julien-c HF staff commited on
Commit
d6db600
1 Parent(s): 14702cc

Migrate model card from transformers-repo

Browse files

Read 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/valhalla/bart-large-finetuned-squadv1/README.md

Files changed (1) hide show
  1. README.md +67 -0
README.md ADDED
@@ -0,0 +1,67 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ datasets:
3
+ - squad
4
+ ---
5
+
6
+ # BART-LARGE finetuned on SQuADv1
7
+
8
+ This is bart-large model finetuned on SQuADv1 dataset for question answering task
9
+
10
+ ## Model details
11
+ BART was propsed in the [paper](https://arxiv.org/abs/1910.13461) **BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension**.
12
+ BART is a seq2seq model intended for both NLG and NLU tasks.
13
+
14
+ To use BART for question answering tasks, we feed the complete document into the encoder and decoder, and use the top
15
+ hidden state of the decoder as a representation for each
16
+ word. This representation is used to classify the token. As given in the paper bart-large achives comparable to ROBERTa on SQuAD.
17
+ Another notable thing about BART is that it can handle sequences with upto 1024 tokens.
18
+
19
+ | Param | #Value |
20
+ |---------------------|--------|
21
+ | encoder layers | 12 |
22
+ | decoder layers | 12 |
23
+ | hidden size | 4096 |
24
+ | num attetion heads | 16 |
25
+ | on disk size | 1.63GB |
26
+
27
+
28
+ ## Model training
29
+ This model was trained on google colab v100 GPU.
30
+ You can find the fine-tuning colab here
31
+ [![Open In Colab](https://colab.research.google.com/assets/colab-badge.svg)](https://colab.research.google.com/drive/1I5cK1M_0dLaf5xoewh6swcm5nAInfwHy?usp=sharing).
32
+
33
+
34
+ ## Results
35
+ The results are actually slightly worse than given in the paper.
36
+ In the paper the authors mentioned that bart-large achieves 88.8 EM and 94.6 F1
37
+
38
+ | Metric | #Value |
39
+ |--------|--------|
40
+ | EM | 86.8022|
41
+ | F1 | 92.7342|
42
+
43
+
44
+ ## Model in Action 🚀
45
+ ```python3
46
+ from transformers import BartTokenizer, BartForQuestionAnswering
47
+ import torch
48
+
49
+ tokenizer = BartTokenizer.from_pretrained('valhalla/bart-large-finetuned-squadv1')
50
+ model = BartForQuestionAnswering.from_pretrained('valhalla/bart-large-finetuned-squadv1')
51
+
52
+ question, text = "Who was Jim Henson?", "Jim Henson was a nice puppet"
53
+ encoding = tokenizer(question, text, return_tensors='pt')
54
+ input_ids = encoding['input_ids']
55
+ attention_mask = encoding['attention_mask']
56
+
57
+ start_scores, end_scores = model(input_ids, attention_mask=attention_mask, output_attentions=False)[:2]
58
+
59
+ all_tokens = tokenizer.convert_ids_to_tokens(input_ids[0])
60
+ answer = ' '.join(all_tokens[torch.argmax(start_scores) : torch.argmax(end_scores)+1])
61
+ answer = tokenizer.convert_tokens_to_ids(answer.split())
62
+ answer = tokenizer.decode(answer)
63
+ #answer => 'a nice puppet'
64
+ ```
65
+
66
+ > Created with ❤️ by Suraj Patil [![Github icon](https://cdn0.iconfinder.com/data/icons/octicons/1024/mark-github-32.png)](https://github.com/patil-suraj/)
67
+ [![Twitter icon](https://cdn0.iconfinder.com/data/icons/shift-logotypes/32/Twitter-32.png)](https://twitter.com/psuraj28)