Skip to content
GitLab
Explore
Sign in
Overview
Active
Stale
All
jiecaoyu_fp8allgather_debug
8bebf152
·
[Not for merge] FP8 allgather changes for debugging in fairscale
·
Nov 08, 2023
pp_fp8_freezing
ed366c9c
·
v2
·
Nov 27, 2023
artk/fix_fsdp_for_8gpu_cases
2ff977dd
·
fix no shard case
·
Nov 28, 2023
keep_grads_fp32
768c5a84
·
changes to keep reduced grad in fp32
·
Dec 04, 2023
jiecaoyu_temp_fp32_main_grad_fix
f79d5014
·
fix .grad=None issue when param is not sharded
·
Dec 05, 2023
ngoyal_main_grad_fix
476aa945
·
fixed broken clipping
·
Dec 06, 2023
ngoyal_temp_fp32_main_grad
30cd0211
·
fixed broken clipping (#1154)
·
Dec 06, 2023
ngoyal_changes_for_pp_fp8
d0b506f7
·
added option for no PG validation for faster init (#1161)
·
Jan 24, 2024
ngoyal_changes_for_pp_fp8_akadian
73df9be2
·
Added reshard hook for frozen params in backward
·
Feb 06, 2024
ngoyal_changes_for_pp_fp8_akadian_jie_fix
5baa10cb
·
Fix silenet errors when requires_grad=False in params_with_grad in FSDP
·
Mar 04, 2024
dependabot/pip/black-24.3.0
2ea9dc28
·
Bump black from 22.3.0 to 24.3.0
·
Mar 20, 2024
ngoyal_changes_for_pp_fp8_jiecaoyu_norm_allreduce
74a7313f
·
Support for only performing norm weights allreduce in last microbatch | fairscale
·
Mar 22, 2024
add_context_parallel_init
66d75ce9
·
add context parallel group init to mp init
·
Apr 03, 2024
ngoyal_changes_for_pp_fp8_jiecaoyu_free_fp16_shard_cast_input
1056da98
·
Add cast input argument
·
Apr 03, 2024
ngoyal_changes_for_pp_fp8_jiecaoyu_debug
7bcbc805
·
Add cast input argument (#1175)
·
Apr 05, 2024
add-get_cp_ranks
09dbfb1f
·
add get_cp_ranks to model_parallel initialize
·
Apr 05, 2024
shikaili_fp8_allgather
57eb557a
·
Clean up shard offset calculation logic.
·
Apr 10, 2024
selective_rebuild_modulelist
19893653
·
Only rebuilds parameters of selected module.
·
Apr 13, 2024
ngoyal_changes_for_pp_fp8_jiecaoyu_free_fp16_shard_shard_vision_encoder
eb96cca2
·
mm numerics Apr 19 fairscale
·
Apr 19, 2024
chriscai_ngoyal_changes_for_pp_fp8
d95ca8da
·
remove .clone()
·
Apr 27, 2024
Prev
1
2
3
Next