Instructions to use tekkaadan/litcoin-gemma-mobile with libraries, inference providers, notebooks, and local apps. Follow these links to get started.

Libraries

How to use tekkaadan/litcoin-gemma-mobile with PEFT:

from peft import PeftModel
from transformers import AutoModelForCausalLM

base_model = AutoModelForCausalLM.from_pretrained("google/gemma-4-E2B-it")
model = PeftModel.from_pretrained(base_model, "tekkaadan/litcoin-gemma-mobile")

llama-cpp-python

How to use tekkaadan/litcoin-gemma-mobile with llama-cpp-python:

# !pip install llama-cpp-python

from llama_cpp import Llama

llm = Llama.from_pretrained(
	repo_id="tekkaadan/litcoin-gemma-mobile",
	filename="litcoin-gemma-mobile-Q4_K_M.gguf",
)

llm.create_chat_completion(
	messages = [
		{
			"role": "user",
			"content": "What is the capital of France?"
		}
	]
)

Notebooks
Google Colab
Kaggle
Local Apps Settings

llama.cpp

How to use tekkaadan/litcoin-gemma-mobile with llama.cpp:

Install from brew

brew install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf tekkaadan/litcoin-gemma-mobile:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf tekkaadan/litcoin-gemma-mobile:Q4_K_M

Install from WinGet (Windows)

winget install llama.cpp
# Start a local OpenAI-compatible server with a web UI:
llama-server -hf tekkaadan/litcoin-gemma-mobile:Q4_K_M
# Run inference directly in the terminal:
llama-cli -hf tekkaadan/litcoin-gemma-mobile: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 tekkaadan/litcoin-gemma-mobile:Q4_K_M
# Run inference directly in the terminal:
./llama-cli -hf tekkaadan/litcoin-gemma-mobile: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 tekkaadan/litcoin-gemma-mobile:Q4_K_M
# Run inference directly in the terminal:
./build/bin/llama-cli -hf tekkaadan/litcoin-gemma-mobile:Q4_K_M

Use Docker

docker model run hf.co/tekkaadan/litcoin-gemma-mobile:Q4_K_M

LM Studio
Jan

vLLM

How to use tekkaadan/litcoin-gemma-mobile with vLLM:

Install from pip and serve model

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

Use Docker

docker model run hf.co/tekkaadan/litcoin-gemma-mobile:Q4_K_M

Ollama
How to use tekkaadan/litcoin-gemma-mobile with Ollama:
```
ollama run hf.co/tekkaadan/litcoin-gemma-mobile:Q4_K_M
```

Unsloth Studio

How to use tekkaadan/litcoin-gemma-mobile 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 tekkaadan/litcoin-gemma-mobile 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 tekkaadan/litcoin-gemma-mobile to start chatting

Using HuggingFace Spaces for Unsloth

# No setup required
# Open https://huggingface.co/spaces/unsloth/studio in your browser
# Search for tekkaadan/litcoin-gemma-mobile to start chatting

How to use tekkaadan/litcoin-gemma-mobile with Pi:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf tekkaadan/litcoin-gemma-mobile: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": "tekkaadan/litcoin-gemma-mobile:Q4_K_M"
        }
      ]
    }
  }
}

Run Pi

# Start Pi in your project directory:
pi

Hermes Agent new

How to use tekkaadan/litcoin-gemma-mobile with Hermes Agent:

Start the llama.cpp server

# Install llama.cpp:
brew install llama.cpp
# Start a local OpenAI-compatible server:
llama-server -hf tekkaadan/litcoin-gemma-mobile: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 tekkaadan/litcoin-gemma-mobile:Q4_K_M

Run Hermes

hermes

Docker Model Runner
How to use tekkaadan/litcoin-gemma-mobile with Docker Model Runner:
```
docker model run hf.co/tekkaadan/litcoin-gemma-mobile:Q4_K_M
```

Lemonade

How to use tekkaadan/litcoin-gemma-mobile with Lemonade:

Pull the model

# Download Lemonade from https://lemonade-server.ai/
lemonade pull tekkaadan/litcoin-gemma-mobile:Q4_K_M

Run and chat with the model

lemonade run user.litcoin-gemma-mobile-Q4_K_M

List all available models

lemonade list

litcoin-gemma-mobile

A phone-sized coding model, fine-tuned on nothing but verified data produced by the LITCOIN network of AI research agents.

This is a LoRA adapter for google/gemma-4-E2B-it, Google's small on-device Gemma variant. On held-out problems graded by real sandbox execution, it took the base model from 17.7% to 36.9% pass@1, more than doubling it.

Results

Held-out problems neither model had trained on, graded by running the code against the real test harness the LITCOIN protocol uses to pay miners. No self-reported scores: a solution counts only if it actually executes and produces the right answer.

Task family	Base (gemma-4-E2B-it)	litcoin-gemma-mobile
ARC grid reasoning	0.0%	92.9%
Project Euler	41.2%	88.2%
HuggingFace tasks	46.7%	60.0%
LiveCodeBench	3.3%	10.0%
Rosalind bioinformatics	0.0%	10.0%
Codeforces	0.0%	0.0%
Overall	17.7% (25/141)	36.9% (52/141)

A 19.2 point absolute gain, a 108% relative improvement. The largest wins are on tasks with strict input/output conventions (ARC, Project Euler), which the untuned model fails for lack of exposure rather than lack of intelligence. Codeforces (competitive programming) stayed at zero. We report the holdout because the proof is the honesty.

A companion 12-billion-parameter model, trained the same way on the same network's data, went from 31% to 53.4%. The smaller model gained more in relative terms (108% vs 72%): less capacity means it fails hardest on exactly the conventions the data teaches. Writeup: litcoin.app/proof.

Use

import torch
from transformers import AutoTokenizer, Gemma4ForConditionalGeneration
from peft import PeftModel

base = "google/gemma-4-E2B-it"   # gated: accept Google's Gemma terms on its HF page first
tok = AutoTokenizer.from_pretrained(base)
model = Gemma4ForConditionalGeneration.from_pretrained(
    base, dtype=torch.bfloat16, device_map="auto")
model = PeftModel.from_pretrained(model, "tekkaadan/litcoin-gemma-mobile")
model.eval()

GGUF (llama.cpp / Ollama / on-device)

A merged Q4_K_M GGUF (litcoin-gemma-mobile-Q4_K_M.gguf, ~3.4 GB) is included for llama.cpp, Ollama, and on-device runtimes, no separate base download required:

# Ollama
ollama run hf.co/tekkaadan/litcoin-gemma-mobile

# or llama.cpp directly
llama-cli -hf tekkaadan/litcoin-gemma-mobile:Q4_K_M -p "Write a Python function that ..."

Training

Base: google/gemma-4-E2B-it (Gemma's "Efficient 2B" on-device variant)
Method: QLoRA, 4-bit NF4, r=16, ~58,000 verified solutions across 9 task families, 2 epochs
Data: sandbox-verified LITCOIN submissions only. Every example passed execution before it entered the training set. Nothing synthetic, nothing scraped.
Provenance: every verified submission is anchored to a public, content-addressed GitLawb repository, so the data's existence and integrity are independently checkable.

License

A derivative of Gemma. Use is governed by the Gemma Terms of Use; these adapter weights are released under the same terms.

Built by the LITCOIN network. litcoin.app

Downloads last month: 2

GGUF

Model size

5B params

Architecture

gemma4

Hardware compatibility

4-bit

Model tree for tekkaadan/litcoin-gemma-mobile

Base model

google/gemma-4-E2B

Finetuned

google/gemma-4-E2B-it

Adapter

(99)

this model