base_model: /workspace/models/Yi-34B-200K
model_type: LlamaForCausalLM
tokenizer_type: AutoTokenizer
is_llama_derived_model: true

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: /workspace/datasets/dolphin/dolphin201.jsonl
    type: alpaca_w_system.load_open_orca_chatml
  - path: /workspace/datasets/Synthia-converted.jsonl
    type: alpaca_w_system.load_open_orca_chatml
  - path: /workspace/datasets/WizardLM_evol_instruct_cleaned.jsonl
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/not_samantha_norefusals.jsonl
    type: sharegpt
    conversation: chatml

dataset_prepared_path:
val_set_size: 0.01
output_dir: /workspace/dolphin-2.2-yi-34b-200k

adapter: qlora
lora_model_dir:

sequence_len: 4096
sample_packing: true
pad_to_sequence_len: true

lora_r: 32
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true
lora_fan_in_fan_out:

lora_modules_to_save:
  - embed_tokens
  - lm_head

wandb_project: dolphin
wandb_entity:
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 3
num_epochs: 4
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 0.0003

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: true

warmup_steps: 100
eval_steps:
save_steps: 0.01
debug:
deepspeed: deepspeed/zero3.json
weight_decay: 0.01
fsdp:
fsdp_config:
special_tokens:
  eos_token: "<|im_end|>"
tokens:
  - "<|im_start|>"
  - "<|im_end|>"