Instructions to use Luigi/minicpm5-1b-cursor with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use Luigi/minicpm5-1b-cursor 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 Luigi/minicpm5-1b-cursor:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Luigi/minicpm5-1b-cursor:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf Luigi/minicpm5-1b-cursor:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf Luigi/minicpm5-1b-cursor: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 Luigi/minicpm5-1b-cursor:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf Luigi/minicpm5-1b-cursor: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 Luigi/minicpm5-1b-cursor:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf Luigi/minicpm5-1b-cursor:Q4_K_M

Use Docker

docker model run hf.co/Luigi/minicpm5-1b-cursor:Q4_K_M

LM Studio
Jan
Ollama
How to use Luigi/minicpm5-1b-cursor with Ollama:
```
ollama run hf.co/Luigi/minicpm5-1b-cursor:Q4_K_M
```

Unsloth Studio

How to use Luigi/minicpm5-1b-cursor with Unsloth Studio:

Install Unsloth Studio (macOS, Linux, WSL)

curl -fsSL https://unsloth.ai/install.sh | sh
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Luigi/minicpm5-1b-cursor to start chatting

Install Unsloth Studio (Windows)

irm https://unsloth.ai/install.ps1 | iex
# Run unsloth studio
unsloth studio -H 0.0.0.0 -p 8888
# Then open http://localhost:8888 in your browser
# Search for Luigi/minicpm5-1b-cursor to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for Luigi/minicpm5-1b-cursor to start chatting

How to use Luigi/minicpm5-1b-cursor with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Luigi/minicpm5-1b-cursor:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @mariozechner/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "Luigi/minicpm5-1b-cursor:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

OpenClaw new

How to use Luigi/minicpm5-1b-cursor with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Luigi/minicpm5-1b-cursor: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 "Luigi/minicpm5-1b-cursor: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"

Docker Model Runner
How to use Luigi/minicpm5-1b-cursor with Docker Model Runner:
```
docker model run hf.co/Luigi/minicpm5-1b-cursor:Q4_K_M
```

Lemonade

How to use Luigi/minicpm5-1b-cursor with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull Luigi/minicpm5-1b-cursor:Q4_K_M

Run and chat with the model

lemonade run user.minicpm5-1b-cursor-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use Luigi/minicpm5-1b-cursor with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf Luigi/minicpm5-1b-cursor: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 Luigi/minicpm5-1b-cursor:Q4_K_M

Run Hermes

hermes

Atomic Chat

MiniCPM5-1B-CURSOR — agentic meeting summarizer (sub-1B on-device)

Fine-tune of openbmb/MiniCPM5-1B on the CURSOR protocol: a streaming agent that converts long meeting transcripts (zh-TW / en) into structured, fully-anchored meeting notes by emitting edit ops (ADD/UPD/DEL/CMP/NOP) against one evolving NOTES state — no map-reduce, no ReAct loops. Same family as the earlier Luigi/lfm2.5-350m-cursor-* releases, but a single model covering BOTH languages (the 350M needed a per-language pair; this 1B holds both).

How it was made

Base: openbmb/MiniCPM5-1B (4k context, linear+full attention hybrid).
Data: teacher traces from Gemma-4-31B (judge-filtered, ~2k steps at 2048-token chunks), sweep-feedback negative samples harvested from the students' own fabrications (170-230 per iteration), and synthetic hard-class counterfactuals (proposed-never- decided, informal-negation, negative-preference, reject-action meetings).
Training: 2-epoch continuation passes, LR 1e-5, effective batch 16, torch DDP on two GPUs. This checkpoint (274) is the G1-verified artifact — verified three times.
Deployment harness: src/voxsum (same repo as the card) — streaming chunker (2048-token chunks, 2-line overlap), op parser, anchor validation, temporal guard, plus the VERIFY/ANCHOR sweep (gpt-oss-20b judge at eval time) and a deterministic UPD→ADD fallback for UPDs against an unmatched prefix.

Measured results (T1 tier, n=20, local judges, 3x majority; see caveats)

metric	this model	map-reduce baseline (Qwen3.5-9B)
G1 capability screen	PASS en + zh (chain, deadlines, anchored, trap)	—
valid-op rate (screen)	en 100% / zh 88% (one redundant duplicate-ADD rejected)	—
raw INVERT (model only, no sweep)	2/20	3/20
swept INVERT (deployed: model + VERIFY/ANCHOR sweep)	0/19-0/20	—
FAITH-claim (1-5)	4.81-4.84	3.50
COVER (1-5)	2.84-2.89	3.05
SYNTH (1-5)	2.32	2.60 (tie within judge noise +-0.4-0.5)

The ship rule (spec section 7.7) is GT2 OR GT3 at equal inversions: GT2 clears decisively (FAITH +1.3 at FEWER inversions than the baseline). The sweep is part of the deployed pipeline and is what yields the 0-inversion deployment.

Usage

llama.cpp server, greedy, thinking OFF (the model emits hybrid <think> unless disabled):

llama-server -m minicpm5-1b-cursor.Q4_K_M.gguf --n-gpu-layers 999 --ctx-size 4096 \
  --parallel 1 --flash-attn on --jinja --reasoning off --temp 0

The client is the CURSOR harness (agentic-summarizer repo, eval/run_arms.py --arms cursor --base-url <server> --tokenizer openbmb/MiniCPM5-1B --budget 2048). The model expects the exact system prompt and NOTES rendering the harness produces — it is NOT a general chat model.

Caveats (must accompany every reported number)

zh T2 tier is synthetic; the zh pool is largely monologic — contested-zh is unmeasured.
Judge-noise floor +-0.4-0.5 (FAITH/SYNTH); n=20 per tier; reduced cells directional.
The zh trap behavior sits near the decision boundary between adjacent checkpoints — use checkpoint-274 (this GGUF); the training final (284) fails the zh trap.
On-device envelope: ~650 MB Q4_K_M, 4k context, per-step ~2.9k prefill tokens.

Downloads last month: 3

GGUF

Model size

1B params

Architecture

llama

Hardware compatibility

4-bit

Model tree for Luigi/minicpm5-1b-cursor

Base model

openbmb/MiniCPM5-1B

Quantized

(92)

this model