39
« Last post by Art on October 06, 2025, 02:26:48 pm »
Very nice work, Cyber Jedi!
I've been experimenting with this, and I am impressed by how well things work.
The new file is called UltraHal7LLM1.exe, and it is the whole package, similar to the existing UltraHal7-5-0-240 or whatever version you might currently have or run.
This new one includes many enhancements/features not found in the original HAL, and they should serve to showcase what HAL is capable of.
These LLMs (Large Language Models) use Ollama models, and you are free to choose which ones you wish to use, but try keeping the size reasonable to avoid lengthy response times or lags. From my experience using an ordinary 64-bit Windows 10 Home laptop, I try keeping the llama model size from 4 to 9 GB, and that strategy seems to work well. There are a huge assortment of llama models and each has its own storehouse of information and useful data that has been amassed and compiled.
Your own experimentation via question and answer will indicate which model or models will serve you best.
Cyber Jedi has done a great job at combining the learning module into HAL and the use of either of them will not and cannot corrupt each other!!
The information can sort of be shared if one wishes to do a Copy and Paste function. Just highlight the desired text, press CTRL C to Copy the selected text, then put your cursor in the desired area and press CTRL V to Paste the text where you choose.
There are many features and choices so play with them and decide which works best for you. The only Wrong move you might make would be to select a model from the list that you do NOT have in your system or have not downloaded. Relax, the world will not end. You will simply get a 404 warning letting you know that you will need to download that particular model, then try the selection again and it should work fine.
Otherwise, the main HAL GUI or skin with your avatar should answer either the Learning Model or the original HAL character of your choice. The character will still lip sync and emote as it always has and the speech will work based on your usual TTS selections within HAL.
Overall, this is a wonderful Hybrid operation that offers the best of both worlds, to allow Hal to learn new things as well as respond to you based on which llama model you have selected, or you can just use Hal as you always have. The choices are yours!
Enjoy!
Feel free to post your experiences as well.
A huge shout-out to our resident guru, Cyber Jedi, for forging this all together to give us this terrific resource and research project!! Thanks Cyber Jedi!!
Recent Posts