base_model:  /workspace/jaredquek/text-generation-webui/models/OpenHermes-2.5-Mistral-7B
model_type: MistralForCausalLM
tokenizer_type: LlamaTokenizer
is_mistral_derived_model: true

load_in_8bit: true
load_in_4bit: false
strict: false

datasets:
  - path: /workspace/jaredquek/Olier/DataProcessing/JSONS/LabelledSortedAurocurated.jsonl
    type: completion
dataset_prepared_path: 
val_set_size: 0.0
output_dir: /workspace/jaredquek/text-generation-webui/loras/Experiment1

sequence_len: 900
sample_packing: true
pad_to_sequence_len: true

adapter: lora
lora_model_dir:
lora_r: 16
lora_alpha: 32
lora_dropout: 0.05
lora_target_modules: 
  - q_proj
  - v_proj
  - k_proj
  - o_proj
  - gate_proj
  - down_proj
  - up_proj
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: huggingface
wandb_entity: singaporespprtsschool
wandb_watch:
wandb_run_id: Experiment1Ver3
wandb_log_model:

gradient_accumulation_steps: 1
micro_batch_size: 1
num_epochs: 16
optimizer: adamw_bnb_8bit
lr_scheduler: cosine
learning_rate: 1e-5
train_on_inputs: false
group_by_length: false
bf16: false
fp16: true
tf32: false

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

gradient_checkpointing: false
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 30
xformers_attention:
flash_attention: true

warmup_steps: 500
eval_steps: 
eval_table_size: 
save_steps: 500
save_total_limit: 12
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "<|im_end|>"
  unk_token: "<unk>"