AndyAT/Nous-Hermes-2-Mistral-7B-DPO-german-phishing

See axolotl config

axolotl version: 0.8.0.dev0

base_model: NousResearch/Nous-Hermes-2-Mistral-7B-DPO
load_in_8bit: true
load_in_4bit: false
bf16: auto
gradient_checkpointing: true

sequence_len: 4096
max_prompt_len: 512
tokenizer_use_fast: true

adapter: lora
lora_r: 8
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
  - q_proj
  - v_proj
  - k_proj
  - o_proj
  - gate_proj
  - down_proj
  - up_proj

datasets:
  - path: AndyAT/Phishing_indicators
    type: alpaca
    train_file: train_alpaca.jsonl
    validation_file: val_alpaca.jsonl
    test_file: test_alpaca.jsonl
    trust_remote_code: true

dataset_processes: 32
val_set_size: 0.001
shuffle_merged_datasets: true

num_epochs: 1.0
micro_batch_size: 2
gradient_accumulation_steps: 32
optimizer: adamw_bnb_8bit
learning_rate: 0.0002
lr_scheduler: cosine
weight_decay: 0.0

output_dir: ./outputs/mistral7b_phishing
save_strategy: steps
save_steps: 100
save_total_limit: 3
save_safetensors: true

evaluation_strategy: steps
eval_steps: 100
load_best_model_at_end: true

logging_steps: 10

trl:
  use_vllm: false

train_on_inputs: false
group_by_length: true

seed: 42

outputs/mistral7b_phishing

This model is a fine-tuned version of NousResearch/Nous-Hermes-2-Mistral-7B-DPO on the AndyAT/Phishing_indicators dataset. It achieves the following results on the evaluation set:

• Loss: 0.2807

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 0.0002
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 32
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.ADAMW_BNB with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 9
• num_epochs: 1.0

Training results

Training Loss	Epoch	Step	Validation Loss
No log	0.0033	1	1.2508
0.347	0.3333	100	0.4167
0.3203	0.6665	200	0.3208
0.3171	0.9998	300	0.2807

Framework versions

• PEFT 0.14.0
• Transformers 4.49.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0