How to inference nanoGPT2 ckpt.pt with c?

[aclie]

Hi a1k0n, I obtained a ckpt.pt file after fine-tuning GPT-2 with my own data. I followed this repository to fine-tune the GPT-2: link. However, I am currently facing issues and am unable to obtain the .bin file required for fast inference with C. I'm wondering if you know what changes I should make to address this problem?

karpathy/nanoGPT#355

[a1k0n]

With some modifications to scripts/download_and_convert_gpt2.py, you can create the necessary weight bin file.. it downloads a .safetensors file from huggingface, not a pytorch model, though. You might be able to convert with something like this? https://huggingface.co/docs/safetensors/index#save-tensors and get rid of the hf_hub stuff in my script to produce a .bin.