You're insane
Hi,
It seems Xwin-LM-13B-V0.2 has surpassed Xwin-LM-70B-V0.1 in AlpacaEval? You're insane!
But, still no paper release yet? Looking forward seeing it soon.
I've been extremely impressed with the result from the xwin models. The 0.2 version of 13b has been great, looking forward to the 70b.
I have tested the Mistral Orca model, and while it provides good performance, it tends to repeat itself frequently and
lacks depth in its responses. In comparison, the AI model I have been using for this project, version 13b, offers more
varied and indexed answers, and it appears to have a preference for phrasing things in the form of equations. However,
the model is not always accurate in the relationships between variables, especially in complex scenarios. Despite this,
it is an impressively strong model that performs exceptionally well.
According to the benchmarks, the 13b model outperforms some 70b models, which is truly remarkable. Additionally, there
is a 70b weights version available, which should be approaching light speed in terms of performance. I encourage you to
explore these models and their capabilities further, as they have shown significant potential.