wyu97
commited on
Commit
•
2df49c8
1
Parent(s):
fb3a0e1
update
Browse files- README.md +24 -0
- config.json +51 -0
- tokenizer.json +0 -0
README.md
CHANGED
@@ -1,3 +1,27 @@
|
|
1 |
---
|
2 |
license: cc-by-4.0
|
3 |
---
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
---
|
2 |
license: cc-by-4.0
|
3 |
---
|
4 |
+
|
5 |
+
# GenRead (MergeDPR): FiD model trained on TQA
|
6 |
+
|
7 |
+
-- This is the model checkpoint of GenRead [2], based on the T5-3B and trained on the TriviaQA [1].
|
8 |
+
|
9 |
+
-- Hyperparameters: 8 x 80GB A100 GPUs; batch size 16; AdamW; LR 5e-5; best dev at 9000 steps
|
10 |
+
|
11 |
+
References:
|
12 |
+
|
13 |
+
[1] TriviaQA: A Large Scale Dataset for Reading Comprehension and Question Answering. ACL 2017
|
14 |
+
|
15 |
+
[2] Generate rather than Retrieve: Large Language Models are Strong Context Generators. arXiv 2022
|
16 |
+
|
17 |
+
## Model performance
|
18 |
+
|
19 |
+
We evaluate it on the TriviaQA dataset, the EM score is 74.41.
|
20 |
+
|
21 |
+
|
22 |
+
<a href="https://huggingface.co/exbert/?model=bert-base-uncased">
|
23 |
+
<img width="300px" src="https://cdn-media.huggingface.co/exbert/button.png">
|
24 |
+
</a>
|
25 |
+
---
|
26 |
+
license: cc-by-4.0
|
27 |
+
---
|
config.json
ADDED
@@ -0,0 +1,51 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
{
|
2 |
+
"architectures": [
|
3 |
+
"FiDT5"
|
4 |
+
],
|
5 |
+
"d_ff": 16384,
|
6 |
+
"d_kv": 128,
|
7 |
+
"d_model": 1024,
|
8 |
+
"decoder_start_token_id": 0,
|
9 |
+
"dropout_rate": 0.1,
|
10 |
+
"eos_token_id": 1,
|
11 |
+
"initializer_factor": 1.0,
|
12 |
+
"is_encoder_decoder": true,
|
13 |
+
"layer_norm_epsilon": 1e-06,
|
14 |
+
"model_type": "t5",
|
15 |
+
"n_positions": 512,
|
16 |
+
"num_heads": 32,
|
17 |
+
"num_layers": 24,
|
18 |
+
"output_past": true,
|
19 |
+
"pad_token_id": 0,
|
20 |
+
"relative_attention_num_buckets": 32,
|
21 |
+
"task_specific_params": {
|
22 |
+
"summarization": {
|
23 |
+
"early_stopping": true,
|
24 |
+
"length_penalty": 2.0,
|
25 |
+
"max_length": 200,
|
26 |
+
"min_length": 30,
|
27 |
+
"no_repeat_ngram_size": 3,
|
28 |
+
"num_beams": 4,
|
29 |
+
"prefix": "summarize: "
|
30 |
+
},
|
31 |
+
"translation_en_to_de": {
|
32 |
+
"early_stopping": true,
|
33 |
+
"max_length": 300,
|
34 |
+
"num_beams": 4,
|
35 |
+
"prefix": "translate English to German: "
|
36 |
+
},
|
37 |
+
"translation_en_to_fr": {
|
38 |
+
"early_stopping": true,
|
39 |
+
"max_length": 300,
|
40 |
+
"num_beams": 4,
|
41 |
+
"prefix": "translate English to French: "
|
42 |
+
},
|
43 |
+
"translation_en_to_ro": {
|
44 |
+
"early_stopping": true,
|
45 |
+
"max_length": 300,
|
46 |
+
"num_beams": 4,
|
47 |
+
"prefix": "translate English to Romanian: "
|
48 |
+
}
|
49 |
+
},
|
50 |
+
"vocab_size": 32128
|
51 |
+
}
|
tokenizer.json
ADDED
The diff for this file is too large to render.
See raw diff
|
|