nielsr HF staff commited on
Commit
76cea08
1 Parent(s): 2605f73

Add model card

Browse files
Files changed (1) hide show
  1. README.md +34 -0
README.md ADDED
@@ -0,0 +1,34 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ This model corresponds to **tapas_masklm_large_reset** of the [original repository](https://github.com/google-research/tapas).
2
+
3
+ Here's how you can use it:
4
+
5
+ ```python
6
+ from transformers import TapasTokenizer, TapasForMaskedLM
7
+ import pandas as pd
8
+ import torch
9
+
10
+ tokenizer = TapasTokenizer.from_pretrained("google/tapas-large-masklm")
11
+ model = TapasForMaskedLM.from_pretrained("google/tapas-large-masklm")
12
+
13
+ data = {'Actors': ["Brad Pitt", "Leonardo Di Caprio", "George Clooney"],
14
+ 'Age': ["56", "45", "59"],
15
+ 'Number of movies': ["87", "53", "69"]
16
+ }
17
+ table = pd.DataFrame.from_dict(data)
18
+ query = "How many movies has Leonardo [MASK] Caprio played in?"
19
+
20
+ # prepare inputs
21
+ inputs = tokenizer(table=table, queries=query, padding="max_length", return_tensors="pt")
22
+
23
+ # forward pass
24
+ outputs = model(**inputs)
25
+
26
+ # return top 5 values and predictions
27
+ masked_index = torch.nonzero(inputs.input_ids.squeeze() == tokenizer.mask_token_id, as_tuple=False)
28
+ logits = outputs.logits[0, masked_index.item(), :]
29
+ probs = logits.softmax(dim=0)
30
+ values, predictions = probs.topk(5)
31
+
32
+ for value, pred in zip(values, predictions):
33
+ print(f"{tokenizer.decode([pred])} with confidence {value}")
34
+ ```