Upload README.md
#2
by
leelearn
- opened
README.md
CHANGED
|
@@ -4,9 +4,6 @@ language:
|
|
| 4 |
- zh
|
| 5 |
tags:
|
| 6 |
- GENIUS
|
| 7 |
-
- conditional text generation
|
| 8 |
-
- sketch-based text generation
|
| 9 |
-
- data augmentation
|
| 10 |
|
| 11 |
license: apache-2.0
|
| 12 |
datasets:
|
|
@@ -24,7 +21,7 @@ widget:
|
|
| 24 |
|
| 25 |
inference:
|
| 26 |
parameters:
|
| 27 |
-
max_length:
|
| 28 |
num_beams: 3
|
| 29 |
do_sample: True
|
| 30 |
---
|
|
@@ -47,7 +44,7 @@ inference:
|
|
| 47 |
```python
|
| 48 |
# genius-chinese
|
| 49 |
from transformers import BertTokenizer, BartForConditionalGeneration, Text2TextGenerationPipeline
|
| 50 |
-
checkpoint = '
|
| 51 |
tokenizer = BertTokenizer.from_pretrained(checkpoint)
|
| 52 |
genius_model = BartForConditionalGeneration.from_pretrained(checkpoint)
|
| 53 |
genius_generator = Text2TextGenerationPipeline(genius_model, tokenizer, device=0)
|
|
@@ -120,7 +117,6 @@ GENIUS-chinese output:
|
|
| 120 |
|
| 121 |
可以看出，BART只能填补简单的一些词，无法对这些片段进行很连贯的连接，而GENIUS则可以扩写成连贯的句子甚至段落。
|
| 122 |
|
| 123 |
-
|
| 124 |
---
|
| 125 |
|
| 126 |
If you find our paper/code/demo useful, please cite our paper:
|
|
|
|
| 4 |
- zh
|
| 5 |
tags:
|
| 6 |
- GENIUS
|
|
|
|
|
|
|
|
|
|
| 7 |
|
| 8 |
license: apache-2.0
|
| 9 |
datasets:
|
|
|
|
| 21 |
|
| 22 |
inference:
|
| 23 |
parameters:
|
| 24 |
+
max_length: 1000
|
| 25 |
num_beams: 3
|
| 26 |
do_sample: True
|
| 27 |
---
|
|
|
|
| 44 |
```python
|
| 45 |
# genius-chinese
|
| 46 |
from transformers import BertTokenizer, BartForConditionalGeneration, Text2TextGenerationPipeline
|
| 47 |
+
checkpoint = 'leelearn/genius-yiyan-prompt-generator'
|
| 48 |
tokenizer = BertTokenizer.from_pretrained(checkpoint)
|
| 49 |
genius_model = BartForConditionalGeneration.from_pretrained(checkpoint)
|
| 50 |
genius_generator = Text2TextGenerationPipeline(genius_model, tokenizer, device=0)
|
|
|
|
| 117 |
|
| 118 |
可以看出，BART只能填补简单的一些词，无法对这些片段进行很连贯的连接，而GENIUS则可以扩写成连贯的句子甚至段落。
|
| 119 |
|
|
|
|
| 120 |
---
|
| 121 |
|
| 122 |
If you find our paper/code/demo useful, please cite our paper:
|