Port Reza's INT8-quantization fix to container architecture #2725
jeffra
merged
20 commits into
master
from
lekurile/container_int8_fix
support int8 for neox
d5513e9d
fix some issues with the replace policies
88a8b149
Merge branch 'master' into neox-q-int8
376b854b
fix merging checkpoint for qkv params
b8d731d4
fix the transposition when using int8
0d201876
reduce the ckpt-loading code
85ab3d10
fix some issue with GPT-J and OPT ckpt-loading
5658141b
Merge branch 'master' into neox-q-int8
847c85f0
Merge branch 'master' into neox-q-int8
77735015
Merge branch 'neox-q-int8' into lekurile/container_int8_fix
8403165e
Port Rezas int8 fix to container architecture
839fc3d7
Fix load_params abstract method + imports in policies
ba37b743
Create new 'ds_inf_config' to differentiate between passed in config
5533125e
Merge branch 'master' into lekurile/container_int8_fix
d0d5829e
replace meta tensor when loading bias
3415d520
add assert to prevent slient failure of copy_() on meta tensor
c1e5746b
Merge branch 'master' into lekurile/container_int8_fix
6a64e1a6
RezaYazdaniAminabadi
approved these changes
on 2023-02-13
Merge branch 'master' into lekurile/container_int8_fix
90b3b4ea
lanking520
commented
on 2023-02-13
Merge branch 'master' into lekurile/container_int8_fix
803e3761
Merge branch 'master' into lekurile/container_int8_fix
2248e721
disabled auto-merge 2 years ago
Manually disabled by user
Manually disabled by user
Assignees
No one assigned
Labels
None yet
Milestone
No milestone
Login to write a write a comment.
Login via GitHub