[flagged]
The true hacker mentality is technically this.
Llm.c – LLM training in simple, pure C/CUDA
11–20 of 189 posts
Re: Llm.c – LLM training in simple, pure C/CUDA
#12[flagged]
Looking at the code it's just loops with simple floating point math. There's absolutely nothing in the code that would benefit from C++ features.
Re: Llm.c – LLM training in simple, pure C/CUDA
#13> LLM training in simple, pure C/CUDA. There is no need for 245MB of PyTorch or 107MB of cPython
Python has been popular for this because it’s convenient to quickly hack on and experiment with, not because it’s the most efficient thing.
The overhead really isn't that bad is it? Since the the python code is mostly about saying multiply matrix A with matrix B, and then that actual computation is done by optimized low level code.
Re: Llm.c – LLM training in simple, pure C/CUDA
#14Re: Llm.c – LLM training in simple, pure C/CUDA
#15Re: Llm.c – LLM training in simple, pure C/CUDA
#16Very sad, shouldve used an agnostic framework instead of CUDA
As far as I can tell, its optional dependency is Open MP, not CUDA. Doesn't seem directly dependent on CUDA.
Re: Llm.c – LLM training in simple, pure C/CUDA
#17Re: Llm.c – LLM training in simple, pure C/CUDA
#18Re: Llm.c – LLM training in simple, pure C/CUDA
#19Re: Llm.c – LLM training in simple, pure C/CUDA
#20Very sad, shouldve used an agnostic framework instead of CUDA
Are there any strong LLMs trained without CUDA?