Some commenters on this post are clearly not aware of PTX being a part of the CUDA environment. If you know this, you aren’t who I’m trying to inform.

There seems to be some confusion here on what PTX is – it does not bypass the CUDA platform at all. Nor does this diminish NVIDIA’s monopoly here. CUDA is a programming environment for NVIDIA GPUs, but many say CUDA to mean the C/C++ extension in CUDA (CUDA can be thought of as a C/C++ dialect here.) PTX is NVIDIA specific, and sits at a similar level as LLVM’s IR. If anything, DeepSeek is more dependent on NVIDIA than everyone else, since PTX is tightly dependent on their specific GPUs. Things like ZLUDA (effort to run CUDA code on AMD GPUs) won’t work. This is not a feel good story here.

You can look up Eric Hartford on Huggingface for more info.

Basically, somebody removes such restrictions from models, and publishes uncensored ones under the name “Dolphin”. Presumably, an uncensored Deepseek would be called something like “Deepseek R1-dolphin”. The full Deepseek R1 is quite large, and I’m not sure when this will happen yet. But there are other great Dolphin models.

Some models like Meta’s Llama are way too censored to be useful for many completely normal use cases, and the guy is doing God’s work in my opinion.