Optimize load distribution between nodes #719
Open
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Related to exo-explore#701 Implement load distribution between nodes based on flops and memory. * **Partitioning Strategy**: - Add `flops` attribute to `Partition` class in `exo/topology/partitioning_strategy.py`. - Update `map_partitions_to_shards` function to consider `flops` when mapping partitions to shards. - Add `get_flops` method to `PartitioningStrategy` class to calculate the flops of each partition. * **Ring Memory Weighted Partitioning Strategy**: - Update `partition` method in `exo/topology/ring_memory_weighted_partitioning_strategy.py` to consider both memory and flops for partitioning. - Add `calculate_flops_weight` helper function to calculate the weight of each node based on its flops. * **Node Class**: - Update `Node` class in `exo/orchestration/node.py` to implement logic to sort nodes by flops for load distribution. - Add `sort_nodes_by_flops` method to sort nodes by their flops. * **Inference Engine**: - Add `get_flops` method to `InferenceEngine` class in `exo/inference/inference_engine.py` to get the flops of the current node. * **Sharded Inference Engine**: - Add `get_flops` method to `MLXDynamicShardInferenceEngine` class in `exo/inference/mlx/sharded_inference_engine.py` to get the flops of the current node.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Related to #701
Implement load distribution between nodes based on flops and memory.
Partitioning Strategy:
flopsattribute toPartitionclass inexo/topology/partitioning_strategy.py.map_partitions_to_shardsfunction to considerflopswhen mapping partitions to shards.get_flopsmethod toPartitioningStrategyclass to calculate the flops of each partition.Ring Memory Weighted Partitioning Strategy:
partitionmethod inexo/topology/ring_memory_weighted_partitioning_strategy.pyto consider both memory and flops for partitioning.calculate_flops_weighthelper function to calculate the weight of each node based on its flops.Node Class:
Nodeclass inexo/orchestration/node.pyto implement logic to sort nodes by flops for load distribution.sort_nodes_by_flopsmethod to sort nodes by their flops.Inference Engine:
get_flopsmethod toInferenceEngineclass inexo/inference/inference_engine.pyto get the flops of the current node.Sharded Inference Engine:
get_flopsmethod toMLXDynamicShardInferenceEngineclass inexo/inference/mlx/sharded_inference_engine.pyto get the flops of the current node.