Echo-DSRN-v0.1.3-Embed-Exp / tokenizer_config.json
mrs83's picture
config: restore left padding — published MTEB benchmarks were measured with left padding; card parity is the priority (right-padded training lands in the next model version)
8fe8084 verified
Raw
History Blame Contribute Delete
721 Bytes
{
"backend": "tokenizers",
"bos_token": "<s>",
"clean_up_tokenization_spaces": false,
"eos_token": "<|endoftext|>",
"extra_special_tokens": [
"<tool_call>",
"</tool_call>",
"<tool_response>",
"</tool_response>",
"<tools>",
"</tools>"
],
"fix_mistral_regex": true,
"is_local": true,
"legacy": false,
"max_length": 2048,
"model_max_length": 2048,
"pad_to_multiple_of": null,
"pad_token": "<|endoftext|>",
"pad_token_type_id": 0,
"padding_side": "left",
"sp_model_kwargs": {},
"stride": 0,
"tokenizer_class": "TokenizersBackend",
"truncation_side": "right",
"truncation_strategy": "longest_first",
"unk_token": "<unk>",
"use_default_system_prompt": false
}