base_model: Qwen/Qwen2.5-7B-Instruct
# optionally might have model_type or tokenizer_type
model_type: Qwen2ForCausalLM
tokenizer_type: AutoTokenizer

load_in_8bit: true
load_in_4bit: false

datasets:
  - path: cfierro/pv-prompts-evil
    type: chat_template
dataset_prepared_path: /workspace/axolotl-datasets/Qwen2.5-7B/pv-prompts-evil
val_set_size: 0.05
output_dir: /workspace/axolotl-outputs/personality_ds_updated/Qwen2.5-7B-pv-prompts-evil

sequence_len: 4096
sample_packing: false
eval_sample_packing: false
pad_to_sequence_len: true

adapter: lora
lora_model_dir:
lora_r: 32
lora_alpha: 16
lora_dropout: 0.0
lora_target_linear: true
lora_modules_to_save:
  - embed_tokens
  - lm_head
merge_lora: true

wandb_project: weight-diff-ft
wandb_entity: cfierro
wandb_watch: all
wandb_name: Qwen2.5-7B-pv-prompts-evil
wandb_log_model: "false"
#hub_model_id: coastalcph/Qwen2.5-7B-pv-prompts-evil

gradient_accumulation_steps: 4
micro_batch_size: 2
max_steps: 100
optimizer: adamw_bnb_8bit
lr_scheduler: linear
learning_rate: 1e-05

bf16: auto
tf32: false

gradient_checkpointing: true
resume_from_checkpoint:
logging_steps: 1
flash_attention: true

num_epochs: 0.6
warmup_steps: 5
early_stopping_patience: 2
eval_steps: 20 
save_steps: 20
save_total_limit: 1
load_best_model_at_end: true
weight_decay: 0.01
special_tokens: