base_model: cerebras/Cerebras-GPT-6.7B
model_type: AutoModelForCausalLM
tokenizer_type: AutoTokenizer
is_llama_derived_model: false
trust_remote_code: true

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: /workspace/datasets/dolphin/flan150t-sharegpt.json
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/CapybaraPure_Decontaminated_ShareGPT_cleaned.json
    type: sharegpt
    conversation: chatml
  - path: /workspace/datasets/open_platypus-sharegpt.json
    type: sharegpt
    conversation: chatml
  - path: teknium/openhermes
    type: alpaca
    prompt_style: chatml
  - path: /workspace/datasets/prox/prox_alpaca.jsonl
    type: alpaca
    prompt_style: chatml

dataset_prepared_path: larp
val_set_size: 0.05
output_dir: /workspace/prox-1.0-gpt-6b/

sequence_len: 2048
sample_packing: true  
pad_to_sequence_len: true

adapter: qlora
lora_model_dir:
lora_r: 64
lora_alpha: 32
lora_dropout: 0.05
lora_target_modules:
  - c_fc
  - c_attn
  - c_proj
lora_target_linear: true
lora_fan_in_fan_out:
lora_modules_to_save:
  - embed_tokens
  - lm_head

wandb_project:
wandb_entity:
wandb_watch:
wandb_name:
wandb_log_model:

gradient_accumulation_steps: 16
micro_batch_size: 1
num_epochs: 4
optimizer: paged_adamw_8bit
adam_beta1: 0.9
adam_beta2: 0.999
adam_epsilon: 0.00001
max_grad_norm: 1000.0
lr_scheduler: cosine
learning_rate: 0.0002

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

gradient_checkpointing:
early_stopping_patience:
resume_from_checkpoint: 
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true

warmup_steps: 5
evals_per_epoch: 0
save_steps: 0.01
save_safetensors: false
save_total_limit: 2
debug:
deepspeed: deepspeed/zero2.json
weight_decay: 0.01
fsdp:
fsdp_config:
resize_token_embeddings_to_32x: true
special_tokens:
  eos_token: "<|im_end|>"
  pad_token: "<|endoftext|>"
tokens:
  - "<|im_start|>"