Changed EOS token to end_of_text to follow LLM2Vec

Not entirely sure if this is legit.. But I am guessing it is because we are finetuning not for a conv style where EOS is typically used, but for embedding purposes closer to the base model training where a sentence ends using EOS end of text. Not sure how much it matters.

tokenizer_config.json

@@ -2052,14 +2052,14 @@
   "bos_token": "<|begin_of_text|>",
   "chat_template": "{% set loop_messages = messages %}{% for message in loop_messages %}{% set content = '<|start_header_id|>' + message['role'] + '<|end_header_id|>\n\n'+ message['content'] | trim + '<|eot_id|>' %}{% if loop.index0 == 0 %}{% set content = bos_token + content %}{% endif %}{{ content }}{% endfor %}{% if add_generation_prompt %}{{ '<|start_header_id|>assistant<|end_header_id|>\n\n' }}{% endif %}",
   "clean_up_tokenization_spaces": true,
-  "eos_token": "<|eot_id|>",
+  "eos_token": "<|end_of_text|>",
   "mask_token": "_",
   "model_input_names": [
     "input_ids",
     "attention_mask"
   ],
   "model_max_length": 1000000000000000019884624838656,
-  "pad_token": "<|eot_id|>",
+  "pad_token": "<|end_of_text|>",
   "padding_side": "left",
   "tokenizer_class": "PreTrainedTokenizerFast"
 }