disclaimer 2
does it mean it only performs well in benchamrks but not in real life or when deployed.
Hey. No it does mean it's not extensively tested for real-life use cases.
Could we get Importance Matrix gguf versions of your current and future best performing models?
I threw about a couple of thousand queries to score some labeling and the output was great. In general just using this model really impressed me. I also threw some general queries about public knowledge and it seem to produce good results. Nothing too detailed but it did a good jobs of analysis and reasoning for few shot prompts.
Could we get Importance Matrix gguf versions of your current and future best-performing models?
Hey sorry that I missed that message, I can consider that for better performing models, might be a good thing to offer so others can also benefit from that.
I threw about a couple of thousand queries to score some labeling and the output was great. In general just using this model really impressed me. I also threw some general queries about public knowledge and it seem to produce good results. Nothing too detailed but it did a good jobs of analysis and reasoning for few shot prompts.
Oh, that's great to hear, I'm glad that the model performs well in your use case. It was only trained with a small portion of a preference dataset, based on what you share I feel like training it for longer could help us improve even more