Hugging Face's logo

jzhoubu
/
vdr-msmarco

Transformers
PyTorch
Inference Endpoints
Model card Files Community
vdr-msmarco
File size: 2,179 Bytes 
57a776c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
 
biencoder:
  alias: vdr
  sequence_length: 512
  dropout: 0.1
  shared_encoder: false
  semiparametric: true
  device: null
  encoder_q:
    model_id: bert-base-uncased
    max_seq_len: 256
    pretrained: true
    norm: false
    shift_vocab_num: 1000
    prefix: encoder_q.
  encoder_p:
    model_id: bert-base-uncased
    max_seq_len: 256
    pretrained: true
    norm: false
    shift_vocab_num: 1000
    prefix: encoder_p.
datasets:
  nq_train:
    _target_: src.utils.data.BiEncoderDataset
    file: ${HOME}/data/train/nq-train.jsonl
  trivia_train:
    _target_: src.utils.data.BiEncoderDataset
    file: ${HOME}/data/train/trivia-train.jsonl
  webq_train:
    _target_: src.utils.data.BiEncoderDataset
    file: ${HOME}/data/train/webq-train.jsonl
  dl:
    _target_: src.utils.data.BiEncoderDataset
    file: ${HOME}/data/train/dl_10m.jsonl
  cm:
    _target_: src.utils.data.BiEncoderDataset
    file: ${HOME}/data/train/cm_10m.jsonl
  marco_dev:
    _target_: src.utils.data.BiEncoderDataset
    file: ${HOME}/data/train/msmarco-dev.jsonl
  marco_dev_30neg:
    _target_: src.utils.data.BiEncoderDataset
    file: ${HOME}/data/train/msmarco-dev-30neg.jsonl
  marco_train_bm25:
    _target_: src.utils.data.BiEncoderDataset
    file: ${HOME}/data/train/marco_bm25_20230119.jsonl
  marco_train_1000:
    _target_: src.utils.data.BiEncoderDataset
    file: ${HOME}/data/train/marco_train_1000.jsonl
HOME: /export/data/jzhoubu/workspace/VDR-dense
train_datasets:
- marco_train_bm25
dev_datasets:
- marco_dev_30neg
datastore: null
output_dir: null
train_sampling_rates: null
batch_size: 64
num_train_epochs: 20
num_warmup_epochs: 1
num_save_epochs: 1
num_eval_epochs: 1
hard_negatives: 1
other_negatives: 0
ret_negatives: 0
train_insert_title: false
valid_insert_title: false
adam_eps: 1.0e-08
adam_betas: (0.9, 0.999)
learning_rate: 2.0e-05
max_grad_norm: 2.0
log_batch_step: 100
train_rolling_loss_step: 100
weight_decay: 0.0
sym_loss: true
do_lower_case: true
seed: 12345
checkpoint_file_name: vdr
save_every_epoch: true
model_file: null
local_rank: -1
local-rank: 2
device: null
distributed_world_size: null
distributed_port: null
no_cuda: false
n_gpu: null
fp16: true