|
--- |
|
dataset_info: |
|
features: |
|
- name: prompt_text |
|
dtype: string |
|
- name: prompt_ids |
|
sequence: int64 |
|
- name: decode_ids |
|
sequence: int64 |
|
- name: prompt_pattern |
|
sequence: |
|
sequence: int64 |
|
- name: decode_pattern |
|
sequence: |
|
sequence: int64 |
|
- name: predictor_pattern |
|
sequence: |
|
sequence: |
|
sequence: int64 |
|
splits: |
|
- name: train |
|
num_bytes: 297949158 |
|
num_examples: 30000 |
|
download_size: 30294841 |
|
dataset_size: 297949158 |
|
configs: |
|
- config_name: default |
|
data_files: |
|
- split: train |
|
path: data/train-* |
|
--- |
|
|
|
|
|
`decode_pattern.shape`: (num_layers, num_decoding_steps) |
|
`predictor_pattern.shape`: (num_decoding_steps, num_layers, top3_indices) |
|
|
|
We need to permute `predictor_pattern` via: |
|
|
|
```python |
|
predictor_pattern = predictor_pattern.permute(1,0,2) # (num_layers, num_decoding_steps, top3_indices) |
|
``` |
