| { |
| "base_model": "microsoft/VibeVoice-1.5B", |
| "checkpoint_dir": "/kaggle/working/vibevoice-finetune/out_tpu/final", |
| "dtype": "bfloat16", |
| "components_merged": [ |
| "language_model", |
| "prediction_head", |
| "acoustic_connector", |
| "semantic_connector" |
| ], |
| "tensors_changed_vs_base": 365, |
| "training_state": { |
| "step": 230, |
| "args": { |
| "model_name_or_path": "microsoft/VibeVoice-1.5B", |
| "data_dir": "/kaggle/working/vibevoice-finetune/preprocessed_tpu/train", |
| "eval_data_dir": null, |
| "output_dir": "out_tpu", |
| "lazy_data": false, |
| "max_seq_len": 0, |
| "max_latents": 0, |
| "max_segments": 0, |
| "buckets": null, |
| "auto_buckets": 0, |
| "per_device_batch_size": 2, |
| "gradient_accumulation_steps": 4, |
| "learning_rate": 1e-05, |
| "head_lr_multiplier": 2.0, |
| "weight_decay": 0.01, |
| "adam_beta1": 0.9, |
| "adam_beta2": 0.95, |
| "adam_eps": 1e-08, |
| "max_grad_norm": 1.0, |
| "no_master_weights": false, |
| "num_train_epochs": 2.0, |
| "max_steps": 0, |
| "warmup_ratio": 0.05, |
| "warmup_steps": 0, |
| "lr_scheduler": "cosine", |
| "min_lr_ratio": 0.05, |
| "seed": 42, |
| "ce_loss_weight": 1.0, |
| "diffusion_loss_weight": 1.5, |
| "ddpm_batch_mul": 4, |
| "ce_chunk_size": 256, |
| "no_semantic": false, |
| "no_latent_noise": false, |
| "ce_acoustic_supervision": "target", |
| "freeze_components": null, |
| "freeze_llm_layers": null, |
| "train_llm_layers": null, |
| "freeze_head_layers": null, |
| "freeze_embeddings": false, |
| "freeze_lm_head": false, |
| "freeze_regex": null, |
| "bf16": true, |
| "gradient_checkpointing": true, |
| "no_spmd": false, |
| "logging_steps": 10, |
| "save_steps": 300, |
| "save_total_limit": 3, |
| "eval_steps": 0, |
| "eval_batches": 20, |
| "resume_from_checkpoint": null, |
| "dataloader_workers": 0, |
| "profile_first_steps": 3 |
| }, |
| "shapes": { |
| "max_seq_len": 512, |
| "max_latents": 224, |
| "max_segments": 8, |
| "acoustic_dim": 64, |
| "semantic_dim": 128 |
| }, |
| "preprocess_metadata": { |
| "model_name_or_path": "microsoft/VibeVoice-1.5B", |
| "processor_name_or_path": "vibevoice/processor", |
| "compress_ratio": 3200, |
| "acoustic_dim": 64, |
| "semantic_dim": 128, |
| "fix_std": 0.5, |
| "latent_dtype": "float16", |
| "speech_scaling_factor": 0.1962890625, |
| "speech_bias_factor": -0.04931640625, |
| "scaling_source": "checkpoint", |
| "corpus_latent_mean": -0.1338784531259102, |
| "corpus_latent_var": 23.753134242682215, |
| "shapes": { |
| "max_seq_len": 512, |
| "max_latents": 224, |
| "max_segments": 8, |
| "acoustic_dim": 64, |
| "semantic_dim": 128 |
| }, |
| "pad_token_id": 151655, |
| "augment_target_silence": true, |
| "normalize_target_audio": false |
| } |
| } |
| } |
