Getting likelihoods for a given target
#16
by
adymaharana
- opened
Is there a way to get likelihoods from the model for a pre-defined answer? Something like this:
inputs = processor(text=prompt, images=image, return_tensors="pt")
targets = processor(text=answer, return_tensors="pt")
output = model(
pixel_values=inputs["pixel_values"],
input_ids=inputs["input_ids"][:, :-1],
attention_mask=inputs["attention_mask"][:, :-1],
img_features=None,
img_attn_mask=inputs["img_attn_mask"][:, :-1],
use_cache=False,
labels=targets
)
nll_loss = output[0]
Hi, the implementation (should) have the labels argument already. But I didn't try it myself. I would suggest to wait for the official port, see https://github.com/huggingface/transformers/pull/24709
ydshieh
changed discussion status to
closed