Releases: allenai/OLMo-in-loop-evals
Releases · allenai/OLMo-in-loop-evals
v0.5.0
What's new
Changed ⚠️
- You can pass
None
for lm_logits
in ICLMetric.update()
to support pipeline parallelism.
Commits
ba30253 (chore) prepare for release v0.5.0
193fe15 Add support for pipeline parallelism (#3)
72d8a83 (chore) prepare for release v0.4.0
5f3db3c Sync eval changes in OLMo/ladder-1xC to here (#1)
v0.3.0
What's new
Added 🎉
- Allowed passing additional kwargs to the task through
build_task()
.
- Added the option to fix the context length of every batch to the model's context length.
Commits
35f072a Allow fixing sequence length to the model's context length (#2)
v0.2.0
What's new
Added 🎉
- Added
ICLMultiChoiceTaskDataset.max_sequence_length
property.
Commits
ee86f3b Add ICLMultiChoiceTaskDataset.max_sequence_length
property
v0.1.0
What's new
Added 🎉
- Added in-loop evals from original OLMo repo.
Commits
bfdc809 (chore) prepare for release v0.1.0
302e3a4 add license
9b5f070 make mypy happy
78de296 Add readme
b1c8aea initial commit