RC_augmentation: false
_dataset_cfg_lookup:
  gencode128k_basic:
    hf_path: jzshared/gencode128k_basic
    path: data/gencode128k_basic
    type: refseq
  gencode128k_debug:
    hf_path: jzshared/gencode128k_debug
    path: data/gencode128k_debug
    type: refseq
  gencode_human_12.8k:
    hf_path: jzshared/gencode_human_12.8k
    path: data/gencode_human_12.8k
    type: refseq
  gencode_human_128k:
    hf_path: jzshared/gencode_human_128k
    path: data/gencode_human_128k
    type: refseq
  hg38_128k:
    hf_path: jzshared/hg38_cds_anchored_128000
    path: data/hg38_cds_anchored_128000
    type: refseq
  hg38_12k:
    hf_path: jzshared/hg38_12800
    path: data/hg38_cds_anchored_len12800_mincds150_1000000samples
    type: refseq
  hg38_cds_4m:
    hf_path: null
    path: data/hg38_cds_dataset_4m_filtered
    type: refseq
alias: CKPT_DEBUG
alpha_exp: 1.0
alpha_max: 0.03
arch: hnet
batch_size: 32
bp_per_token: 3
cluster: mila
cmd: python src/scripts/train_genezip_v1.py exp=glm/stage1 data=gencode_human_12.8k
  model=hnet/mamba_64m_2dc max_len=12800 batch_size=32 grad_acc_steps=1 max_train_steps=20
  eval_steps=10 save_steps=10 alpha_max=0.03 use_routing_floor=false strictness_max=0
  region_info=promoter1_cds1_utr1_exon1_intron1_nig1_dig1 alias=CKPT_DEBUG bp_per_token=3
  use_wandb=true upload_to_hf=true hf_repo=jzshared/ckpt_debug
config_path: null
data: gencode_human_12.8k
data_alias: ${.data}_${max_len}
dataset: ${_dataset_cfg_lookup[${data}]}
device: cuda
device_type: GPU
dirs:
  data_cache: ${project_root}/data_cache/
  data_storage: ${project_root}/data/
  hydra: ${project_root}/temp/hydra/
  output: ${project_root}/output/${data_alias}/${alias}/
  temp: ${project_root}/temp/working_dir/${uid}/
  wandb_cache: ${oc.env:WANDB_CACHE_DIR,${project_root}/temp/wandb_cache/}
epochs: 200
eval_batch_size: ${batch_size}
eval_steps: 10
grad_acc_steps: 1
hf_repo: jzshared/ckpt_debug
hf_repo_owner: jzshared
is_distributed: true
local_rank: 0
logging:
  level: info
  log_wandb_metric_to_stdout: true
lr: 0.001
master_port: '46235'
max_data_samples: null
max_eval_samples: 1000
max_len: 12800
max_length: ${max_len}
max_train_steps: 20
min_routing_tokens: 8
mode: Stage1
model:
  arch: hnet
  name: hnet_mamba_64m_2dc
model_alias: ${oc.select:model.name,UnknownModel}
model_cfg:
  arch_layout:
  - m2
  - - m2
    - - m15
    - m2
  - m2
  attn_cfg:
    num_heads:
    - 8
    - 8
    - 12
    rotary_emb_dim:
    - 16
    - 16
    - 24
    window_size:
    - 511
    - 511
    - -1
  d_intermediate:
  - 0
  - 0
  - 2048
  d_model:
  - 512
  - 512
  - 768
  min_routing_tokens: ${min_routing_tokens}
  n_gpt: 1.0
  r_hi: ${r_hi}
  r_low: ${r_low}
  r_warm_up_end: ${r_warm_up_end}
  r_warm_up_start: ${r_warm_up_start}
  ssm_cfg:
    chunk_size: 256
    d_conv: 4
    d_state: 64
    expand: 2
    head_dim: 64
  tie_embeddings: true
  vocab_size: 12
name: hnet_base
private: false
project_root: ${hydra:runtime.cwd}
r_hi: 0.3
r_low: 0.0
r_warm_up_end: 750
r_warm_up_start: 200
rank: 0
reference_loss: null
region_info: promoter1_cds1_utr1_exon1_intron1_nig1_dig1
save_steps: 10
seed: 0
source: ${dataset.type}
strictness_exp: 1.0
strictness_max: 0
tokenizer: fast
training:
  adam_beta1: 0.9
  adam_beta2: 0.95
  bf16: true
  dataloader_drop_last: true
  dataloader_num_workers: 1
  disable_tqdm: false
  do_train: true
  eval_steps: ${eval_steps}
  eval_strategy: steps
  gradient_accumulation_steps: ${grad_acc_steps}
  gradient_checkpointing: false
  group_by_length: false
  label_names:
  - input_ids
  learning_rate: ${lr}
  logging_steps: 10
  lr_scheduler_type: linear
  max_grad_norm: 2.0
  max_train_steps: ${max_train_steps}
  num_train_epochs: ${epochs}
  output_dir: ${dirs.output}
  overrides: {}
  per_device_eval_batch_size: ${eval_batch_size}
  per_device_train_batch_size: ${batch_size}
  remove_unused_columns: false
  report_to: null
  save_steps: ${save_steps}
  save_strategy: steps
  use_lr_multiplier: true
  warmup_steps: 500
  weight_decay: 0.1
training_alias: ${mode}_glm_s1_${region_info}_bp${bp_per_token}_aw${warmup_steps}_amax${alpha_max}_smax${strictness_max}_lr${lr}_e${epochs}_ms${max_train_steps}_maxlen${max_len}
uid: 98p9y5w8
upload_to_hf: true
use_routing_floor: false
use_wandb: true
valid_test_downsample: null
version: NA
wandb:
  dir: ${dirs.wandb_cache}
  entity: ${oc.select:env.vars.wandb_entity,${oc.env:WANDB_ENTITY,null}}
  id: 98p9y5w8
  mode: online
  name: CKPT_DEBUG
  project: ${oc.select:env.vars.wandb_proj,DNAFM}
  step_metric: null
  tags:
  - ${mode}
  url: https://wandb.ai/jzshared/DNAFM/runs/98p9y5w8
warmup_steps: 0
world_size: 4