Hi, I am using Transformers from Git on Linux. I am passing return_attentions=True, return_dict_in_generate=True, and return_attention_mask=True in my GenerationConfig but the keys in the output from Generate are only ‘sequences’, ‘attentions’, and ‘past_key_values’. The attention_mask that I am supposed to be passing back to the model, is not being returned by generate().
Related topics
| Topic | Replies | Views | Activity | |
|---|---|---|---|---|
| Customizing GenerationMixin to output attentions | 4 | 1769 | September 10, 2020 | |
| Problem with returning decoder cross attentions through generate function | 0 | 24 | October 25, 2024 | |
| [Announcement] GenerationOutputs: Scores, Attentions and Hidden States now available as outputs to generate | 1 | 4566 | January 13, 2021 | |
| Attentions not returned from transformers ViT model when using output_attentions=True | 4 | 760 | July 10, 2024 | |
| Understanding attention output from generate method in GPT model | 0 | 575 | November 8, 2023 |