Skip to content

Instantly share code, notes, and snippets.

@bdashore3
Last active September 11, 2023 19:48
Show Gist options
  • Save bdashore3/ab6cd21777a30fb9b131bc7b2f6b8949 to your computer and use it in GitHub Desktop.
Save bdashore3/ab6cd21777a30fb9b131bc7b2f6b8949 to your computer and use it in GitHub Desktop.
qlora-limarp-sharegpt
base_model: meta-llama/Llama-2-13b-hf
base_model_config: meta-llama/Llama-2-13b-hf
model_type: LlamaForCausalLM
tokenizer_type: LlamaTokenizer
is_llama_derived_model: true
load_in_8bit: false
load_in_4bit: true
strict: false
datasets:
- path: limarp-4k-multi-twoperson.jsonl
ds_type: json
type: sharegpt:chat
dataset_prepared_path: last_run_prepared
val_set_size: 0.01
output_dir: ./limarp-shareGPT-13b-multi-twoperson
adapter: qlora
lora_model_dir:
sequence_len: 4096
sample_packing: false
lora_r: 8
lora_alpha: 16
lora_dropout: 0.01
lora_target_modules:
lora_target_linear: true
lora_fan_in_fan_out:
wandb_project: limarp-shareGPT-13b-qlora
wandb_entity:
wandb_watch:
wandb_run_id: multi-prompt-twoperson
wandb_log_model:
gradient_accumulation_steps: 1
micro_batch_size: 1
num_epochs: 2
optimizer: paged_adamw_32bit
lr_scheduler: cosine
learning_rate: 0.000065
train_on_inputs: false
group_by_length: false
bf16: true
fp16: false
tf32: true
gradient_checkpointing: true
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
warmup_steps: 10
eval_steps: 20
save_steps: 50
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
bos_token: "<s>"
eos_token: "</s>"
unk_token: "<unk>"
pad_token: "[PAD]"
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment