first version of the model

config.json

@@ -0,0 +1,87 @@
+{
+  "_name_or_path": "/localdata/jamesbr/dev/pretrained_checkpoints/pytorch_bert_large_phase1/",
+  "architectures": [
+    "PoptorchPipelinedBertForPretraining"
+  ],
+  "async_dataloader": true,
+  "attention_probs_dropout_prob": 0.0,
+  "auto_loss_scaling": false,
+  "batch_size": 2,
+  "batches_per_step": 1,
+  "checkpoint_output_dir": "/localdata/jamesbr/dev/pretrained_checkpoints/pytorch_bert_large_phase2",
+  "checkpoint_steps": null,
+  "compile_only": false,
+  "config": null,
+  "custom_ops": true,
+  "dataloader_workers": 64,
+  "dataset": "pretraining",
+  "disable_progress_bar": true,
+  "embedding_serialization_factor": 2,
+  "enable_half_first_order_momentum": false,
+  "enable_half_partials": true,
+  "executable_cache_dir": "",
+  "file_buffer_size": 10,
+  "global_batch_size": 16384,
+  "gradient_accumulation": 2048,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 1024,
+  "initializer_range": 0.02,
+  "input_files": [
+    "data/wikipedia/384/*.tfrecord"
+  ],
+  "intermediate_size": 4096,
+  "ipus_per_replica": 4,
+  "layer_norm_eps": 0.001,
+  "layers_per_ipu": [
+    3,
+    7,
+    7,
+    7
+  ],
+  "learning_rate": 0.002828,
+  "loss_scaling": 8192.0,
+  "lr_schedule": "linear",
+  "lr_warmup": 0.128,
+  "mask_tokens": 56,
+  "matmul_proportion": [
+    0.15,
+    0.25,
+    0.25,
+    0.25
+  ],
+  "max_position_embeddings": 512,
+  "model_type": "bert",
+  "num_attention_heads": 16,
+  "num_epochs": null,
+  "num_hidden_layers": 24,
+  "optimizer": "LAMB",
+  "optimizer_state_offchip": true,
+  "pad_token_id": 0,
+  "position_embedding_type": "absolute",
+  "pred_head_transform": true,
+  "pretrained_checkpoint": "/localdata/jamesbr/dev/pretrained_checkpoints/pytorch_bert_large_phase1/",
+  "profile": false,
+  "profile_dir": "profile",
+  "random_seed": 42,
+  "recompute_checkpoint_every_layer": true,
+  "replicated_tensor_sharding": true,
+  "replication_factor": 4,
+  "restore_steps_and_optimizer": false,
+  "samples_per_step": 16384,
+  "sdk_version": "poplar_sdk-ubuntu_18_04-2.3.0-EA.1+716-757737e247",
+  "sequence_length": 384,
+  "squad_do_training": true,
+  "squad_do_validation": true,
+  "synthetic_data": false,
+  "training_steps": 2137,
+  "transformers_version": "4.7.0",
+  "type_vocab_size": 2,
+  "use_cache": true,
+  "use_popdist": false,
+  "vocab_size": 30522,
+  "wandb": true,
+  "wandb_param_steps": null,
+  "weight_decay": 0.01
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:0aa7af4e8c9c66b940751f91be95590f3dde4aba5160d65d5ce952dff2835a63
+size 672610403

training_state.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8e170da99eae1aeb46476dfb799e440557c0a07e40043abe92e1e339048452f2
+size 5167