Michael Wyatt
|
0c4b726aa2
moved megatron and meta tensor support to separate containers, added multi-class inheritance
|
1 年之前 |
Lev Kurilenko
|
5d602d62a2
Fix a few bugs after the base re-org
|
1 年之前 |
Lev Kurilenko
|
a0979feaea
First base container re-org
|
1 年之前 |
Lev Kurilenko
|
381e61b3a3
Move weight quantization into separate function in base container
|
1 年之前 |
Lev Kurilenko
|
da475804c5
Small clean-up
|
1 年之前 |
Lev Kurilenko
|
791935cac9
Apply weight quantization in BERT container
|
1 年之前 |
Lev Kurilenko
|
bc9002ae66
Whitespace in replace_layer to match replace_module
|
1 年之前 |
Lev Kurilenko
|
fbd0b65a65
Remove .scale comments from ds_mlp and ds_attention
|
1 年之前 |
Lev Kurilenko
|
794092e31a
Comprehend quantizer in containers + generic checkpoint works for GPTJ w/ container
|
1 年之前 |
Lev Kurilenko
|
7b65351a07
Use self. for tensors in bloom container
|
1 年之前 |
Lev Kurilenko
|
a8f440e31e
Get generic checkpoint loading working w/o containers
|
1 年之前 |
Lev Kurilenko
|
55e929df8c
Merge branch 'master' into inference-refactor-v1
|
1 年之前 |
Jeff Rasley
|
2600db54f9
bump to 0.7.8
|
1 年之前 |
Jeff Rasley
|
2076bf2311
get mask token from tokenizer (#2592)
|
1 年之前 |
Lev Kurilenko
|
3feff1fd01
Update from model_config to config in replace_module.py
|
1 年之前 |
Jeff Rasley
|
35eabb0a33
Fix issues w. python 3.6 + add py-version checks to CI (#2589)
|
1 年之前 |
Lev Kurilenko
|
52bba028c5
Make megatron_v2 part of base container
|
1 年之前 |
Joe Mayer
|
18713c6838
Updating API docs (#2586)
|
1 年之前 |
Joe Mayer
|
377c770a61
Updating docs README (#2587)
|
1 年之前 |
Lev Kurilenko
|
79643351af
Add generic load checkpoint to replace_layer, still need to add to containers
|
1 年之前 |
Lev Kurilenko
|
4e668887ae
Update all policies for generic load checkpoint
|
1 年之前 |
Lev Kurilenko
|
4479a07b12
Merge branch 'master' into inference-refactor-v1
|
1 年之前 |
Michael Wyatt
|
ccb8eb81fb
Add checkpoint sharding unit tests (#2561)
|
1 年之前 |
lokoppakmsft
|
591744eba3
Support N-dimension input in quantization kernel (#2575)
|
1 年之前 |
Lev Kurilenko
|
105ba4b365
Make containers inherit scale_attention from policy
|
1 年之前 |
Lev Kurilenko
|
28e6b617fc
remove TODO
|
1 年之前 |
Lev Kurilenko
|
992ace2291
Add comprehension of megatron_v2 in gptneox, and rotation config
|
1 年之前 |
Lev Kurilenko
|
eecae6fe92
Properly source megatron_v2 flag from policy in containers
|
1 年之前 |
Lev Kurilenko
|
699336bb31
Change _transpose to transpose_qkv_alignment
|
1 年之前 |
Quentin Anthony
|
18d55e54b0
Update barrier and reduce_scatter_base to conform to PyTorch signatures (#2570)
|
1 年之前 |