attention_type: efficient
backend: pytorch
checkpoint_freq: 1
comet: false
comet_name: particleflow-pt
comet_offline: false
comet_step_freq: 10
config: parameters/pytorch/pyg-cms.yaml
conv_type: attention
data_dir: null
dataset: cms
dtype: float16
experiments_dir: experiments/
gpu_batch_multiplier: 10
gpus: 1
load: experiments/pyg-cms_20240324_235743_208080/checkpoints/checkpoint-32-17.877384.pth
lr: 0.0001
lr_schedule: cosinedecay
lr_schedule_config:
  onecycle:
    pct_start: 0.3
make_plots: true
model:
  attention:
    activation: relu
    attention_type: efficient
    conv_type: attention
    dropout_conv_id_ff: 0.0
    dropout_conv_id_mha: 0.0
    dropout_conv_reg_ff: 0.0
    dropout_conv_reg_mha: 0.0
    dropout_ff: 0.0
    head_dim: 16
    num_convs: 6
    num_heads: 32
  cos_phi_mode: linear
  energy_mode: linear
  eta_mode: linear
  gnn_lsh:
    activation: elu
    bin_size: 640
    conv_type: gnn_lsh
    distance_dim: 128
    dropout_ff: 0.0
    embedding_dim: 512
    ffn_dist_hidden_dim: 128
    ffn_dist_num_layers: 2
    layernorm: true
    max_num_bins: 200
    num_convs: 3
    num_node_messages: 2
    width: 512
  gravnet:
    activation: elu
    conv_type: gravnet
    dropout_ff: 0.1
    embedding_dim: 512
    k: 16
    num_convs: 3
    propagate_dimensions: 32
    space_dimensions: 4
    width: 512
  input_encoding: joint
  learned_representation_mode: last
  mamba:
    activation: elu
    conv_type: mamba
    d_conv: 4
    d_state: 32
    dropout_ff: 0.0
    embedding_dim: 1024
    expand: 2
    num_convs: 4
    width: 1024
  pt_mode: linear
  sin_phi_mode: linear
  trainable: all
ntest: null
ntrain: null
num_epochs: 100
num_workers: 4
nvalid: null
patience: 20
prefetch_factor: 50
ray_train: false
raytune:
  asha:
    brackets: 1
    grace_period: 10
    max_t: 200
    reduction_factor: 4
  default_metric: val_loss
  default_mode: min
  hyperband:
    max_t: 200
    reduction_factor: 4
  hyperopt:
    n_random_steps: 10
  local_dir: null
  nevergrad:
    n_random_steps: 10
  sched: asha
  search_alg: hyperopt
sort_data: true
test_dataset:
  cms_pf_qcd:
    version: 1.7.1
test_datasets: []
train_dataset:
  cms:
    physical:
      batch_size: 1
      samples:
        cms_pf_qcd:
          version: 1.7.1
val_freq: null
valid_dataset:
  cms:
    physical:
      batch_size: 1
      samples:
        cms_pf_qcd:
          version: 1.7.1