TomCon.net offers a free public service for hobbyist to experiment with a machine-learning model through kobold.cpp lite webui interface. Here you may use my compute for free to experiment with local quantized models.
Contact me through one of the methods listed in my contacts page, I will send you an API key to login.
I currently have a few modes depending on purpose. Note that for now these need to be manually changed out by me on the server operator end. For now I maintain sole admin rights to swap out model modes, if you would like a mode swapped or exclusive access to the service reserved for the day for a dedicated project feel free to contact me.
Deephermes 8b preview is a Chain of Thought (CoT) reasoning model finetuned from llama3.1 using deepseek R1 distilled reasoning patterns. DeepHermes features a toggleable CoT reasoning with a specific system prompt. CoT is useful for more intricate reasoning task such as solving problems or boilerplating chains of logic in a programming language.
Currently ai.tomcon.net runs on a recycled desktop class PC with an nvidia 1070ti 8gb that consumes about 200 watts at load or 60 watts at idle. All publically available models are fully loaded on the GPU for speed.
The following text contains the privacy policy and rules for using tomcon.net
TomCon.net does not track your IP beyond basic logging of connections. tomcon.net does not collect telemetry or analytics. tomcon.net runs kobold logging in quiet mode so that all input and outputs are suppressed for added privacy (I cant see/read what you send and recieve). Kobold has been set up to be properly delivered over the HTTPS protocol with SSL/TLS encryption certificates to prevent MITM snooping.
I, Thomas Conway, reserve the right to cut off both individual and total usage to ai.tomcon.net as a service I privately own and maintain for public use. The service may go down at any point for an indeterminate amount of time while undergoing maintenance or upgrades to the homelab.