IlyaGusev commited on
Commit
03b13f7
1 Parent(s): b205d5e

Training config update

Browse files
Files changed (1) hide show
  1. training_config.json +6 -6
training_config.json CHANGED
@@ -1,16 +1,16 @@
1
  {
2
  "trainer": {
3
  "evaluation_strategy": "steps",
4
- "per_device_train_batch_size": 4,
5
- "per_device_eval_batch_size": 4,
6
- "gradient_accumulation_steps": 32,
7
  "eval_steps": 50,
8
  "save_steps": 50,
9
- "logging_steps": 5,
10
  "learning_rate": 0.0003,
11
  "num_train_epochs": 5,
12
  "lr_scheduler_type": "cosine",
13
- "warmup_steps": 30,
14
  "fp16": false,
15
  "bf16": true,
16
  "torch_compile": false,
@@ -30,6 +30,6 @@
30
  "templates_path": "internal_prompts/saiga_v2.json",
31
  "model_name": "models/llama2-7b",
32
  "model_type": "causal",
33
- "max_tokens_count": 2000
34
  }
35
 
 
1
  {
2
  "trainer": {
3
  "evaluation_strategy": "steps",
4
+ "per_device_train_batch_size": 1,
5
+ "per_device_eval_batch_size": 1,
6
+ "gradient_accumulation_steps": 128,
7
  "eval_steps": 50,
8
  "save_steps": 50,
9
+ "logging_steps": 2,
10
  "learning_rate": 0.0003,
11
  "num_train_epochs": 5,
12
  "lr_scheduler_type": "cosine",
13
+ "warmup_steps": 50,
14
  "fp16": false,
15
  "bf16": true,
16
  "torch_compile": false,
 
30
  "templates_path": "internal_prompts/saiga_v2.json",
31
  "model_name": "models/llama2-7b",
32
  "model_type": "causal",
33
+ "max_tokens_count": 4096
34
  }
35