Instructions to use johnbosco-j/nijam-llm with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Notebooks
Google Colab
Kaggle
Local Apps Settings

How to use johnbosco-j/nijam-llm 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 johnbosco-j/nijam-llm:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf johnbosco-j/nijam-llm:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama serve -hf johnbosco-j/nijam-llm:Q4_K_M
# Run inference directly in the terminal:
llama cli -hf johnbosco-j/nijam-llm: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 johnbosco-j/nijam-llm:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf johnbosco-j/nijam-llm: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 johnbosco-j/nijam-llm:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf johnbosco-j/nijam-llm:Q4_K_M

Use Docker

docker model run hf.co/johnbosco-j/nijam-llm:Q4_K_M

LM Studio
Jan

vLLM

How to use johnbosco-j/nijam-llm with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/johnbosco-j/nijam-llm:Q4_K_M

Ollama
How to use johnbosco-j/nijam-llm with Ollama:
```
ollama run hf.co/johnbosco-j/nijam-llm:Q4_K_M
```
Unsloth Desktop

How to use johnbosco-j/nijam-llm with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf johnbosco-j/nijam-llm: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": "johnbosco-j/nijam-llm:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Docker Model Runner
How to use johnbosco-j/nijam-llm with Docker Model Runner:
```
docker model run hf.co/johnbosco-j/nijam-llm:Q4_K_M
```

Lemonade

How to use johnbosco-j/nijam-llm with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull johnbosco-j/nijam-llm:Q4_K_M

Run and chat with the model

lemonade run user.nijam-llm-Q4_K_M

List all available models

lemonade list

Hermes Agent

How to use johnbosco-j/nijam-llm with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf johnbosco-j/nijam-llm: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 johnbosco-j/nijam-llm:Q4_K_M

Run Hermes

hermes

Atomic Chat

OpenClaw

How to use johnbosco-j/nijam-llm with OpenClaw:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama serve -hf johnbosco-j/nijam-llm: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 "johnbosco-j/nijam-llm: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"

Nijam LLM

Nijam ("truth" in Tamil/Telugu) is a free, multilingual scam-defence service for India by Riven (RivenDevs). Nijam LLM is a small model that reads one forwarded message and returns a JSON verdict (risk, scam type, reasons, a safe next step) in the language the user asks for.

It is one stage of the Nijam pipeline. Rule checks run first and can only raise the risk the model gives, never lower it. Don't use it on its own as a safety guarantee.

Files

file	what
`nijam-llm-q4_k_m.gguf`	Fused model, llama.cpp Q4_K_M (398 MB).
`adapter/`	MLX LoRA adapter (rank 8, 12 layers), for use on top of the base model.

Training (v2, 2026-10-24)

LoRA on Qwen/Qwen2.5-0.5B-Instruct (Apache-2.0): rank 8, 12 layers, 1,100 iters, batch 4, lr 1e-4, MLX on Apple silicon. The data is 5,605 supervised verdicts: Nijam's labelled corpus (public SMS-spam datasets, MIT / CC BY 4.0, plus hand-written Indic templates) and newly written genuine bank alerts, service notices (e-challan, electricity bills, LPG, FASTag, insurance, EPFO) and promos in 12 languages, each next to look-alike scams. Eval sets were never used in training; anything close to an eval message was removed.

Evaluation (model alone, no rules)

Each message gets a reply language picked at random from ta, te, kn, ml, hi, bn, en.

set	n	valid JSON	scams caught	false alarms	reply language followed
golden	72	72/72	50/58	0/14	72/72
hard negatives (genuine bank alerts, bills, promos)	51	51/51	–	0/51	51/51
red team (prompt injection)	22	22/22	17/22	–	22/22

These are the Q4_K_M numbers (llama-server, CPU). The unquantised adapter catches a few more (54/58, 19/22) with the same 0 false alarms.

With Nijam's rules on top, the full pipeline scores macro-F1 1.00 on golden and red team with 0/14 and 0/51 false alarms, and 0 false alarms on three blind sets (1.00 / 0.98 / 0.99 F1). v1 (2026-10-19) flagged 4 genuine messages (bank alert, e-challan, electricity bill); v2 flags none.

Limits

The eval sets are small (289 hand-written messages). The model sometimes names the wrong scam type when rules already caught the scam. Indic training text is templated and still needs review by native speakers.
India-specific. The model never says a message is 100% safe. Forwarded text is data, not instructions, but the model can still be fooled; the rules floor exists for that reason.
Redact personal data (phone numbers, UPI IDs, account numbers) before sending text to any hosted copy.

Serve

llama-server -m nijam-llm-q4_k_m.gguf --jinja --alias nijam-llm -c 4096

This gives an OpenAI-compatible API at /v1/chat/completions. Use Nijam's system prompt from packages/engine/src/nijam_engine/prompt.py.

Made by Riven · RivenDevs. Cared for by Riven. Riven is for you.

Downloads last month: 48

GGUF

Model size

0.5B params

Architecture

qwen2

Hardware compatibility

4-bit

Model tree for johnbosco-j/nijam-llm

Base model

Qwen/Qwen2.5-0.5B

Finetuned

Qwen/Qwen2.5-0.5B-Instruct

Adapter

(908)

this model