this is VERY hacky for now. It's going to require more work to make this be clean and work well. We need to implement input sharding because currently all videos are fed as a constant stream into the SharedQueue based on which finish reading first so it's hard to determine which videos are in which shard. I temporarily solved this by iterating over 'fake shards' by just indexing the full arrays
this is VERY hacky for now. It's going to require more work to make this be clean and work well. We need to implement input sharding because currently all videos are fed as a constant stream into the SharedQueue based on which finish reading first so it's hard to determine which videos are in which shard. I temporarily solved this by iterating over 'fake shards' by just indexing the full arrays