Activity
Update gke tpu job to bypass jobset coordinator
Update gke tpu job to bypass jobset coordinator
Update pathways container specs for gke tpu job
Update pathways container specs for gke tpu job
Revert updates to fuji model config
Revert updates to fuji model config
remove unnecessary updates to gke tpu job for pathways workloads
remove unnecessary updates to gke tpu job for pathways workloads
revert changes to trainer and model configs
revert changes to trainer and model configs
refactor pathways jobset to new spec
refactor pathways jobset to new spec
Rebase to axlearn main
Rebase to axlearn main
Introduce BaseAttentionBias.has_value()
. (apple#920)
Introduce
BaseAttentionBias.has_value()
. (apple#920)Implements FlashDecoding with Sparsity Support (apple#899)
Implements FlashDecoding with Sparsity Support (apple#899)
refactor pathways jobset to new spec
refactor pathways jobset to new spec
Implements FlashDecoding with Sparsity Support (apple#899)
Implements FlashDecoding with Sparsity Support (apple#899)
Apply formatting
Apply formatting
Refactor jobset to align with new pathways structure
Refactor jobset to align with new pathways structure
Update compiler options validation
Update compiler options validation
Add remat policy for fuji-70B on tpu-v6e
Add remat policy for fuji-70B on tpu-v6e
Support v6e (apple#879)
Support v6e (apple#879)
dump xla flags to gcs
dump xla flags to gcs
Enable host network on gke pod spec
Enable host network on gke pod spec
Merge branch 'apple:main' into pathways_trillium
Merge branch 'apple:main' into pathways_trillium
Support orbax state builder (apple#866)
Support orbax state builder (apple#866)
update pathways jobset definition
update pathways jobset definition
Temp change to use Orbax checkpointer for Fuji
Temp change to use Orbax checkpointer for Fuji
Launch pathways on trainer_main
Launch pathways on trainer_main
Merge branch 'apple:main' into pathways_trillium
Merge branch 'apple:main' into pathways_trillium
merge trillium changes
merge trillium changes
pathways jobset updates
pathways jobset updates
Deleted branch
snapshot (apple#854)
snapshot (apple#854)
snapshot (apple#854)
snapshot (apple#854)
Fix RLHF slowdown in attention multi steps extend_step. (apple#849)
Fix RLHF slowdown in attention multi steps extend_step. (apple#849)