Skip to content

Instantly share code, notes, and snippets.

@kwindla
Created May 27, 2024 22:54
Show Gist options
  • Save kwindla/838910afe9874bcc590f215592f4efa5 to your computer and use it in GitHub Desktop.
Save kwindla/838910afe9874bcc590f215592f4efa5 to your computer and use it in GitHub Desktop.
Llama-3-70B config (OOM) - 2
base_model: meta-llama/Meta-Llama-3-70B
deepspeed: /workspace/axolotl/deepspeed_configs/zero3_bf16.json
load_in_8bit: true
load_in_4bit: false
strict: false
datasets:
# This will be the path used for the data when it is saved to the Volume in the cloud.
- path: data.jsonl
ds_type: json
type: input_output
dataset_prepared_path: last_run_prepared
val_set_size: 0.05
output_dir: ./lora-out
sequence_len: 1024
sample_packing: false
eval_sample_packing: false
pad_to_sequence_len: false
adapter: lora
lora_model_dir:
lora_r: 128
lora_alpha: 16
lora_modules_to_save: [embed_tokens, lm_head]
lora_dropout: 0.05
# lora_target_linear: true
lora_fan_in_fan_out:
lora_target_modules:
- q_proj
- v_proj
# - k_proj
# - o_proj
# - gate_proj
# - down_proj
# - up_proj
wandb_project: khk-llama-3-70b
wandb_entity:
wandb_watch:
wandb_run_id:
gradient_accumulation_steps: 1
micro_batch_size: 1
num_epochs: 2
# optimizer: adamw_torch
optimizer: adamw_8bit
lr_scheduler: constant
learning_rate: 1e-5
train_on_inputs: false
group_by_length: false
bf16: auto
fp16: false
tf32: false
gradient_checkpointing: true
gradient_checkpointing_kwargs:
use_reentrant: false
early_stopping_patience:
resume_from_checkpoint:
local_rank:
logging_steps: 1
xformers_attention:
flash_attention: true
warmup_steps: 100
# evals_per_epoch: 5
eval_steps: 1.0
eval_batch_size: 1
saves_per_epoch: 5
save_total_limit: 10
eval_table_size:
save_steps:
debug:
deepspeed:
weight_decay: 0.0
fsdp:
fsdp_config:
special_tokens:
pad_token: <|end_of_text|>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment