-
chunk before retrieval
-
add actions to redis
-
train bge-base
-
refactor(train): change objective from finetuning to downtuning
-
revert to WhereIsAI/UAE-Large-V1 & train
-
refactor(train): sync data
-
add model evaluation
-
train!
-
update AnglE
-
add khoa-klaytn/AnglE submodule
-
refactor(script_download_large): --exclude pytorch_model*
-
refactor(script_download_large): revert to angle-llama-7b-nli-v2
-
add DS_DIR env
-
redo config & add SEED env
-
refactor(script_download_large): revert to WhereIsAI/UAE-Large-V1
-
refactor(script_download_large_files): hardcode models
-
move split to klayassist/ds
-
restrict anchor to action & reduce num_actions to 3
-
refactor(split): reverse engineer: propose split first then calculate
-
refactor(split): reduce scalar to 20
-
calculate splits
-
fix(script_download_large): dont use symlinks
-
add pull.sh
-
fix(.env.example): remove trailing slash from ROOT_DIR
-
fix(script_download_large): remove url prefix
-
add models/.gitkeep
-
revert to models/ to experiment w/ many models
-
hardcode model
-
add models dir
-
initial commit
