Professional Documents
Culture Documents
com/
Introduction
This is where the new AI Model comes into play as a true game-changer.
This state-of-the-art LM has undergone extensive fine-tuning on a
diverse dataset comprising over 300,000 instructions, covering a wide
range of topics and tasks. Developed by Nous Research, a prominent AI
The underlying mission behind the creation of this model was to craft a
potent and versatile LM capable of effectively tackling diverse tasks and
domains while maintaining high quality and accuracy. Built upon the
foundation of GPT-3.5-turbo, an enhanced iteration of GPT-3 with
increased parameters and superior training data, this model also
benefits from quantization using GPTQ. This technique reduces the
model's size and memory requirements without compromising
performance. Moreover, the model is fully compatible with Hugging
Face, a popular platform that offers seamless accessibility and utilization
of various LMs. This new AI model is called 'Nous-Hermes-13B'.
Nous-Hermes-13B has several key features that make it stand out from
other LMs:
also provide fun and engaging interactions for users and players.
These are just some of the possible use cases of this model. There are
many more applications that can be explored and discovered by using
this model.
Architecture of Nous-Hermes-13B
The model has 13 layers in both the encoder and the decoder. Each
layer has 12 attention heads and a hidden size of 5120. The total
number of parameters in the model is 48 billion. The model is trained on
a large corpus of text data from various sources and domains. The
fine-tuning process uses a custom dataset of over 300,000 instructions
that cover diverse topics and tasks.
To use the model locally, you will need to install PyTorch, Transformers,
and GPTQ-for-LLaMa. You can find the instructions on how to install and
run these libraries on their respective websites or GitHub repositories.
To use the model online, you will need to create an account on Hugging
Face and get an API key. You can then use the Hugging Face API to
send requests to the model and receive responses. You can also use the
text-generation-webui to enter prompts and see the model outputs in a
browser.
Limitations
Conclusion
source
https://huggingface.co/TheBloke/Nous-Hermes-13B-GPTQ
https://huggingface.co/NousResearch/Nous-Hermes-13b