upload

README.md

@@ -0,0 +1,5 @@
+## Roberta-Base
+
+This repo trains [roberta-base](https://huggingface.co/roberta-base) from scratch on the [Norwegian training subset of Oscar](https://oscar-corpus.com/) containing roughly 4.7 GB of data according to [this](https://github.com/huggingface/transformers/tree/master/examples/flax/language-modeling) example.
+
+Training is done on a TPUv3-8 in Flax. More statistics on the training run can be found under [tf.hub](https://tensorboard.dev/experiment/GdYmdak2TWeVz0DDRYOrrg).

config.json

@@ -0,0 +1,26 @@
+{
+  "_name_or_path": "./",
+  "architectures": [
+    "RobertaForMaskedLM"
+  ],
+  "attention_probs_dropout_prob": 0.1,
+  "bos_token_id": 0,
+  "eos_token_id": 2,
+  "gradient_checkpointing": false,
+  "hidden_act": "gelu",
+  "hidden_dropout_prob": 0.1,
+  "hidden_size": 768,
+  "initializer_range": 0.02,
+  "intermediate_size": 3072,
+  "layer_norm_eps": 1e-05,
+  "max_position_embeddings": 514,
+  "model_type": "roberta",
+  "num_attention_heads": 12,
+  "num_hidden_layers": 12,
+  "pad_token_id": 1,
+  "position_embedding_type": "absolute",
+  "transformers_version": "4.7.0.dev0",
+  "type_vocab_size": 1,
+  "use_cache": true,
+  "vocab_size": 50265
+}

flax_model.msgpack

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:356a3d955697364c8a996cae104843bfc0a8acc398d2b303172cf1f3bfd0fbf2
+size 498796983

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9c65fdff45715ebc7c661001206e9aa32b6707efa1940fec592e7a57285fc22
+size 498875931