transformers
Generate: Add new decoding strategy "DoLa" in `.generate()`
#29619
Merged
Go
Login via GitHub
Home
Pricing
FAQ
Install
Login
via GitHub
Overview
Commits
33
Changes
View On
GitHub
Commits
add support for DoLa decoding
gante
committed
1 year ago
add docs; remove deprecated function
gante
committed
1 year ago
add test code for DoLa decoding
gante
committed
1 year ago
update docs and paper link
gante
committed
1 year ago
solved the issues that made tests failed on CircleCI
gante
committed
1 year ago
ruff reformatted
gante
committed
1 year ago
update DoLa decoding; test cases for llama/mistral/mixtral/gemma; docs
gante
committed
1 year ago
fix formatting; fix failed test cases
gante
committed
1 year ago
fix other test cases
gante
committed
1 year ago
ruff reformatted; mamba cache issue unsolved
gante
committed
1 year ago
remove keyword argument 'model_inputs' to match upstream changes
gante
committed
1 year ago
improve documentation
gante
committed
1 year ago
fixed suggestions from @gante
gante
committed
1 year ago
ruff reformated
gante
committed
1 year ago
moved config warning of dola generation from `utils.py` to `configuration_utils.py`
gante
committed
1 year ago
fixed suggestions from @amyeroberts
gante
committed
1 year ago
fixed format issue; removed print; added explanation
gante
committed
1 year ago
remove trailing whitespace
gante
committed
1 year ago
ruff reformat to pass test
gante
committed
1 year ago
fixed suggestions from @amyeroberts on Mar 28
gante
committed
1 year ago
fix failed CI tests
gante
committed
1 year ago
ruff reformatted; fixed missing argument generation_config
gante
committed
1 year ago
make `dola_layers` not optional
gante
committed
1 year ago
fix divergence w main
gante
committed
1 year ago
fix dola test on mamba
gante
committed
1 year ago
rwkv test (wont fix
gante
committed
1 year ago
slow tests running in fp16
gante
committed
1 year ago
make fixup
gante
committed
1 year ago
remove redundant fn
gante
committed
1 year ago
final rebase divergences
gante
committed
1 year ago
this one was missing
gante
committed
1 year ago
a few more nits
gante
committed
1 year ago
skip stateful models
gante
committed
1 year ago
Loading