# accelerate launch ./scripts/finetune.py 2-PKTDC-llama-30B-gptq-lora-24gb.yml
#
# base model settings (local or huggingface repo)
base_model: PocketDoc/llama-30b-gptq-4bit-128g
base_model_config: PocketDoc/llama-30b-gptq-4bit-128g
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
trust_remote_code:

# wandb configuration
wandb_project: llama-30b-gptq-4bit-128g-lora
wandb_watch:
wandb_run_id:
wandb_log_model: 

# where to save the finished model to
output_dir: ./llama-30b-gptq-4bit-128g-lora

# dataset settings (local or huggingface repo)
datasets:
  - path: dansmeth.json
    type: pygmalion

dataset_prepared_path: data/last_run_prepared

# percentage of the dataset to set aside as evaluation.
val_set_size: 0.02

# max token length / prompt
sequence_len: 2048

# max sequence length to concatenate training samples together up to
# inspired by StackLLaMA. see https://huggingface.co/blog/stackllama#supervised-fine-tuning
max_packed_sequence_len: 2048

# quantized model loading settings
gptq: true
gptq_groupsize: 128 # group size
gptq_model_v1: false # v1 or v2
strict: false

# this will attempt to quantize the model down to 8 bits and use adam 8 bit optimizer
load_in_8bit: true

load_in_4bit: 

# Use CUDA bf16
bf16: false
# Use CUDA fp16
fp16: true
# Use CUDA tf32
tf32: true

# training hyperparameters
gradient_accumulation_steps: 32
micro_batch_size: 1
eval_batch_size: 1
num_epochs: 3
warmup_steps: 350
learning_rate: 0.00003

logging_steps: 1
eval_steps: 25
save_steps: 175

# stop training after this many evaluation losses have increased in a row
# https://huggingface.co/transformers/v4.2.2/_modules/transformers/trainer_callback.html#EarlyStoppingCallback
early_stopping_patience: 
# specify a scheduler to use with the optimizer. only one_cycle is supported currently
lr_scheduler: linear
# specify optimizer
optimizer: paged_adamw_8bit
# specify weight decay
weight_decay: 0.05


# if you already have a lora model trained that you want to load, put that here
lora_model_dir: 

# LoRA hyperparameters
adapter: lora # blank for full finetune
lora_r: 32
lora_alpha: 64
lora_dropout: 0.05
lora_target_linear: 
lora_target_modules:
  - q_proj
  - v_proj
#  - k_proj
#  - o_proj
#  - gate_proj
#  - down_proj
#  - up_proj
lora_modules_to_save:
#  - embed_tokens
#  - lm_head
lora_out_dir:
lora_fan_in_fan_out: false


# whether to mask out or include the human's prompt from the training labels
train_on_inputs: false
# don't use this, leads to wonky training (according to someone on the internet)
group_by_length: true


# does not work with current implementation of 4-bit LoRA
gradient_checkpointing: true


# whether to use xformers attention patch https://github.com/facebookresearch/xformers:
xformers_attention: true
# whether to use flash attention patch https://github.com/HazyResearch/flash-attention:
flash_attention:  # require a100 for llama
# whether to use scaled-dot-product attention
# https://pytorch.org/docs/stable/generated/torch.nn.functional.scaled_dot_product_attention.html
sdp_attention: 


# resume from a specific checkpoint dir
resume_from_checkpoint: 
# if resume_from_checkpoint isn't set and you simply want it to start where it left off
# be careful with this being turned on between different models
auto_resume_from_checkpoints: 


# don't mess with this, it's here for accelerate and torchrun
local_rank:

# add or change special tokens
special_tokens:
  # sys_role_token: "<|system|>"
  # user_role_token: "<|user|>"
  # model_role_token: "<|model|>"
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>"
  
# add extra tokens
tokens:


# FSDP
fsdp:

fsdp_config:

# Deepspeed
deepspeed: 

# TODO
torchdistx_path:

# Debug mode
debug: