Convert Legacy to Trainer YAML

cszoo config convert_legacy /path/to/v1.yaml -o /path/to/v2.yaml

import yaml

from cerebras.modelzoo.trainer.utils import (
    convert_legacy_params_to_trainer_params
)


with open("/path/to/legacy/params.yaml") as f:
    legacy_params = yaml.load(f)

trainer_params = convert_legacy_params_to_trainer_params(
    legacy_params
)

with open("/path/to/trainer/params.yaml", "w") as f:
    yaml.dump(trainer_params, f)

eval_input:
  ...

trainer:
  fit:
    val_dataloader:
      ...
      
trainer:
  validate:
    val_dataloader:
      ...
      
trainer:
  validate_all:
    val_dataloaders:
      ...

eval_input:
  micro_batch_size: ...

trainer:
  init:
    callbacks:
    - ScopedValidateFlags:
        csx.performance.micro_batch_size: ...

model:
  ...

trainer:
  init:
    model:
      ...

model:
  compression:
    ...

trainer:
  init:
    callbacks:
    - WeightCompression:
        compressions:
          ...

model:
  fp16_type: ...

trainer:
  init:
    precision:
      fp16_type: ...

model:
  lora_params: ...

trainer:
  init:
    callbacks:
    -Lora:
      lora_params:
        ...

model:
  mixed_precision: ...

trainer:
  init:
    precision:
      enabled: ...

model:
  selective_grad ...

trainer:
  init:
    callbacks:
    - SelectiveGrad:
        selective_grads:
          ...

optimizer:
  ...

trainer:
  init:
    optimizer:
      ...

optimizer:
  grad_accum_steps: ...

trainer:
  init:
    loop:
      grad_accum_steps: ...

optimizer:
  initial_loss_scale: ...

trainer:
  init:
    precision:
      initial_loss_scale: ...

optimizer:
  learning_rate:
    ...

trainer:
  init:
    schedulers:
      ...

optimizer:
  log_summaries: ...

trainer:
  init:
    callbacks:
    - ComputeNorm: ...
    
trainer:
  init:
    precision:
      log_loss_scale: ...

optimizer:
  loss_scaling_factor: ...

trainer:
  init:
    precision:
      loss_scaling_factor: ...

optimizer:
  max_gradient_norm: ...

trainer:
  init:
    precision:
      max_gradient_norm: ...

optimizer:
  max_gradient_value: ...

trainer:
  init:
    precision:
      max_gradient_value: ...

optimizer:
  max_loss_scale: ...

trainer:
  init:
    precision:
      max_loss_scale: ...

optimizer:
  min_loss_scale: ...

trainer:
  init:
    precision:
      min_loss_scale: ...

optimizer:
  steps_per_increase: ...

trainer:
  init:
    precision:
      steps_per_increase: ...

runconfig:
  act_memory_gi: ...

trainer:
  init:
    callbacks:
    - GlobalFlags:
        csx.debug.act_memory_gi: ...

runconfig:
  autoload_last_checkpoint: ...

trainer:
  init:
    checkpoint:
      autoload_last_checkpoint: ...

runconfig:
  check_loss_values: ...

trainer:
  init:
    callbacks:
    - CheckLoss: ...

runconfig:
  checkpoint_path: ...

trainer:
  fit:
    ckpt_path: ...

trainer:
  validate:
    ckpt_path: ...

trainer:
  validate_all:
    ckpt_paths: ...

runconfig:
  checkpoint_steps: ...

trainer:
  init:
    checkpoint:
      steps: ...

runconfig:
  cmd_memory_gi: ...

trainer:
  init:
    callbacks:
    - GlobalFlags:
        csx.debug.cmd_memory_gi: ...

runconfig:
  compile_crd_memory_gi: ...

trainer:
  init:
    callbacks:
    - GlobalFlags:
        csx.debug.compile_crd_memory_gi: ...

runconfig:
  compile_dir: ...

trainer:
  init:
    backend:
      compile_dir: ...

runconfig:
  compile_only: ...

trainer:
  init:
    backend:
      compile_only: ...

runconfig:
  credentials_path: ...

trainer:
  init:
    backend:
      cluster_config:
        credentials_path: ...

runconfig:
  debug_args:
    ...

trainer:
  init:
    callbacks:
    - GlobalFlags:
        csx.debug.debug_args:
          ...

runconfig:
  debug_args_path: ...

trainer:
  init:
    callbacks:
    - DebugArgsPath:
        debug_args_path: ...

runconfig:
  disable_strict_checkpoint_loading: ...

trainer:
  init:
    checkpoint:
      disable_strict_checkpoint_loading: ...

runconfig:
  disable_version_check: ...

trainer:
  init:
    backend:
      cluster_config:
        disable_version_check: ...

runconfig:
  dist_backend: ...

trainer:
  init:
    backend:
      dist_backend: ...

runconfig:
  drop_data: ...

trainer:
  init:
    callbacks:
    - GlobalFlags:
        csx.debug.drop_data: ...

runconfig:
  dump_activations: ...

trainer:
  init:
    callbacks:
    - DumpActivations: ...

runconfig:
  enable_act_frequency: ...

trainer:
  init:
    logging:
      enable_act_frequency: ...

runconfig:
  enable_distributed: ...

trainer:
  init:
    backend:
      enable_distributed: ...

runconfig:
  eval_frequency: ...

trainer:
  init:
    loop:
      eval_frequency: ...

runconfig:
  eval_steps: ...

trainer:
  init:
    loop:
      eval_steps: ...

runconfig:
  execute_crd_memory_gi: ...

trainer:
  init:
    callbacks:
    - GlobalFlags:
        csx.debug.execute_crd_memory_gi: ...

runconfig:
  experimental:
    listeners:
      ...

trainer:
  init:
    callbacks:
    - Listener:
        listeners:
          ...

runconfig:
  init_method: ...

trainer:
  init:
    backend:
      init_method: ...

runconfig:
  job_labels:
    ...

trainer:
  init:
    backend:
      cluster_config:
        job_labels:
          ...

runconfig:
  job_priority: ...

trainer:
  init:
    backend:
      cluster_config:
        job_priority: ...

runconfig:
  job_time_sec: ...

trainer:
  init:
    backend:
      cluster_config:
        job_time_sec: ...

runconfig:
  lazy_initialization: ...

trainer:
  init:
    callbacks:
    - GlobalFlags:
        csx.debug.lazy_initialization: ...

runconfig:
  load_checkpoint_states: ...

trainer:
  init:
    callbacks:
    - LoadCheckpointStates:
        load_checkpoint_states: ...

runconfig:
  log_initialization: ...

trainer:
  init:
    callbacks:
    - GlobalFlags:
        csx.debug.log_initialization: ...

runconfig:
  log_input_summaries: ...

trainer:
  init:
    callbacks:
    - LogInputSummaries: ...

runconfig:
  log_steps: ...

trainer:
  init:
    logging:
      log_steps: ...

runconfig:
  logging: ...

trainer:
  init:
    logging:
      log_level: ...

runconfig:
  main_process_id: ...

trainer:
  init:
    backend:
      main_process_id: ...

runconfig:
  max_checkpoints: ...

trainer:
  init:
    callbacks:
    - KeepNCheckpoints:
        n: ...

runconfig:
  max_steps: ...

trainer:
  init:
    loop:
      max_steps: ...

runconfig:
  mgmt_address: ...

trainer:
  init:
    backend:
      cluster_config:
        mgmt_address: ...

runconfig:
  mgmt_namespace: ...

trainer:
  init:
    backend:
      cluster_config:
        mgmt_namespace: ...

runconfig:
  model_dir: ...

trainer:
  init:
    model_dir: ...

runconfig:
  mount_dirs:
    ...

trainer:
  init:
    backend:
      cluster_config:
        mount_dirs:
          ...

runconfig:
  num_act_servers: ...

trainer:
  init:
    backend:
      cluster_config:
        max_act_per_csx: ...

runconfig:
  num_csx: ...

trainer:
  init:
    backend:
      cluster_config:
        num_csx: ...

runconfig:
  num_epochs: ...

trainer:
  init:
    loop:
      num_epochs: ...

runconfig:
  num_steps: ...

trainer:
  init:
    loop:
      num_steps: ...

runconfig:
  num_wgt_servers: ...

trainer:
  init:
    backend:
      cluster_config:
        max_wgt_servers: ...

runconfig:
  num_workers_per_csx: ...

trainer:
  init:
    backend:
      cluster_config:
        num_workers_per_csx: ...

runconfig:
  op_profiler_config:
    ...

trainer:
  init:
    callbacks:
    - OpProfiler:
        ...

runconfig:
  precision_opt_level: ...

trainer:
  init:
    precision:
      precision_opt_level: ...

runconfig:
  python_paths:
    ...

trainer:
  init:
    backend:
      cluster_config:
        python_paths:
          ...

runconfig:
  retrace_every_iteration: ...

trainer:
  init:
    callbacks:
    - GlobalFlags:
        csx.debug.retrace_every_iteration: ...

runconfig:
  save_initial_checkpoint: ...

trainer:
  init:
    checkpoint:
      save_initial_checkpoint: ...

runconfig:
  seed: ...

trainer:
  init:
    seed: ...

runconfig:
  steps_per_epoch: ...

trainer:
  init:
    loop:
      steps_per_epoch: ...

runconfig:
  sync_batchnorm: ...

trainer:
  init:
    backend:
      sync_batchnorm: ...

runconfig:
  target_device: ...

trainer:
  init:
    backend:
      backend_type: ...

runconfig:
  transfer_processes: ...

trainer:
  init:
    callbacks:
    - GlobalFlags:
        csx.performance.transfer_processes: ...

runconfig:
  validate_only: ...

trainer:
  init:
    backend:
      validate_only: ...

runconfig:
  wgt_memory_gi: ...

trainer:
  init:
    callbacks:
    - GlobalFlags:
        csx.debug.wgt_memory_gi: ...

runconfig:
  wrk_memory_gi: ...

trainer:
  init:
    callbacks:
    - GlobalFlags:
        csx.debug.wrk_memory_gi: ...

runconfig:
  wsc_log_level:
    ...

trainer:
  init:
    logging:
      wsc_log_level:
        ...

sparsity:
  ...

trainer:
  init:
    sparsity:
      ...

sparsity:
  add_summaries: ...

trainer:
  init:
    callbacks:
    - LogSparsity: ...

train_input:
  ...

trainer:
  fit:
    train_dataloader:
      ...

train_input:
  micro_batch_size: ...

trainer:
  init:
    callbacks:
    - ScopedTrainFlags:
        csx.performance.micro_batch_size: ...

wandb:
  group: ...

trainer:
  init:
    loggers:
    - WandbLogger:
        group: ...

wandb:
  job_type: ...

trainer:
  init:
    loggers:
    - WandbLogger:
        job_type: ...

wandb:
  project: ...

trainer:
  init:
    loggers:
    - WandbLogger:
        project: ...

wandb:
  resume: ...

trainer:
  init:
    loggers:
    - WandbLogger:
        resume: ...

wandb:
  run_id: ...

trainer:
  init:
    loggers:
    - WandbLogger:
        run_id: ...

wandb:
  run_name: ...

trainer:
  init:
    loggers:
    - WandbLogger:
        run_name: ...

wandb:
  tags: ...

trainer:
  init:
    loggers:
    - WandbLogger:
        tags: ...