Add trained LoRA adapter (main), tokenizer/processor, lineage, standalone inference and model card
36ba721 verified | { | |
| "run_name": "main", | |
| "created_utc": "2026-11-08T18:44:24Z", | |
| "base_model": { | |
| "path": "models/whisper-large-v3", | |
| "repo_id": "openai/whisper-large-v3", | |
| "commit_sha": "06f233fe06e710322aca913c1bc4249a0d71fce1", | |
| "license": "apache-2.0", | |
| "weights_sha256": { | |
| "model.safetensors": "a8e94b85976e5864ba3e9525c7e6c83b2a1eca42d4b797a0c7c24d778e40fd95" | |
| } | |
| }, | |
| "manifests": { | |
| "train": { | |
| "path": "whisper_ft/manifests/train.jsonl", | |
| "sha256": "793e412837b58ce9604a380b23090107de9b87244ffae55823bf9b6d220c838f", | |
| "rows": 24497 | |
| }, | |
| "val": { | |
| "path": "whisper_ft/manifests/val.jsonl", | |
| "sha256": "44c8ad4bc2691e9a4428f85b8ae8c8a1918c38e6439e2d87d4c82159438572e3", | |
| "rows": 8217 | |
| }, | |
| "test": { | |
| "path": "whisper_ft/manifests/test.jsonl", | |
| "sha256": "e5510f27fe6995133f29bca7fc0285fbf34276bb7225f13839c874de744e48aa", | |
| "rows": 8216 | |
| } | |
| }, | |
| "train_rows_used": 24497, | |
| "train_rows_per_dataset": { | |
| "anomaly": 20000, | |
| "engine": 4497 | |
| }, | |
| "eval_subset_rows": 600, | |
| "eval_subset_ids_sha256": "16e9f9af5d09f54b3209ddb8ba98757e610ec3206797849144470c573345aaa7", | |
| "dataset_licenses": { | |
| "anomaly": { | |
| "hf_id": "mandipgoswami/AnomalyMachine-50K", | |
| "license": "cc-by-4.0" | |
| }, | |
| "engine": { | |
| "hf_id": "huggingbear12/procedural-engine-sounds", | |
| "license": "cc-by-nc-4.0", | |
| "note": "NON-COMMERCIAL: models trained on this data inherit the NC restriction" | |
| } | |
| }, | |
| "non_commercial": true, | |
| "config": { | |
| "model": { | |
| "base_path": "../models/whisper-large-v3", | |
| "language": "en", | |
| "task": "transcribe", | |
| "quantization": "8bit", | |
| "attn_implementation": "sdpa", | |
| "apply_spec_augment": false, | |
| "finetuned_decoding": true | |
| }, | |
| "lora": { | |
| "r": 32, | |
| "alpha": 64, | |
| "dropout": 0.05, | |
| "target_modules": [ | |
| "q_proj", | |
| "k_proj", | |
| "v_proj", | |
| "out_proj", | |
| "fc1", | |
| "fc2" | |
| ], | |
| "scope": "all" | |
| }, | |
| "data": { | |
| "train_manifest": "manifests/train.jsonl", | |
| "val_manifest": "manifests/val.jsonl", | |
| "test_manifest": "manifests/test.jsonl", | |
| "max_label_tokens": 128, | |
| "eval_samples": 600, | |
| "seed": 42 | |
| }, | |
| "training": { | |
| "output_root": "out/whisper-lora", | |
| "per_device_train_batch_size": 4, | |
| "per_device_eval_batch_size": 8, | |
| "gradient_accumulation_steps": 8, | |
| "learning_rate": 0.0001, | |
| "warmup_ratio": 0.05, | |
| "lr_scheduler_type": "cosine", | |
| "num_train_epochs": 3, | |
| "weight_decay": 0.0, | |
| "bf16": true, | |
| "gradient_checkpointing": true, | |
| "optim": "adamw_bnb_8bit", | |
| "eval_steps": 500, | |
| "save_steps": 500, | |
| "save_total_limit": 3, | |
| "predict_with_generate": true, | |
| "generation_max_length": 128, | |
| "metric_for_best_model": "wer", | |
| "greater_is_better": false, | |
| "load_best_model_at_end": true, | |
| "remove_unused_columns": false, | |
| "dataloader_num_workers": 6, | |
| "logging_steps": 25, | |
| "report_to": "none", | |
| "seed": 42 | |
| }, | |
| "fallbacks": { | |
| "batch2": { | |
| "training.per_device_train_batch_size": 2, | |
| "training.gradient_accumulation_steps": 16 | |
| }, | |
| "nf4": { | |
| "model.quantization": "4bit" | |
| }, | |
| "decoder_only": { | |
| "lora.scope": "decoder" | |
| }, | |
| "turbo": { | |
| "model.base_path": "../models/whisper-large-v3-turbo" | |
| } | |
| } | |
| }, | |
| "cli": [ | |
| "--run-name", | |
| "main" | |
| ], | |
| "resumed_from": null, | |
| "finished_utc": "2026-11-09T03:40:52Z", | |
| "global_steps": 2298, | |
| "best_checkpoint": "whisper_ft/out/whisper-lora/main/checkpoints/checkpoint-2298", | |
| "best_metric": 0.18471849865951742, | |
| "train_metrics": { | |
| "train_runtime": 31899.7704, | |
| "train_samples_per_second": 2.304, | |
| "train_steps_per_second": 0.072, | |
| "total_flos": 2.5945070514536448e+20, | |
| "train_loss": 0.21723902401870182, | |
| "epoch": 3.0, | |
| "peak_vram_reserved_gb": 10.36, | |
| "sec_per_step": 13.05 | |
| }, | |
| "final_eval": { | |
| "eval_loss": 0.2756240665912628, | |
| "eval_wer": 0.18471849865951742, | |
| "eval_cer": 0.0795668549905838, | |
| "eval_exact": 0.22166666666666668, | |
| "eval_field_acc_mean": 0.5955555555555555, | |
| "eval_anomaly_f1": 1.0, | |
| "eval_wer_anomaly": 0.08876404494382023, | |
| "eval_wer_engine": 0.2723076923076923, | |
| "eval_runtime": 286.9148, | |
| "eval_samples_per_second": 2.091, | |
| "eval_steps_per_second": 0.261, | |
| "epoch": 3.0, | |
| "peak_vram_reserved_gb": 10.36, | |
| "sec_per_step": 13.05 | |
| }, | |
| "perf": { | |
| "optimizer_steps_timed": 2296, | |
| "sec_per_optimizer_step_mean": 13.259, | |
| "sec_per_optimizer_step_median": 13.276, | |
| "peak_vram_allocated_gb": 9.23, | |
| "peak_vram_reserved_gb": 10.36, | |
| "peak_device_used_gb": 10.76 | |
| }, | |
| "train_wall_seconds": 31900.2, | |
| "environment": { | |
| "torch": "2.15.0", | |
| "transformers": "5.22.2", | |
| "peft": "0.21.2", | |
| "bitsandbytes": "0.50.2", | |
| "accelerate": "1.15.0", | |
| "gpu": "NVIDIA GeForce RTX 5070", | |
| "python": "3.11.18" | |
| } | |
| } |