Commit History

wrap every layer in a checkpoint
e0da4c5

Markus28 commited on

fix: remove cleaving (#13)
139b4a5
verified

Markus28 commited on

fix: added trust_remote_code to tokenizer
9689b77

Markus28 commited on

fix: fixed from_bert method
151f328

Markus28 commited on

fix: fix LoRA implementation
20706dd

Markus28 commited on

feat: cleave off layers from encoder (#11)
b641603
verified

Markus28 commited on

feat: only apply select_task_for_layer if task has changed
462e28d

Markus28 commited on

feat: make num of loras part of the config
a416a9d

Markus28 commited on

feat: make main parameters trainable
cdf5490

Markus28 commited on

fix BertForMaskedLM
c0b46cc

Markus28 commited on

feat: added separate BertForMaskedLM class
3cb3930

Markus28 commited on

feat: added return_dict
59c0808

Markus28 commited on

fix: fixed syntax error in LoRA
e93b0fd

Markus28 commited on

feat: add current_task to forward
9410275

Markus28 commited on

feat: use property in LoRA parametrization
0ff7c3d

Markus28 commited on

feat: added LoRA copyright notice
faa9951

Markus28 commited on

feat: use property instead of setter
6aad619

Markus28 commited on

feat: return from_bert for from_pretrained
5549314

Markus28 commited on

support-fast-tokenizer (#6)
ed1b276
verified

jupyterjazz commited on

feat: made from_bert work
851184a

Markus28 commited on

feat: choose flash attention heuristically if not set explicitly
2e2b8d0

Markus28 commited on

feat: select first LoRA upon initialization
fabeb13

Markus28 commited on

feat: formatting and type hints
617fe56

Markus28 commited on

fix: use proper initilization for embedding layer
850b9a2

Markus28 commited on

fix: fixed typo
5c4e4bf

Markus28 commited on

feat: added LoRA
8561a1f

Markus28 commited on

feat: assert return_dict
326b1c4

Markus28 commited on

fix: same assertions in other models
c1d92c9

Markus28 commited on

fix: assert is None for other kwargs too
3f5615c

Markus28 commited on

feat: added head_mask
599c64e

Markus28 commited on

added classifier dropout
767b681

Markus28 commited on

fix: formatting
ae4c28c

Markus28 commited on

fix: formatting
f115a1d

Markus28 commited on

feat: added further GLUE models
ec37ae5

Markus28 commited on

feat: added BertForSequenceClassification
ba24fb1

Markus28 commited on

fix: cast mask to bool
ca5f516

Markus28 commited on

reference the flash attention GitHub
eec6c0e

Markus28 commited on

fix: move flash components into top-level
5944ec8

Markus28 commited on

feat: try to fix import error
4c4562b

Markus28 commited on

feat: moved flash attention code into this repository
46df05d

Markus28 commited on

feat: added encode method
32458be

Markus28 commited on

fix: try to skip initialization of task type embeddings
3b35eab

Markus28 commited on

fix: try to skip initialization of task type embeddings
95ca1a8

Markus28 commited on

feat: added option for QK normalization
463061d

Markus28 commited on

fix: removed obscure config options
2e69073

Markus28 commited on