 AlpinDale
|
afb710e3e4
dry: use gpu for z_array computation
|
hai 1 mes |
|
AlpinDale
|
f72f70cc02
dry: only apply dry to sequences that request it
|
hai 1 mes |
 AlpinDale
|
0035dc42ed
sampler: optimize DRY performance using z-algorithm (#856)
|
hai 1 mes |
|
AlpinDale
|
2150bb5019
sampler: add range parameter for DRY (#855)
|
hai 1 mes |
|
AlpinDale
|
72c505ad84
sampler: fix dry concurrency issue (#852)
|
hai 1 mes |
 Selali
|
14ac216498
sampler: add output_tokens to DRY sampler (#849)
|
hai 1 mes |
|
AlpinDale
|
dfa34d1b24
feat: add sampler_priorty (#837)
|
hai 1 mes |
|
AlpinDale
|
60f7b828d5
feat: add skew sampling (#834)
|
hai 2 meses |
|
AlpinDale
|
ba9d8f631a
feat: add no_repeat_ngram sampler (#832)
|
hai 2 meses |
|
Selali
|
4c4a365f77
feat: Add DRY (Don't Repeat Yourself) sampling (#827)
|
hai 2 meses |
|
AlpinDale
|
bfc8988116
feat: add cuda sampling kernels for top_k and top_p (#828)
|
hai 2 meses |
|
AlpinDale
|
22427602eb
feat: add top-nsigma sampling method
|
hai 2 meses |
 50h100a
|
273c61d406
guard against nan temperature from dynatemp (or anywhere else).
|
hai 2 meses |
|
50h100a
|
a22e887319
why we don't use the github website editor to make changes
|
hai 2 meses |
|
50h100a
|
54a8320638
logit shenanigans to prevent even worse shenanigans
|
hai 2 meses |
|
50h100a
|
b6a897d2a1
fix temperature, and address those pernicious dynatemp NaNs
|
hai 2 meses |
|
50h100a
|
83040c6389
Mask dynatemp using min/max, rather than exp
|
hai 2 meses |
 Pyroserenus
|
99fc6f4697
fix: crash in token bans (#764)
|
hai 3 meses |
|
AlpinDale
|
ad181e3fef
feat: bring back dynatemp (#754)
|
hai 4 meses |
|
AlpinDale
|
6329c2d53f
chore: re-enable custom token bans (#751)
|
hai 4 meses |
|
AlpinDale
|
abd9d5799a
feat: add XTC Sampling (#740)
|
hai 4 meses |
|
AlpinDale
|
198029295c
fix: empty sampler output when temperature is too low (#709)
|
hai 4 meses |
|
AlpinDale
|
1394008421
chore: decouple `should_modify_greedy_probs_inplace (#671)
|
hai 4 meses |
|
AlpinDale
|
2da6a3ec2b
feat: option to apply temperature scaling last (#670)
|
hai 4 meses |
 50h100a
|
cbdf2d986f
quadratic sampling: separate diff from logits to avoid NaNs.
|
hai 4 meses |
|
AlpinDale
|
f1d0b77c92
[0.6.0] Release Candidate (#481)
|
hai 4 meses |
|
AlpinDale
|
9d81716bfd
[v0.5.3] Release Candidate (#388)
|
hai 8 meses |
|
50h100a
|
f663d3fccc
Merge pull request #397 from 50h100a/pr_samplerasserts
|
hai 9 meses |
|
50h100a
|
85ae23ac3c
Missed .items() and assert
|
hai 9 meses |
|
50h100a
|
43c9858854
Merge pull request #244 from PygmalionAI/faster_topk
|
hai 9 meses |
