Initial commit

README.md

@@ -1,3 +1,24 @@
 ---
+language:
+- en
+- et
+- multilingual
 license: cc-by-sa-4.0
+tags:
+- translation
+- wmt20
+datasets:
+- wmt/europarl
+widget:
+- text: "Jupiter on Päikesest kauguselt viies planeet ja Päikesesüsteemi kõige suurem planeet."
+- text: "Plejaadid on Sõnni tähtkujus asuv hajusparv, mille Messier' kataloogi tähiseks on M45."
+- text: "Palju on vaieldud Vikipeedia usaldatavuse ja täpsuse üle. Kritiseeritud on selle avatust vandaalidele, ebaühtlast kvaliteeti ja vasturääkivust, mitteneutraalsust ja konsensuse või populaarsuse eelistamist kvalifitseeritusele."
 ---
+
+# Fairseq Et-En NMT WMT20 MLQE
+
+This repository contains the Estonian-English model trained with the [fairseq toolkit](https://github.com/pytorch/fairseq) that was used to produce translations used in the WMT20 shared task on quality estimation (QE) on the [MLQE dataset](https://github.com/facebookresearch/mlqe).
+
+The checkpoint was converted from the original fairseq checkpoint available [here](https://github.com/facebookresearch/mlqe/tree/master/nmt_models) using the `convert_fsmt_original_pytorch_checkpoint_to_pytorch.py` script from the 🤗 Transformers library (v4.26.0).
+
+Please refer to the repositories linked above for additional information on usage, parameters and training data

config.json

@@ -0,0 +1,37 @@
+{
+  "architectures": [
+    "FSMTForConditionalGeneration"
+  ],
+  "model_type": "fsmt",
+  "activation_dropout": 0.0,
+  "activation_function": "relu",
+  "attention_dropout": 0.0,
+  "d_model": 512,
+  "dropout": 0.3,
+  "init_std": 0.02,
+  "max_position_embeddings": 1024,
+  "num_hidden_layers": 6,
+  "src_vocab_size": 32760,
+  "tgt_vocab_size": 18592,
+  "langs": [
+    "et",
+    "en"
+  ],
+  "encoder_attention_heads": 8,
+  "encoder_ffn_dim": 2048,
+  "encoder_layerdrop": 0,
+  "encoder_layers": 6,
+  "decoder_attention_heads": 8,
+  "decoder_ffn_dim": 2048,
+  "decoder_layerdrop": 0,
+  "decoder_layers": 6,
+  "bos_token_id": 0,
+  "pad_token_id": 1,
+  "eos_token_id": 2,
+  "is_encoder_decoder": true,
+  "scale_embedding": true,
+  "tie_word_embeddings": false,
+  "num_beams": 5,
+  "early_stopping": false,
+  "length_penalty": 1.0
+}

pytorch_model.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1c3dff33aafb02e93e5f10603795ea1b6ff1531b6bce69e54d21ffc0f475dea9
+size 357428997

tokenizer_config.json

@@ -0,0 +1,8 @@
+{
+  "langs": [
+    "et",
+    "en"
+  ],
+  "model_max_length": 1024,
+  "do_lower_case": false
+}