fix: update frequencies when updating the rope base value (#40)

feat: update the readme

fix mixed precision loading with recent transformers versions (#39)

lora-instructions (#36)

fix-device-issues (#38)

fix: override use_flash_attn in lora

Fix rotary backward

cpu-inference (#35)

feat: configurable use_reentrant (#37)

refine-codebase (#33)

fix-adapter-masks (#32)

change rotary base (#31)

2-adapter-tuning (#29)

Update modeling_xlm_roberta.py

2-adapter-tuning-initial-impl (#30)

Update modeling_xlm_roberta.py

make-lora-stateless (#25)

feat-routing (#26)

fix-rope-inference (#28)

no-flash-attention-during-inference (#22)

Rename config to config.json

Create config

Delete config.json

rope-embeddings (#20)

alibi (#19)

lora bugfix (#16)

Update config.json (#15)

Update tokenizer_config.json (#14)

Update modeling_xlm_roberta.py (#13)

Update modeling_xlm_roberta.py (#12)

truncate-embedding-dimension (#10)

lora-multiple-adapters (#11)

Refactor LoRA (#8)

support lora (#1)

Support for SequenceClassification (#7)

Support torch_dtype and CLS pooling (#6)

Update modeling_xlm_roberta_for_glue.py

Create modeling_xlm_roberta_for_glue.py (#4)

add-encode-function (#5)

support-cpu (#2)

Update configuration_xlm_roberta.py

Update configuration_xlm_roberta.py

add stochastic_depth

support activation checkpointing

add mlm model and adjust naming

add script to convert weights

add tokenizer class

change config name

upload model

initial commit