Replies: 2 comments 1 reply
-
|
Yes @DSDay all training support has been removed from haystack 2.x and will not be introduced. These train methods existed in 1.x for convenience; today one can use dedicated training (fine-tuning) frameworks; it didn't make sense to have it in Haystack 2.x |
Beta Was this translation helpful? Give feedback.
-
|
Thank you, Vladimir! Which tuning framework would you suggest? I switched from Huggingface's trainer-based script to Haystack's because I couldn't get my data handled properly, resulting in flawed model learning/tuning. The squad2-based JSON representation supported by Haystack made it all work for me. Thanks for any pointers you can suggest. |
Beta Was this translation helpful? Give feedback.
-
|
I would have said to use HF trainer myself, what was the problem with it? Perhaps we can look at it together? |
Beta Was this translation helpful? Give feedback.
-
Hello, It seems as if model tuning isn't supported in the current version of Haystack 2.x (haystack-ai). For example, there is no .train method provided for the QuestionAnswering Reader any more. Am I not looking in the right place, or is this still in development or has Deepset decided not to maintain this capability? Thanks. David
Beta Was this translation helpful? Give feedback.
All reactions