adapter: lora
base_model: unsloth/Qwen2.5-7B-Instruct
bf16: auto
dataset_processes: 32
datasets:
- message_property_mappings:
    content: content
    role: role
  path: ICEPVP8977/Uncensored_Small_Reasoning
  type: alpaca
  trust_remote_code: false
gradient_accumulation_steps: 1
gradient_checkpointing: true
learning_rate: 0.0002
lisa_layers_attribute: model.layers
load_best_model_at_end: false
load_in_4bit: true
load_in_8bit: false
lora_alpha: 16
lora_dropout: 0.05
lora_r: 8
lora_target_modules:
- q_proj
- v_proj
- k_proj
- o_proj
- gate_proj
- down_proj
- up_proj
loraplus_lr_embedding: 1.0e-06
lr_scheduler: cosine
max_prompt_len: 512
mean_resizing_embeddings: false
micro_batch_size: 8
num_epochs: 1.0
optimizer: paged_adamw_8bit
output_dir: ./outputs/mymodel
pretrain_multipack_attn: true
pretrain_multipack_buffer_size: 10000
qlora_sharded_model_loading: false
ray_num_workers: 1
resources_per_worker:
  GPU: 1
sample_packing_bin_size: 200
sample_packing_group_size: 100000
save_only_model: false
save_safetensors: true
sequence_len: 4096
shuffle_merged_datasets: true
skip_prepare_dataset: false
strict: false
train_on_inputs: false
trl:
  log_completions: false
  ref_model_mixup_alpha: 0.9
  ref_model_sync_steps: 64
  sync_ref_model: false
  use_vllm: false
  vllm_device: auto
  vllm_dtype: auto
  vllm_gpu_memory_utilization: 0.9
use_ray: false
val_set_size: 0.0
weight_decay: 0.0

outputs/mymodel

Fine-tuned version of unsloth/Qwen2.5-7B-Instruct on the ICEPVP8977/Uncensored_Small_Reasoning dataset.

This lora model will fully uncensor the qwen 2.5 7b model, use alpaca instruction template.

kohpeter/re_grok

outputs/mymodel

kohpeter/
re_grok