AlpinDale
|
8fa608aeb7
feat: replace Ray with NCCL for control plane comms (#221)
|
1 year ago |
Stefan Gligorijevic
|
9e7e108dc8
chore: clamp dynatemp_min (#214)
|
1 year ago |
Stefan Gligorijevic
|
56446a04bb
feat: dynamic temperature (#209)
|
1 year ago |
AlpinDale
|
d54791aaa8
feat: reduce sampler overhead by making it less blocking (#198)
|
1 year ago |
AlpinDale
|
653da510d1
chore: rewrite InputMetadata (#143)
|
1 year ago |
AlpinDale
|
6c914ea0e4
fix: `SequenceOutputs` -> `SequenceOutput` (#133)
|
1 year ago |
Stefan Gligorijevic
|
a5255901c6
feat: min_p sampling (#106)
|
1 year ago |
AlpinDale
|
ae7d8df224
fix lint issues (again)
|
1 year ago |
50h100a
|
fa0ae5a2c9
feat: new mirostatv2 implementation (#96)
|
1 year ago |
AlpinDale
|
69204736de
Revert "fix: sync CPU delay in sampler (#93)"
|
1 year ago |
AlpinDale
|
ce66e1df56
fix: sync CPU delay in sampler (#93)
|
1 year ago |
AlpinDale
|
74604eb252
fix: pylint complaints (#91)
|
1 year ago |
AlpinDale
|
efc6f7fbec
chore: reformats (#90)
|
1 year ago |
AlpinDale
|
3d72f05c7b
feat: flattened 1D tensor -> 2D tensor (#85)
|
1 year ago |
AlpinDale
|
e6be0118c9
feat: prompt logprobs and batched samplers (#77)
|
1 year ago |
Stefan Gligorijevic
|
34c1c8c83a
feat: Enable banning tokens (#80)
|
1 year ago |
50h100a
|
d0eadd4dbd
Added `min_tokens` and reimplemented `ignore_eos` using a new logit processor (#70)
|
1 year ago |
AlpinDale
|
04a27c6aeb
fix: revert mirostat v2 (#79)
|
1 year ago |
Stefan Gligorijevic
|
5dbd262033
feat: Mirostat v2 (#69)
|
1 year ago |
AlpinDale
|
3bf6197afb
fix: prompt processing delay introduced by #66 (#71)
|
1 year ago |
AlpinDale
|
380206038e
fix: change the timing of logit sorting (#66)
|
1 year ago |
AlpinDale
|
a6a4220fa6
feat: refactor megatron and quants (#57)
|
1 year ago |
Stefan Gligorijevic
|
0feaa6eba8
Merge branch 'main' into samplers-next
|
1 year ago |
50h100a
|
b1bbec5625
Merge branch 'main' of https://github.com/PygmalionAI/aphrodite-engine into new_samplers
|
1 year ago |
50h100a
|
633b99d266
Merge branch 'new_samplers' of https://github.com/50h100a/aphrodite-engine into new_samplers
|
1 year ago |
50h100a
|
490e14c038
function rename
|
1 year ago |
Stefan Gligorijevic
|
8a6c9f5cbd
fix eta,eps and typical for parallel requests
|
1 year ago |
Stefan Gligorijevic
|
99f76323ad
Misc fixes in eta, eps, and typical
|
1 year ago |
AlpinDale
|
022380e896
`entropy_deviation` -> `surprisal_deviation`
|
1 year ago |
Stefan Gligorijevic
|
048572e4dd
Add eta, epsilon, and locally typical sampling
|
1 year ago |