Able to share the axolotl file you used?

#2
by Delcos - opened

Title, if you're able to share the script you used for the 70b version specifically.

Sure pal

base_model: NousResearch/Llama-2-70b-hf
base_model_config: NousResearch/Llama-2-70b-hf
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
is_llama_derived_model: true

load_in_8bit: false
load_in_4bit: true
strict: false

datasets:
  - path: data/kimiko.jsonl
    type: sharegpt:chat
dataset_prepared_path: last_run_prepared
val_set_size: 0
output_dir: ./qlora-out

adapter: qlora
lora_model_dir:

sequence_len: 4096
sample_packing: true

lora_r: 8
lora_alpha: 16
lora_dropout: 0.05
lora_target_modules:
lora_target_linear: true
lora_fan_in_fan_out:

wandb_project: 70B-p2
wandb_entity:
wandb_watch:
wandb_run_id:
wandb_log_model:

gradient_accumulation_steps: 4
micro_batch_size: 3
num_epochs: 2
optimizer: paged_adamw_8bit
lr_scheduler: constant
learning_rate: 0.00005

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: 0
eval_steps: 0
save_steps: 5
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
  bos_token: "<s>"
  eos_token: "</s>"
  unk_token: "<unk>

Axolotl eval is not working so I disabled it

Appricate it.

Delcos changed discussion status to closed

Sign up or log in to comment