AI Models Helix Uses
Text Models
We use Model aliases for different use cases. We update the underlying model to the most performant and efficient model available for that role.
You can specify the
Helix 3.5
- Utilizes Llama3-8B for fast and efficient performance, ideal for everyday tasks.
- Ollama tag
llama3:instruct
Helix 4
- Powered by Llama3-70B, this model offers deeper insights and although a bit slower, it’s smarter for complex queries.
- Ollama tag
llama3:70b
Helix Mixtral
- Features CodeLlama-70B from Meta, which excels in programming and coding tasks, surpassing the capabilities of GPT-4 in software development contexts.
- Ollama tag
mixtral:instruct
Helix JSON
- Operates on Nous Hermes 2 Theta Llama3 7B, specialized for function calling and generating JSON outputs, enhancing automation and integration tasks.
- Ollama tag
adrienbrault/nous-hermes2theta-llama3-8b:q8_0
Helix Small
- A smaller model based on Phi-3 Mini 3.8B, fast and memory efficent.
- Ollama tag
phi3:instruct
Helix Text Fine Tuning
- Utilizes Mistral-7B-Instruct via Axolotl for Fine-tuning
Text models use Ollama for inference and axolotl for fine-tuning (and inference on a fine-tune).
See this for the fine-tuning configuration we use.
Image Models
Image models use the excellent cog (in particular cog-sdxl).
See this for the settings we use.
Video Models
Coming soon.
What else would you like to see? Let us know on Discord!
Last updated on