|
{ |
|
"auto_loss_scaling": true, |
|
"device_iterations": 1, |
|
"embedding_serialization_factor": 1, |
|
"enable_half_partials": true, |
|
"executable_cache_dir": "/tmp/exe_cache/3.2.1/roberta1", |
|
"execute_encoder_on_cpu_for_generation": false, |
|
"gradient_accumulation_steps": 64, |
|
"inference_device_iterations": 5, |
|
"inference_replication_factor": 1, |
|
"ipus_per_replica": 3, |
|
"layers_per_ipu": [ |
|
1, |
|
1, |
|
1, |
|
1 |
|
], |
|
"matmul_proportion": 0.25, |
|
"optimizer_state_offchip": true, |
|
"optimum_version": "1.6.1", |
|
"output_mode": "final", |
|
"recompute_checkpoint_every_layer": true, |
|
"replicated_tensor_sharding": false, |
|
"replication_factor": 1, |
|
"seed": 42, |
|
"sharded_execution_for_inference": false, |
|
"transformers_version": "4.20.1" |
|
} |
|
|