Initial commit: FlowDock pipeline configured for WES execution

configs/strategy/ddp.yaml

@@ -0,0 +1,4 @@
+_target_: lightning.pytorch.strategies.DDPStrategy
+static_graph: false
+gradient_as_bucket_view: false
+find_unused_parameters: true

configs/strategy/ddp_spawn.yaml

@@ -0,0 +1,5 @@
+_target_: lightning.pytorch.strategies.DDPStrategy
+static_graph: false
+gradient_as_bucket_view: false
+find_unused_parameters: true
+start_method: spawn

configs/strategy/deepspeed.yaml

@@ -0,0 +1,5 @@
+_target_: lightning.pytorch.strategies.DeepSpeedStrategy
+stage: 2
+offload_optimizer: false
+allgather_bucket_size: 200_000_000
+reduce_bucket_size: 200_000_000

configs/strategy/default.yaml

@@ -0,0 +1,2 @@
+defaults:
+  - _self_

configs/strategy/fsdp.yaml

@@ -0,0 +1,12 @@
+_target_: lightning.pytorch.strategies.FSDPStrategy
+sharding_strategy: ${resolve_variable:torch.distributed.fsdp.ShardingStrategy.FULL_SHARD}
+cpu_offload: null
+activation_checkpointing: null
+mixed_precision:
+  _target_: torch.distributed.fsdp.MixedPrecision
+  param_dtype: null
+  reduce_dtype: null
+  buffer_dtype: null
+  keep_low_precision_grads: false
+  cast_forward_inputs: false
+  cast_root_forward_inputs: true

configs/strategy/optimized_ddp.yaml

@@ -0,0 +1,4 @@
+_target_: lightning.pytorch.strategies.DDPStrategy
+static_graph: true
+gradient_as_bucket_view: true
+find_unused_parameters: false