babybabellm-gpt_bert-all

haznitrama/babybabellm-gpt_bert-all

Rehosted from suchirsalhan/babybabellm-multi-all with standardized remote code and auto_map.

• Original model_type preserved.
• Default AutoModel mapping points to GPTBertForCausalLM.
• Added both causal & masked LM wrappers for evaluation.

Example:

from transformers import AutoTokenizer, AutoModel
m='haznitrama/babybabellm-gpt_bert-all'
tok=AutoTokenizer.from_pretrained(m, trust_remote_code=True)
model=AutoModel.from_pretrained(m, trust_remote_code=True)
print(model(**tok('Hello world', return_tensors='pt')).logits.shape)

Generation:

from transformers import AutoModelForCausalLM, AutoTokenizer
m='haznitrama/babybabellm-gpt_bert-all'
tok=AutoTokenizer.from_pretrained(m, trust_remote_code=True)
model=AutoModelForCausalLM.from_pretrained(m, trust_remote_code=True)
print(tok.decode(model.generate(**tok('Hello', return_tensors='pt'), max_new_tokens=20)[0], skip_special_tokens=True))