best model, 7K steps

tokenizer_config.json

@@ -105,13 +105,13 @@
   "clean_up_tokenization_spaces": true,
   "eos_token": "</s>",
   "language_codes": "m2m100",
-  "model_max_length": 1000000000000000019884624838656,
+  "model_max_length": 1024,
   "num_madeup_words": 8,
   "pad_token": "<pad>",
   "sep_token": "</s>",
   "sp_model_kwargs": {},
-  "src_lang": null,
-  "tgt_lang": null,
+  "src_lang": "ru",
+  "tgt_lang": "zu",
   "tokenizer_class": "M2M100Tokenizer",
   "tokenizer_file": null,
   "unk_token": "<unk>"