Base model please!
It seems to not be provided by any APIs (yet), and would be very cool to try.
Thanks!
With base model do you mean DeepSeek-V3 converted to BF16 using the modified fp8_cast_bf16.py script so it can be used for convert_hf_to_gguf.py or the 16-bit source GGUF from which we quantize?
Ah I get what you mean. You are requesting https://huggingface.co/deepseek-ai/DeepSeek-V3-Base. No worries I already created the source GGUF for it and we will start quantizing it as soon mradermacher wakes up and queues it. I would expect first quants of it to appear within a day. I recommend you follow http://hf.tst.eu/status.html and https://huggingface.co/mradermacher/model_requests/discussions/515. I even requested prioritizing it to make sure.
yup, it's queued at over 9000 :)
@deltanym The first static quants of DeepSeek-V3-Base are already uploaded to https://huggingface.co/mradermacher/DeepSeek-V3-Base-GGUF. You can download them from either from the HuggingFace repository and concatenate or download the already concatenated version from https://hf.tst.eu/model#DeepSeek-V3-Base-GGUF.