Tarun Anand
commited on
Commit
•
a6f1db9
1
Parent(s):
dd45730
added
Browse files- README.md +61 -0
- added_tokens.json +1 -0
- config.json +30 -0
- eval.csv +14 -0
- pytorch_model.bin +3 -0
- special_tokens_map.json +1 -0
- spiece.model +3 -0
- tokenizer_config.json +1 -0
- training_args.bin +3 -0
README.md
ADDED
@@ -0,0 +1,61 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
### Model
|
2 |
+
**[`albert-xlarge-v2`](https://huggingface.co/albert-xlarge-v2)** fine-tuned on **[`SQuAD V2`](https://rajpurkar.github.io/SQuAD-explorer/)** using **[`run_squad.py`](https://github.com/huggingface/transformers/blob/master/examples/question-answering/run_squad.py)**
|
3 |
+
|
4 |
+
### Training Parameters
|
5 |
+
Trained on 4 NVIDIA GeForce RTX 2080 Ti 11Gb
|
6 |
+
```bash
|
7 |
+
BASE_MODEL=albert-xlarge-v2
|
8 |
+
python run_squad.py \
|
9 |
+
--version_2_with_negative \
|
10 |
+
--model_type albert \
|
11 |
+
--model_name_or_path $BASE_MODEL \
|
12 |
+
--output_dir $OUTPUT_MODEL \
|
13 |
+
--do_eval \
|
14 |
+
--do_lower_case \
|
15 |
+
--train_file $SQUAD_DIR/train-v2.0.json \
|
16 |
+
--predict_file $SQUAD_DIR/dev-v2.0.json \
|
17 |
+
--per_gpu_train_batch_size 3 \
|
18 |
+
--per_gpu_eval_batch_size 64 \
|
19 |
+
--learning_rate 3e-5 \
|
20 |
+
--num_train_epochs 3.0 \
|
21 |
+
--max_seq_length 384 \
|
22 |
+
--doc_stride 128 \
|
23 |
+
--save_steps 2000 \
|
24 |
+
--threads 24 \
|
25 |
+
--warmup_steps 814 \
|
26 |
+
--gradient_accumulation_steps 4 \
|
27 |
+
--fp16 \
|
28 |
+
--do_train
|
29 |
+
```
|
30 |
+
|
31 |
+
### Evaluation
|
32 |
+
|
33 |
+
Evaluation on the dev set. I did not sweep for best threshold.
|
34 |
+
|
35 |
+
| | val |
|
36 |
+
|-------------------|-------------------|
|
37 |
+
| exact | 84.41842836688285 |
|
38 |
+
| f1 | 87.4628460501696 |
|
39 |
+
| total | 11873.0 |
|
40 |
+
| HasAns_exact | 80.68488529014844 |
|
41 |
+
| HasAns_f1 | 86.78245127423482 |
|
42 |
+
| HasAns_total | 5928.0 |
|
43 |
+
| NoAns_exact | 88.1412952060555 |
|
44 |
+
| NoAns_f1 | 88.1412952060555 |
|
45 |
+
| NoAns_total | 5945.0 |
|
46 |
+
| best_exact | 84.41842836688285 |
|
47 |
+
| best_exact_thresh | 0.0 |
|
48 |
+
| best_f1 | 87.46284605016956 |
|
49 |
+
| best_f1_thresh | 0.0 |
|
50 |
+
|
51 |
+
|
52 |
+
### Usage
|
53 |
+
|
54 |
+
See [huggingface documentation](https://huggingface.co/transformers/model_doc/albert.html#albertforquestionanswering). Training on `SQuAD V2` allows the model to score if a paragraph contains an answer:
|
55 |
+
```python
|
56 |
+
start_scores, end_scores = model(input_ids)
|
57 |
+
span_scores = start_scores.softmax(dim=1).log()[:,:,None] + end_scores.softmax(dim=1).log()[:,None,:]
|
58 |
+
ignore_score = span_scores[:,0,0] #no answer scores
|
59 |
+
|
60 |
+
```
|
61 |
+
|
added_tokens.json
ADDED
@@ -0,0 +1 @@
|
|
|
1 |
+
{}
|
config.json
ADDED
@@ -0,0 +1,30 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"architectures": [
|
3 |
+
"AlbertForQuestionAnswering"
|
4 |
+
],
|
5 |
+
"attention_probs_dropout_prob": 0.1,
|
6 |
+
"bos_token_id": 2,
|
7 |
+
"classifier_dropout_prob": 0.1,
|
8 |
+
"down_scale_factor": 1,
|
9 |
+
"embedding_size": 128,
|
10 |
+
"eos_token_id": 3,
|
11 |
+
"gap_size": 0,
|
12 |
+
"hidden_act": "gelu",
|
13 |
+
"hidden_dropout_prob": 0.1,
|
14 |
+
"hidden_size": 2048,
|
15 |
+
"initializer_range": 0.02,
|
16 |
+
"inner_group_num": 1,
|
17 |
+
"intermediate_size": 8192,
|
18 |
+
"layer_norm_eps": 1e-12,
|
19 |
+
"max_position_embeddings": 512,
|
20 |
+
"model_type": "albert",
|
21 |
+
"net_structure_type": 0,
|
22 |
+
"num_attention_heads": 16,
|
23 |
+
"num_hidden_groups": 1,
|
24 |
+
"num_hidden_layers": 24,
|
25 |
+
"num_memory_blocks": 0,
|
26 |
+
"output_past": true,
|
27 |
+
"pad_token_id": 0,
|
28 |
+
"type_vocab_size": 2,
|
29 |
+
"vocab_size": 30000
|
30 |
+
}
|
eval.csv
ADDED
@@ -0,0 +1,14 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
,val
|
2 |
+
exact,84.41842836688285
|
3 |
+
f1,87.4628460501696
|
4 |
+
total,11873.0
|
5 |
+
HasAns_exact,80.68488529014844
|
6 |
+
HasAns_f1,86.78245127423482
|
7 |
+
HasAns_total,5928.0
|
8 |
+
NoAns_exact,88.1412952060555
|
9 |
+
NoAns_f1,88.1412952060555
|
10 |
+
NoAns_total,5945.0
|
11 |
+
best_exact,84.41842836688285
|
12 |
+
best_exact_thresh,0.0
|
13 |
+
best_f1,87.46284605016956
|
14 |
+
best_f1_thresh,0.0
|
pytorch_model.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:3b759731572f038d3f3d9cb1ef02fac448233dd3d3e4c1b9bfc59e49e87864e5
|
3 |
+
size 234922444
|
special_tokens_map.json
ADDED
@@ -0,0 +1 @@
|
|
|
1 |
+
{"bos_token": "[CLS]", "eos_token": "[SEP]", "unk_token": "<unk>", "sep_token": "[SEP]", "pad_token": "<pad>", "cls_token": "[CLS]", "mask_token": "[MASK]"}
|
spiece.model
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:fefb02b667a6c5c2fe27602d28e5fb3428f66ab89c7d6f388e7c8d44a02d0336
|
3 |
+
size 760289
|
tokenizer_config.json
ADDED
@@ -0,0 +1 @@
|
|
|
1 |
+
{"do_lower_case": true, "max_len": 512, "init_inputs": []}
|
training_args.bin
ADDED
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
1 |
+
version https://git-lfs.github.com/spec/v1
|
2 |
+
oid sha256:c32566fecd5fffd748f6ab2d71404c2a42d714391eaca4453d3e16c2da226284
|
3 |
+
size 1418
|