Looking for some guidance
I'm in the process of building a model currently myself. I actually have a 94 million parameters model. I'm looking to have someone test and verify possibly partner. This model does not focus on general instruct only. Its trained on public and locally curated coding and agentic datasets. My model is in my opinion a different kind of model than before. Would you be interested in taking a look?
yes
I want to check the details later
Just so your aware. I don't have the same type of hardware I'm sure you have. I have a gaming desktop with 16gb shared ram 6gb vram. I'm actually able to run a hive of agents
okay, I have an MI50, but in fact there is only one, and I accidentally set it to 11
My project , VITALIS-X is an embodied AI research project exploring persistent digital agents that learn through experience, memory, feedback, and simulated interaction rather than isolated prompt/response sessions.
The goal is to study how AI agents can develop continuity, adapt over time, and improve through repeated learning cycles in interactive environments.
I'm sure you understand why I'm being vague here
okay
How exactly does the AI of work?
(running example of VITALIS-X)
Can we talk off of here... I'm telling you this is very intricate and I don't want the larger companies reading this
At a high level, VITALIS-X works by treating the AI less like a one-off chatbot and more like a persistent digital agent inside a learning environment.
A normal chatbot usually receives a prompt, generates a response, and then the interaction mostly ends. VITALIS-X is exploring a different approach: the agent keeps state, receives feedback, remembers patterns from prior attempts, and improves through repeated cycles.
A simple running example would be:
- The agent is given a task or situation.
- It responds or takes an action.
- The system checks whether the result worked.
- Mistakes are turned into training feedback.
- The agent tries again with that feedback in mind.
- Over time, the goal is for the agent to become better at reasoning, communication, coding, and decision-making.
So the core idea is experience-based learning: repeated interaction, feedback, memory, and improvement over time.
I’m keeping the implementation details private for now, but that’s the general direction of the project.
This is all I can elaborate on here. Id love to tell you exact parameters etc but I can't do that here
I'm rate limited for comments. Hopefully you see my edit.
And I'm not worried about huggingface itself or the open source community. I'm worried about the big companies coming in and taking this idea an making it fall under their own umbrella in essence. .. I think if done right this agent type could be the robotics future.
ok,sorry
I'm an individual developer
Let me confirm: Are you saying that you don't want to be read by the huggingface team?
It's not a local llm, so I can't upload it to huggingface,
but I'm currently making an HTML agent using Gemini
I was rate limited for comments.
And I'm not worried about huggingface itself or the open source community. I'm worried about the big companies coming in and taking this idea an making it fall under their own umbrella in essence. .. I think if done right this agent type could be the robotics future.
oh, I finally understood
You don't want to have your ideas stolen by a large company after it's released and be more popular.
I don't care about the popularity part. I don't want them to monopolize
Oh well.
I would like to become a tester later, but if possible, please convert GGUF (if you can't convert, I run the safetensors as it is)