from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Tarxxxxxx/TX-8G"

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    device_map="auto",
    torch_dtype="auto"
)

messages = [
    {"role": "user", "content": "Explain how local AI protects privacy."}
]

input_ids = tokenizer.apply_chat_template(
    messages,
    return_tensors="pt"
).to(model.device)

outputs = model.generate(
    input_ids,
    max_new_tokens=512,
    do_sample=True,
    temperature=0.7
)

print(tokenizer.decode(outputs[0], skip_special_tokens=True))

With llama.cpp

bash

# Download GGUF
wget https://huggingface.co/Tarxxxxxx/TX-8G/resolve/main/tx-8g.Q8_0.gguf

# Run with llama.cpp
./main -m tx-8g.Q8_0.gguf -p "Hello, I'm TARX." -n 256

With Ollama

ollama run tarx/tx-8g

Hardware Requirements

Hardware	Performance
Apple M1/M2/M3 (8GB)	⭐⭐⭐⭐⭐ Excellent
Apple M1/M2/M3 (16GB+)	⭐⭐⭐⭐⭐ Excellent
Intel i5 + 16GB RAM	⭐⭐⭐⭐ Good
Intel i7 + NVIDIA GPU	⭐⭐⭐⭐⭐ Excellent
AMD Ryzen + 16GB	⭐⭐⭐⭐ Good

Quantization Options

Format	Size	RAM	Speed	Quality
Q8_0	7.2 GB	8 GB	⭐⭐⭐⭐	⭐⭐⭐⭐⭐
Q6_K	5.5 GB	6 GB	⭐⭐⭐⭐⭐	⭐⭐⭐⭐
Q4_K_M	4.1 GB	5 GB	⭐⭐⭐⭐⭐	⭐⭐⭐

Training

TX-8G is fine-tuned from Qwen2.5-7B-Instruct with:

Additional instruction tuning for local-first use cases
Optimization for consumer hardware inference
Enhanced privacy-aware responses

Training data does not include any TARX user conversations (we don't have access to them).

Ethical Considerations

TX-8G is designed for local, private use. Because it runs on user devices:

No user data is collected
No conversations are logged
No usage is monitored
Users have complete control

Citation

bibtex

@misc{tarx2026tx8g,
  title={TX-8G: Local-First Language Model for Consumer Hardware},
  author={TARX Team},
  year={2026},
  publisher={HuggingFace},
  url={https://huggingface.co/Tarxxxxxx/TX-8G}
}

Links

Built by TARX | tarx.com