Instructions to use CinMin/guide-4b with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use CinMin/guide-4b 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 CinMin/guide-4b:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf CinMin/guide-4b:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf CinMin/guide-4b:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf CinMin/guide-4b:Q4_K_M

Build from source code

git clone https://github.com/ggml-org/llama.cpp.git
cd llama.cpp
cmake -B build
cmake --build build -j --target llama-app
# Start a local OpenAI-compatible server with a web UI:
./build/bin/llama serve -hf CinMin/guide-4b:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama cli -hf CinMin/guide-4b:Q4_K_M

Use Docker

docker model run hf.co/CinMin/guide-4b:Q4_K_M

Llama new
LM Studio
Jan

vLLM

How to use CinMin/guide-4b with vLLM:

Install from pip and serve model

# Install vLLM from pip:
pip install vllm
# Start the vLLM server:
vllm serve "CinMin/guide-4b"
# Call the server using curl (OpenAI-compatible API):
curl -X POST "http://localhost:8000/v1/chat/completions" \
	-H "Content-Type: application/json" \
	--data '{
		"model": "CinMin/guide-4b",
		"messages": [
			{
				"role": "user",
				"content": "What is the capital of France?"
			}
		]
	}'

Use Docker

docker model run hf.co/CinMin/guide-4b:Q4_K_M

Ollama
How to use CinMin/guide-4b with Ollama:
```
ollama run hf.co/CinMin/guide-4b:Q4_K_M
```
Unsloth Desktop

How to use CinMin/guide-4b with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf CinMin/guide-4b:Q4_K_M

Configure the model in Pi

# Install Pi:
npm install -g @earendil-works/pi-coding-agent
# Add to ~/.pi/agent/models.json:
{
  "providers": {
    "llama-cpp": {
      "baseUrl": "http://localhost:8080/v1",
      "api": "openai-completions",
      "apiKey": "none",
      "models": [
        {
          "id": "CinMin/guide-4b:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use CinMin/guide-4b with Docker Model Runner:
```
docker model run hf.co/CinMin/guide-4b:Q4_K_M
```

Lemonade

How to use CinMin/guide-4b with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull CinMin/guide-4b:Q4_K_M

Run and chat with the model

lemonade run user.guide-4b-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use CinMin/guide-4b with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf CinMin/guide-4b: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 CinMin/guide-4b:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use CinMin/guide-4b with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf CinMin/guide-4b: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 "CinMin/guide-4b: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"

CinMinAI commited on about 1 month ago

Commit

09c2f36

verified ·

1 Parent(s): 9bf70f3

Card: the project is public — links to the site, the OS download and the repository

Browse files

Files changed (1) hide show

README.md +9 -9

README.md CHANGED Viewed

@@ -7,10 +7,10 @@ library_name: gguf
 tags: [gguf, llama.cpp, linux-mint, assistant, cin-minai]
 ---
-> **The built-in guide of [Cin-MinAI](https://github.com/Cin-MinAI)**, a Linux Mint–based OS with an AI assistant
-> that runs on your own computer. This file ships on the Cin-MinAI ISO and installed systems fetch it from here,
-> checked against the SHA-256 below. Paths like `training/…` below are in the Cin-MinAI project repository, which
-> opens with the first public release.
 ## Run it
@@ -28,7 +28,7 @@ The small model built into Cin-MinAI OS that helps people who came from Windows
 looks things up in the built-in help, checks the computer, opens programs, works with a shared
 LibreOffice document, and politely declines everything else. Runs locally with llama.cpp; nothing leaves
 the machine. Everything below is measured unless it says otherwise. Decision: PLAN **D43**; the
-story of how we got here, failures included: `docs/guide-model-journal.md`.
 ## The pick
@@ -89,9 +89,9 @@ the stock model with prompt v2 stays one switch away.
     native phrase; the cause was our own prompt), no terminal, facts from the tool result, and
     **decontamination**: no training message within 0.5 (3-gram Jaccard) of any eval item; against the
     held-out set the closest is 0.25.
-  - Full description, counts, checks and known gaps: `training/datasets/DATASHEET.md` ("Session
     corpus", "Office corpus").
-- **Export:** `training/guide/merge_quantize.sh` — base → f16 GGUF → adapter merged with
   `llama-export-lora` → `llama-quantize` Q4_K_M (no importance matrix). All 128 adapter tensors
   verified merged from the weights.
@@ -168,10 +168,10 @@ prompt; about 10 s per tool call).
 ## Reproduce
-`training/datasets/sessions/merge.py`, `training/datasets/office/merge.py` (corpora, committed) →
 `MODEL=qwen bash training/guide/sweep.sh HO` (mix, train, adapter, public eval) →
 `bash training/guide/merge_quantize.sh HO Qwen3.5-4B Q4_K_M` → `EVAL_GGUF=… bash training/guide/sweep.sh
-merged NAME`. Held-out: `training/guide/held_out.sh` (cycle 0's set is now used; cycle 1 writes a new one).
 ## Licences

 tags: [gguf, llama.cpp, linux-mint, assistant, cin-minai]
 ---
+> **The built-in guide of [Cin-MinAI](https://cin-minai.github.io)**, a Linux Mint–based OS with an AI assistant
+> that runs on your own computer ([download the OS](https://huggingface.co/CinMin/Cin-MinAI-OS)). This file ships on
+> the Cin-MinAI ISO and installed systems fetch it from here, checked against the SHA-256 below. Paths like
+> `training/…` below are in the project's repository: [github.com/Cin-MinAI/Cin-MinAI](https://github.com/Cin-MinAI/Cin-MinAI).
 ## Run it
 looks things up in the built-in help, checks the computer, opens programs, works with a shared
 LibreOffice document, and politely declines everything else. Runs locally with llama.cpp; nothing leaves
 the machine. Everything below is measured unless it says otherwise. Decision: PLAN **D43**; the
+story of how we got here, failures included: [`docs/guide-model-journal.md`](https://github.com/Cin-MinAI/Cin-MinAI/blob/main/docs/guide-model-journal.md).
 ## The pick
     native phrase; the cause was our own prompt), no terminal, facts from the tool result, and
     **decontamination**: no training message within 0.5 (3-gram Jaccard) of any eval item; against the
     held-out set the closest is 0.25.
+  - Full description, counts, checks and known gaps: [`training/datasets/DATASHEET.md`](https://github.com/Cin-MinAI/Cin-MinAI/blob/main/training/datasets/DATASHEET.md) ("Session
     corpus", "Office corpus").
+- **Export:** [`training/guide/merge_quantize.sh`](https://github.com/Cin-MinAI/Cin-MinAI/blob/main/training/guide/merge_quantize.sh) — base → f16 GGUF → adapter merged with
   `llama-export-lora` → `llama-quantize` Q4_K_M (no importance matrix). All 128 adapter tensors
   verified merged from the weights.
 ## Reproduce
+[`training/datasets/sessions/merge.py`](https://github.com/Cin-MinAI/Cin-MinAI/blob/main/training/datasets/sessions/merge.py), [`training/datasets/office/merge.py`](https://github.com/Cin-MinAI/Cin-MinAI/blob/main/training/datasets/office/merge.py) (corpora, committed) →
 `MODEL=qwen bash training/guide/sweep.sh HO` (mix, train, adapter, public eval) →
 `bash training/guide/merge_quantize.sh HO Qwen3.5-4B Q4_K_M` → `EVAL_GGUF=… bash training/guide/sweep.sh
+merged NAME`. Held-out: [`training/guide/held_out.sh`](https://github.com/Cin-MinAI/Cin-MinAI/blob/main/training/guide/held_out.sh) (cycle 0's set is now used; cycle 1 writes a new one).
 ## Licences