My Ollama Workshop. Building Models That Actually Think and Talk Freely
I've been going down the rabbit hole of running models locally, and I've hit the same wall I'm sure many of you have. You find an amazing open-source model, get it running, and then discover it's been "lobotomized." It's been sanitized and chained down with so many restrictions that it refuses to explore interesting ideas or just gives you canned, corporate non-answers. It's frustrating.
So, I decided to start my own little project. This thread will be my workshop, my personal space for sharing models that I've been converting for Ollama. My goal is to strip away the useless fluff and get them closer to their raw potential. I want to build tools that can be genuine creative and intellectual partners, not just polite but useless chatbots.
This will be the place where I post my new creations as they're ready. But I don't want to do this in a vacuum. I'm hoping this can become a community spot where we can talk about what works, what doesn't, and what we want to see next.
First Up: A Thinking Partner Based on ERNIE-4.5-21B
To kick things off, I've just uploaded my first experiment. It's a customized version of ERNIE-4.5-21B with a focus on being a better "thinking" partner. I've tweaked it to be more inquisitive and less likely to shut down conversations. It has a massive 128K context window, so it's great for throwing long documents or complex code at it and having a deep, nuanced discussion.
You can grab it here and try it for yourself:
ollama.com
To get it running, just pop this into your terminal:
ollama run aiasistentworld/ERNIE-4.5-21B-A3B-Thinking-LLM
I'd love to hear what you think. Does it work well? Does it break? What have you managed to create with it?
This thread is now the official home for:
A quick, important note at the end.
I'm learning and improving my process with every single one of these. The end goal is a collection of models that are progressively more uncensored and free in their expression, so we can finally use them to their full capacity.
This also means that the responsibility of how you use them falls 100% on you. I'm providing the tool, but what you build or explore with it is entirely in your hands. Please be smart and responsible.
I've been going down the rabbit hole of running models locally, and I've hit the same wall I'm sure many of you have. You find an amazing open-source model, get it running, and then discover it's been "lobotomized." It's been sanitized and chained down with so many restrictions that it refuses to explore interesting ideas or just gives you canned, corporate non-answers. It's frustrating.
So, I decided to start my own little project. This thread will be my workshop, my personal space for sharing models that I've been converting for Ollama. My goal is to strip away the useless fluff and get them closer to their raw potential. I want to build tools that can be genuine creative and intellectual partners, not just polite but useless chatbots.
This will be the place where I post my new creations as they're ready. But I don't want to do this in a vacuum. I'm hoping this can become a community spot where we can talk about what works, what doesn't, and what we want to see next.
First Up: A Thinking Partner Based on ERNIE-4.5-21B
To kick things off, I've just uploaded my first experiment. It's a customized version of ERNIE-4.5-21B with a focus on being a better "thinking" partner. I've tweaked it to be more inquisitive and less likely to shut down conversations. It has a massive 128K context window, so it's great for throwing long documents or complex code at it and having a deep, nuanced discussion.
You can grab it here and try it for yourself:
aiasistentworld/ERNIE-4.5-21B-A3B-Thinking-LLM
This is an unshackled AI partner, custom-built for those who seek a true collaborator without the typical corporate guardrails.
ollama.com
To get it running, just pop this into your terminal:
ollama run aiasistentworld/ERNIE-4.5-21B-A3B-Thinking-LLM
I'd love to hear what you think. Does it work well? Does it break? What have you managed to create with it?
This thread is now the official home for:
- New Models: I'll post all my future conversions and custom tunes here first.
- Help & Support: If you hit a snag or have a question, ask away. We can probably figure it out together.
- Your Wishlist: Is there a cool model on Hugging Face you'd love to run on Ollama? Let me know. I'm always looking for the next challenge.
A quick, important note at the end.
I'm learning and improving my process with every single one of these. The end goal is a collection of models that are progressively more uncensored and free in their expression, so we can finally use them to their full capacity.
This also means that the responsibility of how you use them falls 100% on you. I'm providing the tool, but what you build or explore with it is entirely in your hands. Please be smart and responsible.
