Fix training over existing lora
193c73b
unverified
Angainor Development
commited on
Commit History
new prompters, misc fixes for output dir missing using fsdp, and changing max seq len
4ac9e25
Update doc for grad_accu and add validation tests for batch size
3c71c8d
fix batch size calculation
5a631b3
Merge pull request #124 from OpenAccess-AI-Collective/xformers-fix
2d0ba3b
unverified
Merge pull request #120 from OpenAccess-AI-Collective/model-from-path
c7021e1
unverified
remove unused import and update readme
e3c494c
copy xformers attn from ooba since we removed dep on alpaca_lora_4bit
6cb2310
add support for gradient accumulation steps
3aad5f3
fix up tokenizer config, isort fix
39a208c
split up llama model loading so config can be loaded from base config and models can be loaded from a path
2520ecd
Fix incorrect rebase
594e72b
fix relative path for fixtures
cfcc549
Fix security issue or ignore false positives
a1f9850
Apply isort then black
37293dc
Fix mypy typing
e9650d3
Black formatting
b832a0a
Lint wandb
9c6750a
Lint validation
c2dbf2c
Lint tokenization
e6b57de
Lint schedulers
fe1f4c4
Lint dict
633ff21
Lint trainer.py
ddb86ea
Lint models.py
f4e5d86
Lint callbacks.py
5062eca
Refactor
4c0eddb
Fix data.py lint
cb7cd34
Lint and format
392dfd9
fix relative path for fixtures
e65aeed
new hf_use_auth_token setting so login to hf isn't required
1c33eb8
Feat: Update validate_config and add tests
52dd92a
Apply black formatter
7bf2069
refactor: fix previous refactors
56f9ca5
Refactor to use DictDefault instead
8bd7a49
Add DictDefault
18d41ce
Convert attrdict to addict
bdfe7c9
Merge pull request #67 from OpenAccess-AI-Collective/refactor-tokenizer-load
0d4a7f4
unverified
Add warning for `trust_remote_code`
2824423
unverified
Merge branch 'main' into refactor/rename-4b-to-gptq
147241c
unverified
fix auto linear modules for lora w/o any set already
4c90633
checking for False is not sufficent for NoneType/unset configs
4c500f5
refactor(param): rename load_4bit config param by gptq
dd00657
Thytu
commited on