{ "embedding_serialization_factor": 1, "recompute_checkpoint_every_layer": false, "optimizer_state_offchip": true, "replicated_tensor_sharding": true, "enable_half_partials": true, "device_iterations": 1, "inference_device_iterations": 4, "replication_factor": 1, "inference_replication_factor": 1, "gradient_accumulation_steps": 32, "executable_cache_dir": "./exe_cache", "ipus_per_replica": 8, "layers_per_ipu": [4,5,3,4,4,4,4,5], "matmul_proportion": [0.2,0.2,0.1,0.1,0.1,0.1,0.1,1.0] }