Add Optuna search + Hub checkpoints + CTranslate2 conversion 4359490 verified LogosAccessibleExpression commited on May 25
Upload train_hf_jobs.py with huggingface_hub 86827db verified LogosAccessibleExpression commited on May 24
Upload train_hf_jobs.py with huggingface_hub f7ca822 verified LogosAccessibleExpression commited on May 24
Upload train_hf_jobs.py with huggingface_hub 533890c verified LogosAccessibleExpression commited on May 24
Skip Optuna search, add hub checkpointing + faster-whisper CT2 conversion 3fe2cd1 verified LogosAccessibleExpression commited on May 24
Upload train_hf_jobs.py with huggingface_hub 47d44bc verified LogosAccessibleExpression commited on May 23
Upload train_hf_jobs.py with huggingface_hub a6f81e9 verified LogosAccessibleExpression commited on May 23
Upload train_hf_jobs.py with huggingface_hub 6f65de4 verified LogosAccessibleExpression commited on May 22
Upload train_hf_jobs.py with huggingface_hub a37d91b verified LogosAccessibleExpression commited on May 22
Upload train_hf_jobs.py with huggingface_hub 91bdba2 verified LogosAccessibleExpression commited on May 22
Upload train_hf_jobs.py with huggingface_hub a2963a2 verified LogosAccessibleExpression commited on May 22
Upload train_hf_jobs.py with huggingface_hub c026992 verified LogosAccessibleExpression commited on May 22
Upload train_hf_jobs.py with huggingface_hub 34451a3 verified LogosAccessibleExpression commited on May 22
Upload train_hf_jobs.py with huggingface_hub d08a4f0 verified LogosAccessibleExpression commited on May 22
Upload train_hf_jobs.py with huggingface_hub c1bbffa verified LogosAccessibleExpression commited on May 22
Upload train_hf_jobs.py with huggingface_hub fc871f3 verified LogosAccessibleExpression commited on May 21
Upload train_hf_jobs.py with huggingface_hub 489017d verified LogosAccessibleExpression commited on May 21
fix: LR range 1e-5 to 2e-4 matching Colab 8499c45 verified LogosAccessibleExpression commited on May 21
fix: fallback to random split when single-recording-per-phrase; guard empty val c2b9c34 verified LogosAccessibleExpression commited on May 21
improve accuracy: stratified split, SpecAugment search, TRIAL_STEPS=250, FINAL_STEPS=1500, train-sample eval 87d0c80 verified LogosAccessibleExpression commited on May 21
fix: disable_tqdm, patch_whisper logging, audio diagnostics, dedupe score_on_val 7da4848 verified LogosAccessibleExpression commited on May 21
Match Colab: load processor without language/task, use set_prefix_tokens ed162d8 verified LogosAccessibleExpression commited on May 21
Add language/task to generate(), restore TRIAL_STEPS=600 eacd576 verified LogosAccessibleExpression commited on May 21
Revert training label normalization — normalize eval only 126d32a verified LogosAccessibleExpression commited on May 21
Cap warmup_steps to [0, 25, 50] for 300-step trials 82a4d85 verified LogosAccessibleExpression commited on May 20
Normalize punctuation in eval and training labels using BasicTextNormalizer 3bd609c verified LogosAccessibleExpression commited on May 20
Reduce TRIAL_STEPS to 300 (keep higher LR floor 5e-5) 5df381e verified LogosAccessibleExpression commited on May 20
Fix: use keyword arg input_features= in model.generate() for PEFT model eval d831303 verified LogosAccessibleExpression commited on May 20
Upload train_hf_jobs.py with huggingface_hub 7c0f39a verified LogosAccessibleExpression commited on May 20
Upload train_hf_jobs.py with huggingface_hub cf0cad3 verified LogosAccessibleExpression commited on May 20
Upload train_hf_jobs.py with huggingface_hub 2dfc4db verified LogosAccessibleExpression commited on May 20
Upload train_hf_jobs.py with huggingface_hub 8327710 verified LogosAccessibleExpression commited on May 20
