{ | |
"auto_loss_scaling": false, | |
"device_iterations": 1, | |
"embedding_serialization_factor": 2, | |
"enable_half_partials": true, | |
"executable_cache_dir": "./exe_cache", | |
"gradient_accumulation_steps": 2048, | |
"inference_device_iterations": 5, | |
"inference_replication_factor": {"pod4": 1, "pod8": 2, "pod16": 4, "pod32": 8, "pod64": 16, "default": 4}, | |
"ipus_per_replica": 4, | |
"layers_per_ipu": [ | |
0, | |
4, | |
4, | |
4 | |
], | |
"matmul_proportion": 0.1, | |
"optimizer_state_offchip": true, | |
"optimum_version": "1.4.1", | |
"output_mode": "final", | |
"recompute_checkpoint_every_layer": true, | |
"replicated_tensor_sharding": true, | |
"replication_factor": {"pod4": 1, "pod8": 2, "pod16": 4, "pod32": 8, "pod64": 16, "default": 4}, | |
"seed": 42 | |
} | |