Hardcoded llama65 multigpu with accelerate

EleutherAI / elk

Keeping language models honest by directly eliciting knowledge encoded in their activations.

MIT License

178 stars 33 forks source link

Closed thejaminator closed 1 year ago

thejaminator commented 1 year ago

Try it with elk elicit huggyllama/llama-65b imdb --num_gpus 8