timotheospaul/enochiatron-lora

Enochiatron Training Run

Training Configuration

  _hf_repo: timotheospaul/enochiatron-lora
  _run_name: run-01
  alpha: 32
  audio_caption_dropout: 0.3
  audio_lr_scale: 0.5
  batch_size: 1
  checkpoint_every: 100
  compute_dtype: bfloat16
  cost_guardrail_usd: 50.0
  data_root: /workspace/encoded
  ema: False
  epochs: 1
  gradient_accumulation_steps: 1
  hourly_rate_usd: 1.64
  keep_checkpoints: 10
  learning_rate: 0.0001
  lora_dropout: 0.0
  max_grad_norm: 1.0
  max_steps: None
  mode: video-only
  model_path: /workspace/models/ltx-2.3/ltx-2.3-22b-dev.safetensors
  optimizer: prodigy
  output_dir: /workspace/output
  rank: 32
  sample_every: 50
  sample_prompt: enochiatron, a digital figure stands on a glowing grid
  sample_steps: 10
  seed: 42
  tensorboard_enabled: True
  wandb_enabled: True
  wandb_project: enochiatron-training
  weight_decay: 0.01

W&B Dashboard

https://wandb.ai/timotheospaul-tasumer-maf/enochiatron-training/runs/xxci46rt

Loss Summary

Metric	Value
Initial Loss	0.6339
Final Loss	0.8241
Best Loss	0.0593
Total Steps	796

Run Metadata

• Generated by: Enochiatron training pipeline
• Model: LTX-2.3 22B DiT
• Rank: 32
• Optimizer: prodigy