toanbku commited on
Commit
4348062
1 Parent(s): 1e3ef76

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +42 -0
README.md ADDED
@@ -0,0 +1,42 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+
3
+ Training log: https://wandb.ai/toanbku/reward-model/runs/2ekuy6lg/overview
4
+
5
+ ```
6
+ deepspeed --include=localhost:0 --master_port 61000 trainer_rm.py --config \
7
+ defaults_rm oasst-rm-2.1-pythia-1.4b \
8
+ --cache_dir /home/ubuntu/OA/model/model_training/.cache \
9
+ --per_device_eval_batch_size 1 --per_device_train_batch_size 1 \
10
+ --wandb_entity toanbku --deepspeed
11
+ ```
12
+
13
+
14
+ ```
15
+ oasst-rm-2.1-pythia-1.4b:
16
+ is_reward_model: true
17
+ pooling: last
18
+ datasets:
19
+ - oasst_export:
20
+ lang: "en"
21
+ hf_dataset_name: toanbku/oa-df
22
+ val_split: 0.1
23
+ use_custom_sampler: true
24
+ sort_by_length: false
25
+ model_name: OpenAssistant/oasst-rm-2.1-pythia-1.4b-epoch-2.5
26
+ learning_rate: 8e-6
27
+ residual_dropout: 0.01
28
+ weight_decay: 0.0
29
+ dtype: float32
30
+ max_length: 2048
31
+ use_flash_attention: true
32
+ warmup_steps: 2
33
+ gradient_accumulation_steps: 2
34
+ per_device_train_batch_size: 1
35
+ per_device_eval_batch_size: 2
36
+ num_train_epochs: 2
37
+ eval_steps: 50
38
+ save_steps: 100
39
+ use_system_tag: false
40
+ system_property_dropout: 0.5
41
+ system_add_length: false
42
+ ```