Add backend sampler for penalties sampler #25262
sampling: enhance penalty handling in common_sampler_init
0ce6faf5
tests: add backend penalties sampling tests and utility functions
e82f0e7d
sampling: add support for top-k penalties in backend sampling
af9cf9c4
sampling: add fix to ensure stable numerical results. Preserve maske…
3ffeaeae
sampling: enhance penalty comparison tests with masking penalties logic
888f264d
add comments on padding
1834d9f6
sampling: add comments on modifications
38ad11c7
kmorennv
requested a review
102 days ago
ORippler
changed the title Kmoren/add penalties cu backend Add backend sampler for penalties sampler 96 days ago
kmorennv
changed the title Add backend sampler for penalties sampler CUDA: Add backend sampler for penalties sampler 95 days ago
add the unit test to cover masked-out token as -INF
d7b05513
validate repeat penalty to ensure it is finite and greater than 0; ad…
2adb9571
refactor: test functions to share logic and be less verbose
760b7e55
add test to cover case where previously penalized token is not part o…
65180ab4
remove comments
98b1bea5
remove redundant penalty_last_n initialization and validation in comm…
381226d7
add support for penalties in sampler chain with configurable positions
91341fe1
add validation for penalty parameters and enhance tests for non-finit…
2e54fe32
add context parameter to common_sampler_init and set default for pena…
ec2f07a4
add llama_n_ctx parameter to common_sampler_init for improved sampler…
e650b603
replace penalty_last_n x n_candidates comparison matrix with a vocabu…
cd20503d
add tests for backend penalties sampling without filler entries , tok…
e124d900
add test for backend penalties sampling after top-p with large histo…
39cabdce
Merge branch 'master' into kmoren/add_penalties_cu_backend
b5757809
remove as unused
84326a2e
danbev
commented
on 2026-07-29
danbev
commented
on 2026-07-29
danbev
commented
on 2026-07-29
add is_disabled method, tensor logits reshape, add rest review sugges…
6e0e28d1
clarify comment
3eb4dfc2
danbev
approved these changes
on 2026-07-31
ORippler
approved these changes
on 2026-07-31
ORippler
merged
96278e39
into master 71 days ago
ORippler
changed the title CUDA: Add backend sampler for penalties sampler Add backend sampler for penalties sampler 71 days ago
Login to write a write a comment.
Login via GitHub