Commit
·
0ec3178
1
Parent(s):
b59f050
Readme
Browse files
Readme.md
ADDED
|
@@ -0,0 +1,82 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
## German NER Albert Model
|
| 2 |
+
|
| 3 |
+
This is a trained Albert model for Token Classification in German ,[Germeval](https://sites.google.com/site/germeval2014ner/) and can be used for Inference.
|
| 4 |
+
|
| 5 |
+
|
| 6 |
+
## Model Specifications
|
| 7 |
+
|
| 8 |
+
- MAX_LENGTH=128
|
| 9 |
+
- MODEL='albert-base-v1'
|
| 10 |
+
- BATCH_SIZE=32
|
| 11 |
+
- NUM_EPOCHS=3
|
| 12 |
+
- SAVE_STEPS=750
|
| 13 |
+
- SEED=1
|
| 14 |
+
- SAVE_STEPS = 100
|
| 15 |
+
- LOGGING_STEPS = 100
|
| 16 |
+
- SEED = 42
|
| 17 |
+
|
| 18 |
+
|
| 19 |
+
|
| 20 |
+
### Usage Specifications
|
| 21 |
+
|
| 22 |
+
This model is trained on Tensorflow version and is compatible with the 'ner' pipeline of huggingface.
|
| 23 |
+
|
| 24 |
+
```python
|
| 25 |
+
from transformers import AutoTokenizer,TFAutoModelForTokenClassification
|
| 26 |
+
from transformers import pipeline
|
| 27 |
+
|
| 28 |
+
model=TFAutoModelForTokenClassification.from_pretrained('abhilash1910/albert-german-ner')
|
| 29 |
+
tokenizer=AutoTokenizer.from_pretrained('abhilash1910/albert-german-ner')
|
| 30 |
+
ner_model = pipeline('ner', model=model, tokenizer=tokenizer)
|
| 31 |
+
seq='Berlin ist die Hauptstadt von Deutschland'
|
| 32 |
+
ner_model(seq)
|
| 33 |
+
```
|
| 34 |
+
|
| 35 |
+
The Tensorflow version of Albert is used for training the model and the output for the above mentioned segment is as follows:
|
| 36 |
+
|
| 37 |
+
```
|
| 38 |
+
[{'entity': 'B-PERderiv',
|
| 39 |
+
'index': 1,
|
| 40 |
+
'score': 0.09580112248659134,
|
| 41 |
+
'word': '▁berlin'},
|
| 42 |
+
{'entity': 'B-ORGpart',
|
| 43 |
+
'index': 2,
|
| 44 |
+
'score': 0.08364498615264893,
|
| 45 |
+
'word': '▁is'},
|
| 46 |
+
{'entity': 'B-LOCderiv',
|
| 47 |
+
'index': 3,
|
| 48 |
+
'score': 0.07593920826911926,
|
| 49 |
+
'word': 't'},
|
| 50 |
+
{'entity': 'B-PERderiv',
|
| 51 |
+
'index': 4,
|
| 52 |
+
'score': 0.09574996680021286,
|
| 53 |
+
'word': '▁die'},
|
| 54 |
+
{'entity': 'B-LOCderiv',
|
| 55 |
+
'index': 5,
|
| 56 |
+
'score': 0.07097965478897095,
|
| 57 |
+
'word': '▁'},
|
| 58 |
+
{'entity': 'B-PERderiv',
|
| 59 |
+
'index': 6,
|
| 60 |
+
'score': 0.07122448086738586,
|
| 61 |
+
'word': 'haupt'},
|
| 62 |
+
{'entity': 'B-PERderiv',
|
| 63 |
+
'index': 7,
|
| 64 |
+
'score': 0.12397754937410355,
|
| 65 |
+
'word': 'stadt'},
|
| 66 |
+
{'entity': 'I-OTHderiv',
|
| 67 |
+
'index': 8,
|
| 68 |
+
'score': 0.0818650871515274,
|
| 69 |
+
'word': '▁von'},
|
| 70 |
+
{'entity': 'I-LOCderiv',
|
| 71 |
+
'index': 9,
|
| 72 |
+
'score': 0.08271490037441254,
|
| 73 |
+
'word': '▁'},
|
| 74 |
+
{'entity': 'B-LOCderiv',
|
| 75 |
+
'index': 10,
|
| 76 |
+
'score': 0.08616268634796143,
|
| 77 |
+
'word': 'deutschland'}]
|
| 78 |
+
```
|
| 79 |
+
|
| 80 |
+
## Resources
|
| 81 |
+
|
| 82 |
+
For all resources , please look into [huggingface](https://huggingface.com).
|