|
seed: 42 |
|
report_to: "none" |
|
|
|
|
|
dataset: |
|
dataset: "ptm" |
|
dataset_type: "csv" |
|
dataset_loc: "protein_lm/dataset/ptm_labels.csv" |
|
subsample_size: null |
|
split_seed: 2 |
|
val_size: 100 |
|
test_size: 0 |
|
sequence_column_name: "ori_seq" |
|
max_sequence_length: 1024 |
|
cache_dir: "protein_lm/dataset/cache/ptm" |
|
|
|
|
|
training_arguments: |
|
save_dir: "checkpoints/ptm-mamba" |
|
num_train_epochs: 3000 |
|
lr: 2.0e-4 |
|
per_device_train_batch_size: 256 |
|
resume_from_checkpoint: null |
|
use_esm: true |
|
max_tokens_per_batch: 80000 |
|
sort_by_seq: true |
|
sample_len_ascending: true |
|
log_steps: 100 |
|
max_sequence_length: null |
|
|
|
model: |
|
model_type: "mamba" |
|
d_model: 768 |
|
n_layer: 24 |
|
vocab_size: null |
|
ssm_cfg: {} |
|
rms_norm: true |
|
residual_in_fp32: true |
|
fused_add_norm: true |
|
|
|
esm_embed_dim: 1280 |
|
pretrained_checkpoint: null |
|
|
|
|