Hi,

Thanks for your great work.
When I use "T5_base_finetuned_common_gen" to generate sentences with "dog" as inputs, the model outputs are always same.
Can I get diverse sentences with the same input?

@mrm8488
Buenos días, me preguntaba si podrías darme alguna explicación sobre la generación de preguntas que utilizas en el proyecto T5-base fine-tuned on SQuAD for Question Generation de huggingface. He visto que te basas en un script de QA pero me preguntaba como modificarlo correctamente. Si pudieras darme alguna clave más o alguna parte del codigo.

Otra opción es el entrenamiento para la realización de QG en español, no se si esta entre tus próximos proyectos o no pero sería interesante ver como se haría.
Muchas gracias.

Hi Manuel,
Thanks for this amazing repo, it's a treasure trove.
In the flan_t5_text_to_sql finetune, what parameter can I optimize to increase the output size.
Also is it better to give the entire schema in the prompt or just the question and expected SQL query?

@mrm8488 hi thanks for opensourcing and sharing the colab to run the yolact++ but when i try to do the same by myself by creating a repo and uploading all the files and changing the git clone from author to my repo and run it i get error in evaluation as mentioned below "File "/content/yolact_adas/backbone.py", line 11, in DCN
raise Exception('DCN could not be imported. If you want to use YOLACT++ models, compile DCN. Check the README for instructions.')
Exception: DCN could not be imported. If you want to use YOLACT++ models, compile DCN. Check the README for instructions." is it not possbile to do it ?? or should we perform only authors repo

I want to implement this model to convert Natural Language to SQL but on my own database. When I give a question outside of the database, it should give me an error and prompt me to ask a question on my database. I tried out a code but it is not getting connected with my database. I am using Chinook.db. Can anyone help in the same. If you have implemented the do let me know how did you do it.
I am using Hugging face(without API) and Langchain agent framework.

@mrm8488
Hey i show you have done a great work in huggingface community. I have some question on your https://huggingface.co/mrm8488/bert-multi-cased-finetuned-xquadv1 this model.
You have mention about augmentation technique.
1.so can you please tell me which neural machine translation model you used?
2.how you use scriping?
3.which are other methods that we can apply for creating more data?
4.which dataset you use for machine translation ?

I love the interface, but am curious to how I can train my own dataset on this. I tried copying a pre-trained checkpoint, but I've been continually met with errors.

Hi Manuel, Thank you for sharing the pretrained T5 model. I would like to finetune your model on a few examples of my own. Could you please share the code you used to preprocess wikisql data, and loaded, trained, evaluated the T5 model?

I am running the colab notebook shared here:

https://github.com/mrm8488/shared_colab_notebooks/blob/bf6d578042bbb393e8cfcb336e2909c9f460b91c/T5_wikiSQL_multitask_with_HF_transformers.ipynb

When I get to trainer.evaluate() I get the following error message:

RuntimeError: output with shape [16, 8, 1, 1] doesn't match the broadcast shape [16, 8, 1, 64]

I've attempted to search for solutions, but I can't find many instances where this type of error comes up with NLP training. It seems to most often occur with image raster data.

I would greatly appreciate any insight that you may have. Thanks!

Eric