Training in progress, step 1000

config.json

@@ -0,0 +1,47 @@
+{
+  "activation_dropout": 0.0,
+  "activation_function": "gelu",
+  "apply_spec_augment": false,
+  "architectures": [
+    "WhisperForConditionalGeneration"
+  ],
+  "attention_dropout": 0.0,
+  "begin_suppress_tokens": null,
+  "bos_token_id": 50257,
+  "classifier_proj_size": 256,
+  "d_model": 1280,
+  "decoder_attention_heads": 20,
+  "decoder_ffn_dim": 5120,
+  "decoder_layerdrop": 0.0,
+  "decoder_layers": 32,
+  "decoder_start_token_id": 50258,
+  "dropout": 0.0,
+  "encoder_attention_heads": 20,
+  "encoder_ffn_dim": 5120,
+  "encoder_layerdrop": 0.0,
+  "encoder_layers": 32,
+  "eos_token_id": 50257,
+  "forced_decoder_ids": null,
+  "init_std": 0.02,
+  "is_encoder_decoder": true,
+  "mask_feature_length": 10,
+  "mask_feature_min_masks": 0,
+  "mask_feature_prob": 0.0,
+  "mask_time_length": 10,
+  "mask_time_min_masks": 2,
+  "mask_time_prob": 0.05,
+  "max_length": null,
+  "max_source_positions": 1500,
+  "max_target_positions": 448,
+  "median_filter_width": 7,
+  "model_type": "whisper",
+  "num_hidden_layers": 32,
+  "num_mel_bins": 128,
+  "pad_token_id": 50256,
+  "scale_embedding": false,
+  "torch_dtype": "float32",
+  "transformers_version": "4.53.1",
+  "use_cache": true,
+  "use_weighted_layer_sum": false,
+  "vocab_size": 51866
+}

experiment_config.yaml

@@ -0,0 +1,5 @@
+dataset_subset: audio_500h
+experiment_name: whisper-large-v3-kin-500h
+seed: 42
+use_mlflow: true
+use_wandb: false

full_config.yaml

@@ -0,0 +1,78 @@
+num_workers: 4
+pretrained_model: openai/whisper-large-v3
+train:
+  download_datasets_in_parallel: false
+  huggingface_load:
+  - name: audio_500h
+    path: evie-8/kinyarwanda-speech-hackathon
+    split: train
+  shuffle: false
+  source:
+    language:
+    - kin
+    preprocessing:
+    - set_sample_rate:
+        p: 0.05
+        rate: 8000
+    - set_sample_rate:
+        rate: 16000
+    - normalize_audio
+    - augment_audio_speed:
+        high: 1.15
+        low: 0.95
+        p: 0.2
+    - augment_audio_noise:
+        max_relative_amplitude: 0.3
+    type: speech
+  target:
+    language:
+    - kin
+    preprocessing:
+    - lower_case
+    - clean_and_remove_punctuation:
+        allowed_punctuation: ''''
+    type: text
+training_args:
+  eval_steps: 200
+  eval_strategy: steps
+  fp16: true
+  generation_max_length: 200
+  gradient_accumulation_steps: 2
+  gradient_checkpointing: true
+  gradient_checkpointing_kwargs:
+    use_reentrant: false
+  greater_is_better: false
+  hub_model_id: akera/whisper-large-v3-kin-500h_20250712_203553
+  learning_rate: 1.0e-05
+  load_best_model_at_end: true
+  logging_steps: 200
+  max_steps: 10000
+  metric_for_best_model: loss
+  output_dir: whisper-large-v3-kin-500h_20250712_203553
+  per_device_eval_batch_size: 16
+  per_device_train_batch_size: 16
+  predict_with_generate: true
+  push_to_hub: true
+  save_steps: 1000
+  save_total_limit: 2
+  warmup_steps: 100
+use_peft: false
+validation:
+  huggingface_load:
+  - path: jq/kinyarwanda-speech-hackathon
+    split: dev_test[:200]
+  source:
+    language:
+    - kin
+    preprocessing:
+    - set_sample_rate:
+        rate: 16000
+    type: speech
+  target:
+    language:
+    - kin
+    preprocessing:
+    - lower_case
+    - clean_and_remove_punctuation:
+        allowed_punctuation: ''''
+    type: text

model-00001-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1b364747c08659b57b30d9b4623b96e90f0f78efffdfc0990435a496ec4be8b5
+size 4993448880

model-00002-of-00002.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:71f580e2fa8769d3a93815eca05460e8422d4cbb917089f49455a97fa02abc58
+size 1180663192

preprocessor_config.json

@@ -0,0 +1,15 @@
+{
+  "chunk_length": 30,
+  "dither": 0.0,
+  "feature_extractor_type": "WhisperFeatureExtractor",
+  "feature_size": 128,
+  "hop_length": 160,
+  "n_fft": 400,
+  "n_samples": 480000,
+  "nb_max_frames": 3000,
+  "padding_side": "right",
+  "padding_value": 0.0,
+  "processor_class": "WhisperProcessor",
+  "return_attention_mask": false,
+  "sampling_rate": 16000
+}

training_args.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9ec08fcfe2489a62c47499f1dc0043e2aa4b9bd3ed2bc6ace82ac681c8311667
+size 6033