A MBARTHEZ MODEL TRAINED FOR QUESTION GENERATION

Training

The model has been trained on different french and english corpus (FQuAD, PIAF and SQuAD)

Generate

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

# Getting the data
access_token = "hf_......"
tokenizer = AutoTokenizer.from_pretrained("ThomasGerald/MBARTHEZ-QG", use_auth_token=access_token)
model = AutoModelForSeq2SeqLM.from_pretrained("ThomasGerald/MBARTHEZ-QG", use_auth_token=access_token)

# text input exemple notice we use the token <hl> to delimite the support of the question
text = ("La recherche moderne considère généralement que la langue grecque n'est pas née en Grèce," +
   "mais elle n'est pas arrivée à un consensus quant à la date d'arrivée des groupes parlant un "+
   "« proto-grec », qui s'est produite durant des phases préhistoriques pour lesquelles il n'y a"+
   "pas de texte indiquant quelles langues étaient parlées. Les premiers textes écrits en grec <hl>sont"+
   "les tablettes en linéaire B de l'époque mycénienne<hl>, au XIVe siècle av. J.-C., ce qui indique que"+
   "des personnes parlant un dialecte grec sont présentes en Grèce au plus tard durant cette période."+
   " La linguistique n'est pas en mesure de trancher, pas plus que l'archéologie.")

tokenized_text = tokenizer([text], return_tensors="pt")

# Output conditionnaly to the language (here two tokens possible '[fr_XX]' and '[en_XX]')
output_ids = model.generate(**tokenized_text, forced_bos_token_id=tokenizer.convert_tokens_to_ids(['[fr_XX]']))

# Decoding
output = tokenizer.batch_decode(output_ids, skip_special_tokens=False)

# output:
'''['</s>[fr_XX] Quels sont les premiers textes écrits en grec?</s>']'''

We can also generate question in english from french context by specifying the begining of sentence token ('[en_XX]'). Considering the previous code prepending the following one we can generate english questions executing :

output_ids = model.generate(**tokenized_text, forced_bos_token_id=tokenizer.convert_tokens_to_ids(['[en_XX]']))
output = tokenizer.batch_decode(output_idsskip_special_tokens=False)

# output:
'''['</s>[en_XX] What are the first texts written in grec?</s>']'''

Of course you can also generate questions from english text :

# text input exemple notice we use the token <hl> to delimite the support of the question
text = ("By 371 BC, Thebes was in the ascendancy, defeating Sparta at" +
        "<hl>the Battle of Leuctra<hl>, killing the Spartan king Cleombrotus I" +
        ", and invading Laconia. Further Theban successes against Sparta" +
        "in 369 led to Messenia gaining independence; Sparta never recovered" +
        "from the loss of Messenia's fertile land and the helot workforce it" +
        "provided.[50] The rising power of Thebes led Sparta and Athens to join" +
        "forces; in 362 they were defeated by Thebes at the Battle of Mantinea." +
        " In the aftermath of Mantinea, none of the major Greek states were able" +
        "to dominate. Though Thebes had won the battle, their general Epaminondas" +
        "was killed, and they spent the following decades embroiled in wars with"+
        "their neighbours; Athens, meanwhile, saw its second naval alliance," + 
        " formed in 377, collapse in the mid-350s.")

tokenized_text = tokenizer([text], return_tensors="pt")

# French question
output_ids = model.generate(**tokenized_text, forced_bos_token_id=tokenizer.convert_tokens_to_ids(['[fr_XX]']))

# Decoding
output = tokenizer.batch_decode(output_ids, skip_special_tokens=False)

# Notice it does not translate "Sparta" which is "Sparte" in french 
'''['</s>[fr_XX] À quelle bataille Sparta a-t-il été vaincu par Thebes?</s>']'''

# English question
output_ids = model.generate(**tokenized_text, forced_bos_token_id=tokenizer.convert_tokens_to_ids(['[en_XX]']))

# Decoding
output = tokenizer.batch_decode(output_ids, skip_special_tokens=False)

'''['</s>[en_XX] At what battle did Thebes defeat Sparta?</s>']'''