Added model weights, vocabs, readme.

README.md

@@ -0,0 +1,78 @@
+---
+language:
+- en
+- el
+tags:
+- translation
+widget:
+- text: "Not all those who wander are lost."
+license: apache-2.0
+metrics:
+- bleu
+---
+
+## English to Greek NMT (lower-case output)
+## By the Hellenic Army Academy (SSE) and the Technical University of Crete (TUC)
+
+* source languages: en
+* target languages: el
+* licence: apache-2.0
+* dataset: Opus, CCmatrix
+* model: transformer(fairseq)
+* pre-processing: tokenization + lowercasing + BPE segmentation
+* metrics: bleu, chrf
+* output: lowercase only, for mixed-cased model use this: https://huggingface.co/lighteternal/SSE-TUC-mt-en-el-cased
+
+### Model description
+
+Trained using the Fairseq framework, transformer_iwslt_de_en architecture.\
+BPE segmentation (10k codes).\
+Lower-case model. 
+
+### How to use
+
+```
+from transformers import FSMTTokenizer, FSMTForConditionalGeneration
+
+mname = " <your_downloaded_model_folderpath_here> "
+
+tokenizer = FSMTTokenizer.from_pretrained(mname)
+model = FSMTForConditionalGeneration.from_pretrained(mname)
+
+text = "Not all those who wander are lost."
+
+encoded = tokenizer.encode(text, return_tensors='pt')
+
+outputs = model.generate(encoded, num_beams=5, num_return_sequences=5, early_stopping=True)
+for i, output in enumerate(outputs):
+    i += 1
+    print(f"{i}: {output.tolist()}")
+    
+    decoded = tokenizer.decode(output, skip_special_tokens=True)
+    print(f"{i}: {decoded}")
+```
+
+
+## Training data
+
+Consolidated corpus from Opus and CC-Matrix (~6.6GB in total)
+
+
+## Eval results
+
+
+Results on Tatoeba testset (EN-EL): 
+
+| BLEU | chrF  |
+| ------ | ------ |
+| 77.3 |  0.739 |
+
+
+Results on XNLI parallel (EN-EL): 
+
+| BLEU | chrF  |
+| ------ | ------ |
+| 66.1 |  0.606 |
+
+### BibTeX entry and citation info
+TODO

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": 9936,
+  "tgt_vocab_size": 12896,
+  "langs": [
+    "en",
+    "el"
+  ],
+  "encoder_attention_heads": 4,
+  "encoder_ffn_dim": 1024,
+  "encoder_layerdrop": 0,
+  "encoder_layers": 6,
+  "decoder_attention_heads": 4,
+  "decoder_ffn_dim": 1024,
+  "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 sha256:04df29cda5952138f6dd7e03361f0e3cb1219625791b7c61f5b8475de20f2311
+size 172976478

tokenizer_config.json

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