moonshotai/Kimi-K2-Instruct-0905
It's queued (and Kimi-K2-Base is still in our queue...)
It requires special handling (1TB XD :) so might take a while. The status page will have you covered.
You can check for progress at http://hf.tst.eu/status.html or regularly check the model
summary page at https://hf.tst.eu/model#Kimi-K2-Instruct-0905-GGUF for quants to appear.
The F16 source GGUF in fact is 2.1 TB (1.9 TiB). With mainline llama.cpp I would have required 2.1 TB additional space for the BF16 SafeTensor version and if I had used --outtype=source it would have chosen F32 for most tensors (as llama.cpp internally stores F8_E4M3 as F32) and so created a 4.2 TB source GGUF. Don't forget the fact that the source SafeTensor model needs to be downloaded somewhere as well. So even in best case with heavily modified llama.cpp over 3 TB of storage are required just to convert the model into a GGUF. What also didn't help that I downloaded Intern-S1 to exactly the storage I need hours before your request. In the end I had so separate myself from some older GFUFs and moved some data to HDD just to free up the required storage.
In any case after over 4.5 hours of converting we now finally have Kimi-K2-Instruct-0905.gguf and nico1 started working on creating quants for it.
You can check for progress at http://hf.tst.eu/status.html or regularly check the model
summary page at https://hf.tst.eu/model#Kimi-K2-Instruct-0905-GGUF for quants to appear.
Brothers, thank you for your hard work. XD
GGUF divorces are the worst.
Thanks, Good job,XD