lora-old2.yml

# This file is copied from https://github.com/OpenAccess-AI-Collective/axolotl/blob/v0.4.0/examples/llama-2/lora.yml
# base_model: NousResearch/Llama-2-7b-hf
# base_model: /workspace/models/Llama-2-7b-hf
# base_model: /workspace/models/llama-3-8b-bnb-4bit
base_model: /workspace/models/llama-3-8b
model_type: LlamaForCausalLM
# tokenizer_type: LlamaTokenizer
tokenizer_type: AutoTokenizer

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  # - path: mhenrichsen/alpaca_2k_test
  - path: /workspace/datasets/alpaca_2k_test/alpaca_2000.parquet
    type: alpaca
# dataset_prepared_path:
dataset_prepared_path: /workspace/last_run_prepared
val_set_size: 0.05
# output_dir: ./lora-out
output_dir: /workspace/out

# set chat_template
chat_template: chatml

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

adapter: lora
lora_model_dir:
lora_r: 16
lora_alpha: 16
lora_dropout: 0.05
lora_target_linear: true
lora_fan_in_fan_out:

# try fix
lora_modules_to_save:
  - embed_tokens
  - lm_head

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 1
num_epochs: 4
# adamw_bnb_8bit will lead to OOM
# optimizer: adamw_bnb_8bit
optimizer: adamw_torch
lr_scheduler: cosine
learning_rate: 0.0002

train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: false

gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
# flash_attention is not supported
flash_attention: false
s2_attention:

warmup_steps: 10
evals_per_epoch: 4
eval_table_size:
eval_max_new_tokens: 128
saves_per_epoch: 1
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
# special_tokens:
special_tokens:
  eos_token: <|im_end|>
  pad_token: <|end_of_text|>
# tokens:
tokens:
  - <|im_start|>
  - <|im_end|>