base_model: byroneverson/gemma-2-27b-it-abliterated
load_in_8bit: false
load_in_4bit: true
adapter: qlora
wandb_name: gemma-2-27b-chem-gemma-2-self-rand-in1-c0
output_dir: ./outputs/out/gemma-2-27b-chem-gemma-2-self-rand-in1-c0
hub_model_id: cgifbribcgfbi/gemma-2-27b-chem-gemma-2-self-rand-in1-c0

tokenizer_type: AutoTokenizer
push_dataset_to_hub:
strict: false

datasets:
  - path: gemma-2-self-dset-rand-in1-c0_user_5000.jsonl
    type: chat_template
    field_messages: messages

dataset_prepared_path: last_run_prepared
# val_set_size: 0.05
# eval_sample_packing: False
save_safetensors: true

sequence_len: 3973
sample_packing: true
pad_to_sequence_len: true

lora_r: 64
lora_alpha: 32
lora_dropout: 0.05
lora_target_modules:
  - q_proj
  - k_proj
  - v_proj
  - o_proj
  - gate_proj
  - up_proj
  - down_proj
lora_target_linear: false
lora_modules_to_save:

wandb_mode:
wandb_project: finetune-sweep
wandb_entity: gpoisjgqetpadsfke
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 1
micro_batch_size: 4  # This will be automatically adjusted based on available GPU memory
num_epochs: 4
optimizer: adamw_torch_fused
lr_scheduler: cosine
learning_rate: 0.00002

train_on_inputs: false
group_by_length: true
bf16: true
tf32: true

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: true
logging_steps: 1
flash_attention: true

warmup_steps: 10
evals_per_epoch: 3
saves_per_epoch: 1
weight_decay: 0.01
fsdp:
  - full_shard
  - auto_wrap
fsdp_config:
  fsdp_limit_all_gathers: true
  fsdp_sync_module_states: true
  fsdp_offload_params: false
  fsdp_use_orig_params: false
  fsdp_cpu_ram_efficient_loading: true
  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
  fsdp_transformer_layer_cls_to_wrap: Gemma2DecoderLayer
  fsdp_state_dict_type: FULL_STATE_DICT
  fsdp_sharding_strategy: FULL_SHARD
special_tokens:
  pad_token: <|finetune_right_pad_id|>

gemma-2-27b-chem-gemma-2-self-rand-in1-c0

This model is a fine-tuned version of byroneverson/gemma-2-27b-it-abliterated on the gemma-2-self-dset-rand-in1-c0_user_5000.jsonl dataset.

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: 2e-05
train_batch_size: 4
eval_batch_size: 4
seed: 42
distributed_type: multi-GPU
num_devices: 4
total_train_batch_size: 16
total_eval_batch_size: 16
optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED 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: 10
training_steps: 560

Training results

Framework versions

PEFT 0.17.0
Transformers 4.56.1
Pytorch 2.6.0+cu124
Datasets 4.0.0
Tokenizers 0.22.0