 AlpinDale
|
2242b356a0
apply dry first
|
1 mēnesi atpakaļ |
|
AlpinDale
|
2242c5756a
add co-author
|
1 mēnesi atpakaļ |
|
AlpinDale
|
224219dd6d
fix dry penalties
|
1 mēnesi atpakaļ |
|
AlpinDale
|
22424ec91c
fix: init the prompt and output token tensors for dry
|
1 mēnesi atpakaļ |
|
AlpinDale
|
2242958d91
fix: sequence breaker ids are a list of int
|
1 mēnesi atpakaļ |
|
AlpinDale
|
22428f934d
formatting
|
1 mēnesi atpakaļ |
|
AlpinDale
|
224256d679
take sequence breakers as both a string literal list and a List[int]
|
1 mēnesi atpakaļ |
|
AlpinDale
|
22427af50b
Merge branch 'main' into dry-sampler
|
1 mēnesi atpakaļ |
 AlpinDale
|
3d83e64f8e
feat: add metrics for prefix cache hit rate (#829)
|
1 mēnesi atpakaļ |
|
AlpinDale
|
22425b689d
fix: XPU build
|
1 mēnesi atpakaļ |
|
AlpinDale
|
bfc8988116
feat: add cuda sampling kernels for top_k and top_p (#828)
|
1 mēnesi atpakaļ |
 Selali Adobor
|
aa7e47cf92
Revert "Refactor based on closer examination of other examples"
|
1 mēnesi atpakaļ |
|
Selali Adobor
|
d9562b96f8
Refactor based on closer examination of other examples
|
1 mēnesi atpakaļ |
|
Selali Adobor
|
07e1175587
Fix naming bug
|
1 mēnesi atpakaļ |
|
Selali Adobor
|
c63672dce3
Fix refactor bug
|
1 mēnesi atpakaļ |
|
Selali Adobor
|
81ae51b1e8
Set useful defaults for DRY API
|
1 mēnesi atpakaļ |
|
Selali Adobor
|
5c6fa79193
feat(sampling): Add DRY (Do not Repeat Yourself) sampling
|
1 mēnesi atpakaļ |
|
AlpinDale
|
22427602eb
feat: add top-nsigma sampling method
|
1 mēnesi atpakaļ |
|
AlpinDale
|
22429e4a10
fix: sampler test with new transformers version
|
1 mēnesi atpakaļ |
|
AlpinDale
|
2f61644f6e
SPMD optimizations (#824)
|
1 mēnesi atpakaļ |
|
AlpinDale
|
32a37e8107
tests: partially fix tensorizer and logprobs tests
|
1 mēnesi atpakaļ |
|
AlpinDale
|
7f1c9af5e2
fix: fp8 quant test
|
1 mēnesi atpakaļ |
|
AlpinDale
|
173ac23399
fix: experts int8 quant test
|
1 mēnesi atpakaļ |
|
AlpinDale
|
68f050129d
fix: lora worker manager test import
|
1 mēnesi atpakaļ |
|
AlpinDale
|
3661de812d
fix: lora layer test
|
1 mēnesi atpakaļ |
|
AlpinDale
|
0a369f9171
feat: support chunked prefill with LoRA (#823)
|
1 mēnesi atpakaļ |
|
AlpinDale
|
e5b1afe625
feat: add chat method for LLM class (#822)
|
1 mēnesi atpakaļ |
|
AlpinDale
|
262cbc63b7
fix: vision api test template path
|
1 mēnesi atpakaļ |
|
AlpinDale
|
b0113a1eaa
fix: tokenization api test (#821)
|
1 mēnesi atpakaļ |
|
AlpinDale
|
c6c91edab7
ci: update & overhaul test units (#769)
|
1 mēnesi atpakaļ |
