Update README.md
Browse files
README.md
CHANGED
@@ -1,9 +1,61 @@
|
|
1 |
-
|
2 |
-
|
3 |
-
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
|
8 |
-
|
9 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
```
|
2 |
+
biencoder:
|
3 |
+
alias: vdr
|
4 |
+
sequence_length: 512
|
5 |
+
dropout: 0.1
|
6 |
+
shared_encoder: false
|
7 |
+
semiparametric: true
|
8 |
+
device: null
|
9 |
+
encoder_q:
|
10 |
+
model_id: bert-base-uncased
|
11 |
+
max_seq_len: 256
|
12 |
+
pretrained: true
|
13 |
+
norm: false
|
14 |
+
shift_vocab_num: 1000
|
15 |
+
prefix: encoder_q.
|
16 |
+
encoder_p:
|
17 |
+
model_id: bert-base-uncased
|
18 |
+
max_seq_len: 256
|
19 |
+
pretrained: true
|
20 |
+
norm: false
|
21 |
+
shift_vocab_num: 1000
|
22 |
+
prefix: encoder_p.
|
23 |
+
train_datasets:
|
24 |
+
- marco_train_bm25
|
25 |
+
dev_datasets:
|
26 |
+
- marco_dev_30neg
|
27 |
+
datastore: null
|
28 |
+
output_dir: null
|
29 |
+
train_sampling_rates: null
|
30 |
+
batch_size: 64
|
31 |
+
num_train_epochs: 20
|
32 |
+
num_warmup_epochs: 1
|
33 |
+
num_save_epochs: 1
|
34 |
+
num_eval_epochs: 1
|
35 |
+
hard_negatives: 1
|
36 |
+
other_negatives: 0
|
37 |
+
ret_negatives: 0
|
38 |
+
train_insert_title: false
|
39 |
+
valid_insert_title: false
|
40 |
+
adam_eps: 1.0e-08
|
41 |
+
adam_betas: (0.9, 0.999)
|
42 |
+
learning_rate: 2.0e-05
|
43 |
+
max_grad_norm: 2.0
|
44 |
+
log_batch_step: 100
|
45 |
+
train_rolling_loss_step: 100
|
46 |
+
weight_decay: 0.0
|
47 |
+
sym_loss: true
|
48 |
+
do_lower_case: true
|
49 |
+
seed: 12345
|
50 |
+
checkpoint_file_name: vdr
|
51 |
+
save_every_epoch: true
|
52 |
+
model_file: null
|
53 |
+
local_rank: -1
|
54 |
+
local-rank: 2
|
55 |
+
device: null
|
56 |
+
distributed_world_size: null
|
57 |
+
distributed_port: null
|
58 |
+
no_cuda: false
|
59 |
+
n_gpu: null
|
60 |
+
fp16: true
|
61 |
+
```
|