Conversation
Signed-off-by: adil-a <adil.asif2000@hotmail.com>
Signed-off-by: adil-a <adil.asif2000@hotmail.com>
Signed-off-by: adil-a <adil.asif2000@hotmail.com>
Signed-off-by: adil-a <adil.asif2000@hotmail.com>
Signed-off-by: adil-a <adil.asif2000@hotmail.com>
Signed-off-by: adil-a <adil.asif2000@hotmail.com>
Signed-off-by: adil-a <adil.asif2000@hotmail.com>
Signed-off-by: adil-a <adil.asif2000@hotmail.com>
Signed-off-by: adil-a <adil.asif2000@hotmail.com>
Signed-off-by: adil-a <adil.asif2000@hotmail.com>
Signed-off-by: adil-a <adil.asif2000@hotmail.com>
| ) | ||
| with train_ctx: | ||
| outputs = model(**batch, return_dict=True) | ||
| logits = outputs.logits.view(-1, self.train_n_passages) |
There was a problem hiding this comment.
we basically want to use same value both for train_n_passage both for training and validation? in the yaml file we have train_n_passage defined twice under dataloader and validation_dataloader. what if user put different value for each. according to line above we'd be still using train_n_passages defined under dataloader field in the yaml right?
| # model_type: crossencoder | ||
| # data_dir_list: training_datasets/validation.json | ||
| # data_type: eval | ||
| # train_n_passages: 5 |
There was a problem hiding this comment.
I think we better name this to val_n_passages? bcs in the current case it should be identical to train_n_passages, I wonder what happens if we set it to a different number than the training. I left a related comment about this below.
|
|
||
| seed: 42 | ||
|
|
||
| train_n_passages: 5 |
There was a problem hiding this comment.
why we duplicate this in three different places? isnt that risky? why not define only once.
| seed: 42 | ||
|
|
||
| train_n_passages: 5 | ||
| eval_negative_size: 4 |
There was a problem hiding this comment.
I think we are overcomplicating the yaml. why we need the eval_neg_size?
option 1:
- set train_n_passages only once in the yaml, use it everywhere, both for training and evaluation
option 2: - set train_n_passages, and val_n_passages, and use train_n_passages for training process, the second for val process.
There was a problem hiding this comment.
I will make the change to support the second option
| q_reps = model(query) | ||
| p_reps = model(passage) | ||
|
|
||
| n_passages = self.eval_negative_size + 1 |
There was a problem hiding this comment.
a bit unnecessary, related to my comment above. if we go with option 1 then n_passages is basically same as self.train_n_passages.
Signed-off-by: adil-a <adil.asif2000@hotmail.com>
2 participants
Uh oh!
There was an error while loading. Please reload this page.