Migrate model card from transformers-repo
Browse filesRead announcement at https://discuss.huggingface.co/t/announcement-all-model-cards-will-be-migrated-to-hf-co-model-repos/2755
Original file history: https://github.com/huggingface/transformers/commits/master/model_cards/wptoux/albert-chinese-large-qa/README.md
README.md
ADDED
@@ -0,0 +1,34 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
# albert-chinese-large-qa
|
2 |
+
Albert large QA model pretrained from baidu webqa and baidu dureader datasets.
|
3 |
+
|
4 |
+
## Data source
|
5 |
+
+ baidu webqa 1.0
|
6 |
+
+ baidu dureader
|
7 |
+
|
8 |
+
## Traing Method
|
9 |
+
We combined the two datasets together and created a new dataset in squad format, including 705139 samples for training and 69638 samples for validation.
|
10 |
+
We finetune the model based on the albert chinese large model.
|
11 |
+
|
12 |
+
## Hyperparams
|
13 |
+
+ learning_rate 1e-5
|
14 |
+
+ max_seq_length 512
|
15 |
+
+ max_query_length 50
|
16 |
+
+ max_answer_length 300
|
17 |
+
+ doc_stride 256
|
18 |
+
+ num_train_epochs 2
|
19 |
+
+ warmup_steps 1000
|
20 |
+
+ per_gpu_train_batch_size 8
|
21 |
+
+ gradient_accumulation_steps 3
|
22 |
+
+ n_gpu 2 (Nvidia Tesla P100)
|
23 |
+
|
24 |
+
## Usage
|
25 |
+
```
|
26 |
+
from transformers import AutoModelForQuestionAnswering, BertTokenizer
|
27 |
+
|
28 |
+
model = AutoModelForQuestionAnswering.from_pretrained('wptoux/albert-chinese-large-qa')
|
29 |
+
tokenizer = BertTokenizer.from_pretrained('wptoux/albert-chinese-large-qa')
|
30 |
+
```
|
31 |
+
***Important: use BertTokenizer***
|
32 |
+
|
33 |
+
## MoreInfo
|
34 |
+
Please visit https://github.com/wptoux/albert-chinese-large-webqa for details.
|