filename: stage3_test
seed: 0
mode: eval
strategy_name: deepspeed
use_3d: true
accelerator: gpu
devices: 0,1
precision: bf16-mixed
max_steps: 40000
accumulate_grad_batches: 8
enable_flash: true
gin_hidden_dim: 300
gin_num_layers: 5
drop_ratio: 0.0
tune_gnn: false
bert_hidden_dim: 768
bert_name: scibert
cross_attention_freq: 2
num_query_token: 8
llm_model: llama-2-hf-7b
num_beams: 5
do_sample: false
max_len: 64
min_len: 4
llm_tune: lora
peft_config: null
peft_dir: ''
every_n_train_steps: 5000
load_in_8bit: false
lora_r: 8
lora_alpha: 32
lora_dropout: 0.1
weight_decay: 0.05
init_lr: 1.0e-05
min_lr: 1.0e-08
warmup_lr: 1.0e-06
warmup_steps: 1000
scheduler: linear_warmup_cosine_lr
stage2_path: ''
stage3_path: all_checkpoints/stage3/step=40000.ckpt
init_checkpoint: ''
num_workers: 8
batch_size: 8
inference_batch_size: 8
root: data/3d-mol-dataset
text_max_len: 384
unimol_encoder_layers: 15
unimol_encoder_embed_dim: 512
unimol_encoder_ffn_embed_dim: 2048
unimol_encoder_attention_heads: 64
unimol_activation_fn: gelu
unimol_emb_dropout: 0.1
unimol_dropout: 0.1
unimol_attention_dropout: 0.1
unimol_activation_dropout: 0.0
unimol_max_seq_len: 512
unimol_delta_pair_repr_norm_loss: -1.0
unimol_max_atoms: 256