jprafael/mpt-7b-instruct-sharded

Update README.md

Replace model with mpt-7b-instruct, loaded in f16 and sharded to 2GB chunks

update commit to use for revision

use_cache by default

✨ gradient checkpointing

add einops

add details on usage

🔧 add a requirements from pipreqs

🎨 format for readability

add MPTBlock to _no_split_modules

Update README.md

increase max_new_tokens default

format

initial support for device_map=auto

better generation params

Update README.md

Update README.md

add sharded checkpoint

initial commit