GPT

You can chat with the decentrally hosted model at chat.hypertensor.org hosting an uncensored version of Llama 3.1 8B. This model is validated in a subnet using the Decentralized Subnet Standard.

This is a beta version of a chat GPT, expect bugs.

Features

• Text generation

• Voice to text

• Text to voice

Performance

With good GPUs, internet connection, and geography, expect between 8-20 tokens/s.

Model	GPU	Tokens Per Second
Llama 3.1 8B	NVIDIA 3070	6-15
Llama 3.1 8B	NVIDIA T4	4-8.5
Llama 3.1 8B	NVIDIA 4090	7.1-20

The tokens per second can slow down when the subnet has many clients using it and when there aren't nodes in your region due to latency.

As this is a testnet and unincentivized, most of the nodes in the subnet won't be using high-performance hardware.