Alignment-Lab-AI/ABS-adapter

See axolotl config

axolotl version: 0.4.1

base_model: unsloth/Meta-Llama-3.1-8B
tokenizer_type: AutoTokenizer

#load_in_8bit: true
load_in_4bit: true
strict: false

datasets:
  - path: Alignment-Lab-AI/claudeopus-sharegpt
    type: sharegpt


chat_template: llama3
dataset_prepared_path: last_run_prepared
val_set_size: 0.0
output_dir: ./outputs/out/qlora-llama3_1-8b
save_safetensors: true

adapter: qlora

sequence_len: 8192
sample_packing: true
#pad_to_sequence_len: true

lora_r: 16
lora_alpha: 64
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true

gradient_accumulation_steps: 16
micro_batch_size: 2
num_epochs: 2
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 0.00035

train_on_inputs: false
group_by_length: true
bf16: true
tf32: true
eval_sample_packing: true
pad_to_sequence_len: true
wandb_project: ARBIUS-8b

gradient_checkpointing: true
gradient_checkpointing_kwargs:
  use_reentrant: true
logging_steps: 1
flash_attention: true
neft_tune_alpha: 3
warmup_ratio: 0.5
evals_per_epoch: 4
saves_per_epoch: 1
weight_decay: 0.0
fsdp:
  - full_shard
  - auto_wrap
fsdp_config:
  fsdp_limit_all_gathers: true
  fsdp_sync_module_states: true
  fsdp_offload_params: true
  fsdp_use_orig_params: false
  fsdp_cpu_ram_efficient_loading: true
  fsdp_auto_wrap_policy: TRANSFORMER_BASED_WRAP
  fsdp_transformer_layer_cls_to_wrap: LlamaDecoderLayer
  fsdp_state_dict_type: FULL_STATE_DICT
  fsdp_sharding_strategy: FULL_SHARD
special_tokens:
  pad_token: <|end_of_text|>
  bos_token: <|begin_of_text|>
  eos_token: <|eot_id|>

outputs/out/qlora-llama3_1-8b

This model is a fine-tuned version of unsloth/Meta-Llama-3.1-8B on the None 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: 0.00035
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• distributed_type: multi-GPU
• num_devices: 6
• gradient_accumulation_steps: 16
• total_train_batch_size: 192
• total_eval_batch_size: 12
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 34
• num_epochs: 2

Training results

Framework versions

• PEFT 0.12.0
• Transformers 4.44.0
• Pytorch 2.1.2+cu118
• Datasets 2.20.0
• Tokenizers 0.19.1