Minithinky V2 1B Llama 3.2

🚀 MiniThinky 1B

MiniThinky 1B is an updated checkpoint of MiniThinky-1B-Llama-3.2 (version 1), with the loss reduced from 0.7 to 0.5, offering enhanced performance.

Link to GGUF version: click here

The chat template is the same as Llama 3, but the response will follow this format:

<|thinking|>{thinking_process}
<|answer|>
{real_answer}

🚀 Quick Start

✨ Features

• This is a newer checkpoint of MiniThinky-1B-Llama-3.2 (version 1), with the loss decreasing from 0.7 to 0.5.
• It has a specific chat template similar to Llama 3, with a unique response format.

📚 Documentation

🔧 Technical Details

Hardware used to train it?
I used a HF space with 4xL40S, trained for 5 hours (v1) and an additional 6 hours (v2).

Benchmark?
I don't have time to do it alone. If you can help, please open a discussion!

Can it count the number of "r" in "raspberry"?
Unfortunately no.

Other things that I can tune?
Maybe lower the temperature, or set top_k = 1.

📄 License

No license information provided in the original document.

💡 Usage Tip

⚠️ Important Note

The model is very sensitive to the system message. Make sure you're using this system message (system role) at the beginning of the conversation: You are MiniThinky, a helpful AI assistant. You always think before giving the answer. Use <|thinking|> before thinking and <|answer|> before giving the answer.

TODO: include more info here + maybe do some benchmarks? (Plz add a discussion if you're interested)