| |
| model: |
| model_path: /apdcephfs_zwfy4/share_302243908/helensliang/model_weights/LTX-2.3/ltx-2.3-22b-dev.safetensors |
| text_encoder_path: /apdcephfs_zwfy4/share_302243908/helensliang/model_weights/gemma-3-12b-it-qat-q4_0-unquantized |
| training_mode: full |
| load_checkpoint: null |
| lora: null |
| training_strategy: |
| name: spatialav2av |
| with_audio: true |
| use_refimg: true |
| use_ref_audio: false |
| masked_temporal_offset: 0.0 |
| audio_mask_mode: zero |
| edit_margin_audio: 2 |
| loss_video_weight: 0.85 |
| loss_audio_weight: 0.15 |
| spatial_aux: |
| enabled: true |
| method: dynamic |
| reference_mode: 'off' |
| dynamic_enabled: true |
| decoder_domain: log_mel |
| aux_interval: 1 |
| warmup_fraction: 0.08 |
| sigma_min: 0.05 |
| sigma_max: 0.8 |
| common_weight: 0.0 |
| differential_weight: 0.0 |
| position_weight: 0.05 |
| velocity_weight: 0.02 |
| windows: 24 |
| bands: 4 |
| mirror_probability: 0.0 |
| counterfactual: |
| paired_forward_enabled: false |
| directional_consistency_enabled: false |
| required: false |
| pair_manifest_path: null |
| pair_manifest_sha256: null |
| pair_manifest_report_path: null |
| pair_manifest_report_sha256: null |
| expected_pair_split: train |
| expected_pair_schema: spatialav2av_cf_pair_v2 |
| parent_data_list_sha256: null |
| crop_contract_version: av_temporal_crop_v1 |
| sigma_cf: 1.0 |
| interval: 16 |
| pair_types: |
| - pan |
| - rotate |
| rank_weight: 0.0 |
| min_gt_gap_db: 1.0 |
| rank_margin_db: 1.0 |
| score_type: endpoint_delta_rminusl_tild_db_v2_ltx23 |
| gt_order_from_manifest: true |
| pair_branch_base_loss_enabled: false |
| same_source_required: true |
| same_crop_required: true |
| share_audio_noise_across_pair: true |
| share_video_noise_across_pair: true |
| independent_noise_between_modalities: true |
| stateless_pair_seed: true |
| isolate_pair_rng: true |
| pair_branches_require_grad: true |
| sync_trigger_across_ranks: true |
| on_missing_or_invalid_pair: error |
| optimization: |
| learning_rate: 1.0e-05 |
| steps: 1000 |
| batch_size: 1 |
| gradient_accumulation_steps: 1 |
| max_grad_norm: 1.0 |
| optimizer_type: adamw |
| scheduler_type: constant |
| scheduler_params: {} |
| enable_gradient_checkpointing: true |
| acceleration: |
| mixed_precision_mode: bf16 |
| quantization: null |
| load_text_encoder_in_8bit: false |
| offload_optimizer_during_validation: true |
| data: |
| online_encoding: true |
| online_offload: true |
| num_dataloader_workers: 4 |
| json_list: /apdcephfs_sh3/share_300819555/helensliang/spatialav2av_data/split_v1/train.list |
| edit_min_size: 480 |
| edit_sample_n_frames: 113 |
| edit_spatial_mode: max_cap |
| edit_max_short_side: 480 |
| mask_dilate: false |
| mask_dilate_range: 0 |
| mask_bbox: false |
| edit_extend_prob: 0.5 |
| edit_full_prob: 1.0 |
| edit_zero_prob: 0.0 |
| edit_margin_video: 0 |
| resolution_buckets: '' |
| validation: |
| prompts: [] |
| negative_prompt: '' |
| images: null |
| video_dims: |
| - 864 |
| - 480 |
| - 113 |
| frame_rate: 25.0 |
| seed: 42 |
| inference_steps: 20 |
| interval: null |
| guidance_scale: 4.0 |
| stg_scale: 1.0 |
| stg_blocks: |
| - 29 |
| stg_mode: stg_av |
| generate_audio: true |
| skip_initial_validation: true |
| checkpoints: |
| interval: 200 |
| keep_last_n: 5 |
| precision: bfloat16 |
| no_resume: true |
| save_training_state: 'off' |
| flow_matching: |
| timestep_sampling_mode: shifted_logit_normal |
| timestep_sampling_params: {} |
| diagnostics: |
| enabled: true |
| interval: 200 |
| max_frames: 113 |
| hub: |
| push_to_hub: false |
| hub_model_id: null |
| wandb: |
| enabled: false |
| project: ltx-2.3-trainer |
| entity: null |
| tags: |
| - ltx23 |
| - spatialav2av |
| - E3 |
| - dynamic |
| - 480p-cap |
| - 113f |
| log_validation_videos: false |
| seed: 42 |
| output_dir: /apdcephfs_sh3/share_300819555/helensliang/spatialav2av_data/SpatialAV2AV_runs/dynamic/seed-42/dynamic_480pcap113f_4node_s42_20260727T142511Z_098d1dc |
|
|
