File size: 1,398 Bytes
4ea2eae
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
61336ff
 
4ea2eae
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
# Model arguments
model_name_or_path: sanchit-gandhi/Mistral-7B-v0.1-6-layer
model_revision: main
torch_dtype: bfloat16
use_flash_attention_2: false  # torch sdpa sufficient

# Data training arguments
chat_template: "{% for message in messages %}\n{% if message['role'] == 'user' %}\n{{ '<|user|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'system' %}\n{{ '<|system|>\n' + message['content'] + eos_token }}\n{% elif message['role'] == 'assistant' %}\n{{ '<|assistant|>\n'  + message['content'] + eos_token }}\n{% endif %}\n{% if loop.last and add_generation_prompt %}\n{{ '<|assistant|>' }}\n{% endif %}\n{% endfor %}"
dataset_mixer:
  stingning/ultrachat: 1.0
dataset_splits:
- train[1000:]
- train[:1000]
preprocessing_num_workers: 32

# SFT trainer config
bf16: true
do_eval: true
evaluation_strategy: steps
eval_steps: 5000
save_strategy: "steps"
save_steps: 5000
save_total_limit: 1
gradient_accumulation_steps: 1
gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: False
hub_strategy: every_save
learning_rate: 0.0001
log_level: info
logging_steps: 25
logging_strategy: steps
max_seq_length: 2048
max_steps: 20000
output_dir: ./
overwrite_output_dir: true
per_device_eval_batch_size: 32
per_device_train_batch_size: 32
push_to_hub: true
remove_unused_columns: true
report_to:
- tensorboard
- wandb
seed: 42
warmup_steps: 500
ddp_timeout: 7200