karpathy / llm.c

LLM training in simple, raw C/CUDA

Repository from Github https://github.comkarpathy/llm.cRepository from Github https://github.comkarpathy/llm.c

LLM.c in google colab

Eliah7 opened this issue · comments

For those poor in GPUs, Google Colab can be an easy alternative to run and experiment with llm.c. Below is a sample notebook that allowed me to train the fp32 code. Only make sure to select the GPU runtime.

https://colab.research.google.com/drive/1pFCqlGkfJiIr1HLTgaMh4i5OtiY7P96P?usp=sharing

Cool thanks!

When trying to use Google Colab free (T4 GPU), I got this error:

[CUDA ERROR] at file train_gpt2_fp32.cu:701: the provided PTX was compiled with an unsupported toolchain.
can you help me to figure out?