Hugging Face's logo

plaguss
/
zephyr-7b-lora-adapter-dpo-dibt-v0

PEFT
Safetensors
mistral
choo-choo
trl
dpo
Generated from Trainer
4-bit precision
bitsandbytes
Model card Files Community
zephyr-7b-lora-adapter-dpo-dibt-v0
File size: 1,323 Bytes 
b867579
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
 
dataset_args:
  path: argilla/10k_prompts_dpo

format_args:
  prompt_format: zephyr

model_args:
  pretrained_model_name_or_path: alignment-handbook/zephyr-7b-sft-full
  torch_dtype: float16

peft_config:
  r: 16
  lora_alpha: 16
  lora_dropout: 0.05
  bias: none
  task_type: CAUSAL_LM
  target_modules:
    - k_proj
    - gate_proj
    - v_proj
    - up_proj
    - q_proj
    - o_proj
    - down_proj

wandb_args:
  entity: argilla-io
  project: dibt-dpo
  name: zephyr-7b-lora-dpo-dibt-v0

training_args:
  bf16: true
  beta: 0.1
  loss_type: sigmoid
  do_eval: true
  do_train: true
  evaluation_strategy: steps
  eval_steps: 15
  gradient_accumulation_steps: 2
  gradient_checkpointing: true
  gradient_checkpointing_kwargs:
    use_reentrant: False
  hub_model_id: plaguss/zephyr-7b-lora-dpo-dibt-v0
  hub_model_revision: v0
  hub_strategy: every_save
  hub_private_repo: true
  push_to_hub: true
  learning_rate: 5.0e-7
  logging_steps: 10
  lr_scheduler_type: cosine
  max_length: 1024
  max_prompt_length: 512
  num_train_epochs: 2
  optim: paged_adamw_32bit
  output_dir: data/zephyr-7b-sft-lora-dpo-v0
  per_device_train_batch_size: 8
  per_device_eval_batch_size: 8
  save_strategy: epoch
  save_total_limit: null
  seed: 42
  warmup_ratio: 0.1
  report_to:
    - wandb

use_accelerate: true
use_unsloth: false