Ollama

One-binary, no-Python laptop deployment. Consumes the released GGUF.

Required input

Steps

1. Install Ollama (once)

brew install ollama                                # macOS
# or:
curl -fsSL https://ollama.com/install.sh | sh      # Linux

OLLAMA_FLASH_ATTENTION=1 OLLAMA_KV_CACHE_TYPE=q8_0 ollama serve &

2. Download the GGUF + write Modelfile

mkdir -p ~/${MODEL_NAME} && cd ~/${MODEL_NAME}

huggingface-cli download ${GGUF_REPO} MiniCPM5-2B-${QUANT}.gguf --local-dir .

cat > Modelfile <<EOF
FROM ./MiniCPM5-2B-${QUANT}.gguf

# MiniCPM5 basic chat template
TEMPLATE """{{- if .Messages -}}
{{- range .Messages -}}
<|im_start|>{{ .Role }}
{{ .Content }}<|im_end|>
{{ end -}}
<|im_start|>assistant
{{ end -}}"""

PARAMETER stop "<|im_end|>"
PARAMETER stop "</s>"

# Defaults tuned for think mode
PARAMETER temperature 1.0
PARAMETER top_p 0.95
PARAMETER num_ctx 8192
EOF

3. Create + run

ollama create ${MODEL_NAME} -f Modelfile
ollama run ${MODEL_NAME}

4. Validate via OpenAI-compatible API

curl http://localhost:11434/v1/chat/completions \
    -H "Content-Type: application/json" \
    -d '{
        "model": "minicpm5-2b",
        "messages": [{"role": "user", "content": "1+1=?"}],
        "temperature": 1.0, "top_p": 0.95, "max_tokens": 64
    }'

Expected: "2" in the reply.

MiniCPM5-1B Think vs No-think

A MiniCPM5-1B Modelfile configured with temperature=0.7, top_p=0.95 defaults to no-think. To switch a single conversation to think mode, override the sampling params at request time:

ollama run minicpm5-1b --temperature 0.9 --top-p 0.95

Or bake it into a separate model tag by raising the temperature line (top_p stays 0.95):

PARAMETER temperature 0.9
PARAMETER top_p 0.95

Then ollama create minicpm5-1b-think -f Modelfile.think.

Ollama 0.24 does not directly evaluate the GGUF-embedded Jinja chat template. It maps recognized Jinja templates to built-in Go templates, so MiniCPM5 uses the explicit Go TEMPLATE block above. To enter the MiniCPM5-2B think path, use raw mode and prepend <think>\n manually:

curl http://localhost:11434/api/generate -d '{
    "model": "minicpm5-2b",
    "raw": true,
    "prompt": "<|im_start|>user\n鸡兔同笼…<|im_end|>\n<|im_start|>assistant\n<think>\n",
    "options": {"temperature": 1.0, "top_p": 0.95}
}'

Common pitfalls

• Error: invalid file magic: corrupted download. Re-run huggingface-cli download.

When NOT to use

• Highest throughput on Mac → minicpm5-deploy-mlx (Q4 build)
• GUI experience → minicpm5-deploy-lmstudio
• NVIDIA GPU production → minicpm5-deploy-vllm

Reference

docs/deployment/ollama.md