AlpinDale
|
39beed0b87
Revert "Refactor AWQ support."
|
1 year ago |
AlpinDale
|
579071b570
Revert "fix the awq gemm kernels"
|
1 year ago |
AlpinDale
|
663dd09399
Revert "fix: detokenization with special tokens"
|
1 year ago |
AlpinDale
|
20c27863c1
fix the awq gemm kernels
|
1 year ago |
AlpinDale
|
d09e27f5d4
Refactor AWQ support.
|
1 year ago |
AlpinDale
|
cc1d5339dd
fix: detokenization with special tokens
|
1 year ago |
AlpinDale
|
6dfca14e1f
compute logprobs with log_softmax instead of log
|
1 year ago |
AlpinDale
|
6b9561ef07
adapt TGI incremental detokenization
|
1 year ago |
AlpinDale
|
d4cd18bd94
chore: allow user to specify model context length
|
1 year ago |
AlpinDale
|
0115e55972
chore: add max log length
|
1 year ago |
AlpinDale
|
e77960c57e
use float datatype for RoPE
|
1 year ago |
AlpinDale
|
57b5ef31e7
fix: wrong dtype in bias
|
1 year ago |
AlpinDale
|
d71a84b780
fix: ModuleNotFoundError for remote code models
|
1 year ago |
AlpinDale
|
2399cbd3e6
feat: bump up the version to 0.2.1
|
1 year ago |
AlpinDale
|
d949dd306f
add api changes
|
1 year ago |
AlpinDale
|
7a85354b69
add logits back
|
1 year ago |
AlpinDale
|
15a4071e77
Merge pull request #12 from PygmalionAI/feat/refactor
|
1 year ago |
AlpinDale
|
45f6d9f923
initial refactor commit
|
1 year ago |
AlpinDale
|
23389d0108
zero out a variable instead of vector in kernels
|
1 year ago |
AlpinDale
|
bdf264880f
clean up safetensors support
|
1 year ago |
AlpinDale
|
826de3ef93
use flash attention with xformers
|
1 year ago |
AlpinDale
|
aba8b0b17a
add rope theta support and bump transformers
|
1 year ago |
AlpinDale
|
c687430ce7
bump xformers and clean up leftover code
|
1 year ago |
AlpinDale
|
32844c1522
add GELU kernels and remove compile bloat
|
1 year ago |
AlpinDale
|
c318602c42
update setuptools
|
1 year ago |
AlpinDale
|
0aa5d13909
clear cuda cache and state
|
1 year ago |
AlpinDale
|
62dfd1a883
fix top_k implementation
|
1 year ago |
AlpinDale
|
1482542239
fix: forgot an import
|
1 year ago |
AlpinDale
|
f3f31434c6
Merge pull request #10 from PygmalionAI/feat/safetensor-support
|
1 year ago |
AlpinDale
|
91abae0631
fix: typo
|
1 year ago |