Good work - Great potential and current results
Hey, thanks for unleashing this to the open world!
I haven't completed my tests by far and I already got new ideas while working with Kimi-K2.
I can't wait to see what to expect to be capable with more APIs and tools enabled.
To give someone here an idea (or not):
For a test I was letting Kimi-K2 manage to spin up from zero a wsl2 instance on a Windows host with full CUDA 12.8 integration to the hosts hardware, CUDA 12.8, PyTorch, driver all detected. Kimi-K2 was pushing through the "roadblocks" getting all the setup done and to sum up, Kimi only needed a bit guidance on getting the correct URLs and a bit help pointing out a way to solve a problem with nvidia drivers by simply using others.
I have tested other LLMs with this task to troubleshoot or start working on it, and Kimi-K2 was the first to get it succeed. (No trophy I can hand over here, as I have limited time to conduct such kind of PoC/tests usually but to my experience this task was pretty tricky for other models. )
Glad you liked it!