Add generation caching in TextEnvironment and fix bugs in TextEnvironment#2556

Open

konrad-gerlach wants to merge 50 commits intohuggingface:mainfrom konrad-gerlach:text_environment_caching

+803-73

Commits on Jan 10, 2025

feat: add caching for TextEnvironment and fix bugs
Konrad Gerlach
committed
feat: make TextEnvironment caching optional and add documentation
Konrad Gerlach
committed
fix: failing TextEnvironment tests
Konrad Gerlach
committed
test: add tests for TextEnvironment caching and fix cache combining bug
Konrad Gerlach
committed
test: remove unnecessary parametrized class decorator
Konrad Gerlach
committed
docs: update TextEnvironmentDocs with caching
Konrad Gerlach
committed
fix: run linter on TextEnvironment and TextEnvironment tests
Konrad Gerlach
committed
fix: comment
Konrad Gerlach
committed
fix: Args comment
Konrad Gerlach
committed
fix: TextEnvironment cache combination and batching issue
Konrad Gerlach
committed
tests: make caching test more complex
Konrad Gerlach
committed

Commits on Jan 11, 2025

fix: combine caches of different sequence lengths
Konrad Gerlach
committed

Commits on Jan 12, 2025

Commits on Jan 14, 2025

fix: attended to invalid last generated token and off-by-one in StringStoppingCriteria
Konrad Gerlach
committed

Commits on Jan 15, 2025

Merge branch 'main' into text_environment_caching
konrad-gerlach
authored

Commits on Jan 17, 2025

Merge branch 'main' into text_environment_caching
konrad-gerlach
authored

Commits on Jan 21, 2025

fix: off by one error in StringStoppingCriteria
Konrad Gerlach
committed
Merge branch 'main' into text_environment_caching
konrad-gerlach
authored

Commits on Jan 22, 2025

feat: test logits are same with and without caching
Konrad Gerlach
committed
fix: model and tokenizer were called gpt2 but were another model
Konrad Gerlach
committed
docs: add warning for torch.compile with TextEnvironment use_cache
Konrad Gerlach
committed
Merge branch 'text_environment_caching' of https://github.com/konrad-gerlach/trl into text_environment_caching
Konrad Gerlach
committed

Commits on Jan 23, 2025

fix: StringStoppingCriteria and add test
Konrad Gerlach
committed
refactor: move StoppingCriteria test
Konrad Gerlach
committed
feat: add support for models without cache class support
Konrad Gerlach
committed
refactor: make caching code optional in TextEnvironment
Konrad Gerlach
committed
docs: TextEnvironment use_cache note untested Encoder-Decoder architecture
Konrad Gerlach
committed

Commits on Jan 24, 2025

refactor: extract method from _generate_batched
Konrad Gerlach
committed
feat: add optional logits to TextHistory and remove eos segment created by truncation
Konrad Gerlach
committed
tests: remove redundant test code
Konrad Gerlach
committed
tests: update logit similarity calculation
Konrad Gerlach
committed
feat: pad each batch individually in TextEnvironment
Konrad Gerlach
committed
refactor: remove redundant past_input_ids manipulation in TExtEnvironment caching
Konrad Gerlach
committed
fix: remove redundant attention blocking for padding and eos token and add tests for utility methods in TextEnvironment
Konrad Gerlach
committed

Commits on Jan 25, 2025

fix: code and documenation cleanup for TextEnvironment caching
Konrad Gerlach
committed

Commits on Jan 26, 2025

Commits on Jan 30, 2025