Model downloaded through hugging-face CLI cannot be loaded using AutoModel classes
I downloaded the model using the huggingface-cli command:huggingface-cli download meta-llama/Llama-3.3-70B-Instruct --include "original/*" --token hf_xxx --local-dir /Models/Llama-3.3-70B-Instruct
The downloaded files contain consolidated_xx.pth files.
When trying to load from the same path using AutoModelForCausalLM I encounter the error:Should have a
model_type key in its config.json
Does the downloaded weights need to be converted to safetensor files for it to be loadable by transformers? If so, how to do this? If not, how can I use the downloaded weights to load the model?
I think you want to download the safetensors that are stored in the parent directory, ie you should exclude "original/*" rather than include it
I think you want to download the safetensors that are stored in the parent directory, ie you should exclude "original/*" rather than include it
Yes, I realized that. However, that needs a re-download. I was asking for a way to make use of the .pth files I already downloaded.
Pretty sure you have an error in the --local-dir flag.
You have a value of;
--local-dir /Models/Llama-3.3-70B-Instruct
Where from the command I used below the value would be;
--local-dir meta-llama/Llama-3.3-70B-Instruct
This command is functioning for me;
huggingface-cli download meta-llama/Llama-3.3-70B-Instruct --include "original/*" --local-dir meta-llama/Llama-3.3-70B-Instruct
Also does the --token flag work in these commands (as for logging in without the huggingface-cli login command?)