qhduan

/

aquilachat-7b

like 17

I updated the readme, check the model card please

I updated the readme, check the model card please

According to your code, the model can't stop correctly and need manually truncate the result. Model generation will waste a lot of time. So that means you didn't use eos_token during model trainning? Why there are 2 eos_token_ids?

I updated the readme, check the model card please

According to your code, the model can't stop correctly and need manually truncate the result. Model generation will waste a lot of time. So that means you didn't use eos_token during model trainning? Why there are 2 eos_token_ids?

According to https://github.com/FlagAI-Open/FlagAI/issues/389, I modify following in config.json:
"bos_token_id": 100006,
"eos_token_id": 100007,
"pad_token_id": 0,
"unk_token_id": 0,

Then model can stop correctly without additional truncation of results.

But eos_token_id in config.json is still 2. And tokenizer should be updated too.

But eos_token_id in config.json is still 2. And tokenizer should be updated too.

@JaheimLee , sorry I am not owner, just met same issue, and try to find a fix.
@qhduan , could you please verify comments above and update configuration in repo?

Thank you @sammysun0711 I'll update the config