VdrontV
Collection
1 item • Updated
VdrontV1 (gpt-like) is the first full model (no experiment) that is neither a test nor an experimental version; it was trained on FineWeb-Edu and FineWeb-Instruct.
It can answer simple questions phrased like "What is Python?" or "How many people are on Earth?"; however, it lacks a sense of self ("I") and cannot handle small talk.
According to Dront Model Tester results, it outperformed VdrontGPT50m-Base2 by a factor of two and came close to Supra50m-Base-Exp (which was trained on a significantly larger number of tokens).
Examples (use.py in files, t 0.4):
You: What is the most popular programming language?
Ai: The most popular programming language is Python. It is a general-purpose programming language that is used to create programs and applications. It is a powerful tool for creating complex programs, and it is also used by many computer programmers.
You: What is a Wi-Fi?
AI: Wi-Fi is a type of wireless network that connects your home to your phone's internet connection. It allows you to connect to a wide variety of devices, including laptops, tablets, and other devices. Wi-Fi is also known as Wi-Fi. You can connect to any device through Wi-Fi, such as the Internet, tablet, or smartphone. The internet is a great way to connect with friends, family, and colleagues. Wi-Fi is also known as "smart" or "smart" or "smart."
You: What is a Site?
AI: A Site is a web site that allows users to create, manage and share their own websites. It is used by many websites, including Facebook, Twitter, Instagram, and YouTube. The site is also used by many websites, such as Google, Yahoo, and others.
| Feature | Detail |
|---|---|
| Architecture | Transformer-based |
| Parameters | +-100,000,000 |
| Precision | F32 (Full float32) |
| Tokenizer | GPT-2 Tokenizer |
| Format | Hugging Face |
| Special Tokens | <|user|> <|assistant|> |
| Dront Model Tester | Supra1.5 Base Exp | VdrontGPT50m-Base2 | VdrontV1 |
|---|---|---|---|
| Gramm | 13% | 1% | 7% |
| Talk | 38% | 17% | 30% |
| Info | 17% | 6% | 12% |
| Train tokens | 23b | 0.5b | 2.1b |
P.S. next, the Vdront V1.5 (150m) is planned—an expanded version of this model.
P.S. train 21.04 token per param! (2.1b tokens)
:-)
- Downloads last month
- -
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support