How to convert the weights of pyllama 7B into lit-llama?

[mmdrahmani]

Hi. Thanks for the great work.
You provided this instruction below:
For using the LLaMa model weight, follow [pyllama](https://github.com/juncongmoo/pyllama) to download the original LLaMA model, and then follow [Lit-LLaMA](https://github.com/Lightning-AI/lit-llama) to convert the weights to the Lit-LLaMA format. After this process, please move the lit-llama/ directory under the checkpoints/ directory.
However, when I check the Lit-LLaMA repository, I cannot find any explanation for how to convert the weights of pyllama 7B (consolidated.00.pth) into lit-llama format. Could you share the link/instructions?
Thank you
Mohammad

[41xu]

see this: https://github.com/Lightning-AI/lit-llama/blob/main/howto/download_weights.md
convert the weights to the lit-llama format part
in lit-llama:

python scripts/convert_checkpoint.py --model_size 7B

@mmdrahmani

[mmdrahmani]

@41xu Thanks.
The downloaded llama 7B consolidated.00.pth is around 12GB. When I convert it to lit-llama format, it becomes 26GB.
Is this correct?

I download the model from huggingface, shown below, because the python -m llama.download --model_size 7B gets stuck.
wget https://huggingface.co/nyanko7/LLaMA-7B/resolve/main/consolidated.00.pth

[41xu]

HI @mmdrahmani
for me, the situation is the same. the process of downloading and converting weights made me wonder if my network was interrupted. it needs a long time (if i remember correctly)

@41xu Thanks. The downloaded llama 7B consolidated.00.pth is around 12GB. When I convert it to lit-llama format, it becomes 26GB. Is this correct?

I download the model from huggingface, shown below, because of the python -m llama.download --model_size 7B gets stuck. wget https://huggingface.co/nyanko7/LLaMA-7B/resolve/main/consolidated.00.pth