v0.9.3

Updates

• Specify what number of samples you want in your train and test set using test_num_samples and train_num_samples in load_data module.
• All possible trainer parameters from the Seq2SeqTrainingArguments available when configuring training run.

What's Changed

• Update: Add test_num_samples and train_num_samples to load_data module. by @KevKibe in #116

Full Changelog: v0.9.2...v0.9.3

v0.9.2

Improvements for Inference Pipeline for Fully Finetuned Models

• Model Optimization: Uses CTranslate2 for optimized model performance.
• Backend: Leverages the faster-whisper backend (previously based on the Transformers Pipeline) for faster processing.
• Alignment: Uses Wav2vec models to ensure precise audio alignment.
• Multi-Speaker Diarization: Integrates pyannote-audio for speaker identification and labeling.
• Voice Activity Detection (VAD): Employs VAD preprocessing to reduce hallucination and improve batching without compromising Word Error Rate (WER).

Fixes

• Fixes #109

What's Changed

• Chore: Update deployment documentation, Fix: remove uid by @KevKibe in #93
• Chore: Add unittests for training.whisper_model_prep module. Modify training.load_dataset module by @KevKibe in #94
• Chore: Add tests for training.audio_data_processor module. by @KevKibe in #95
• Feat: Faster Whisper backend for deployment.transcription_pipeline module, transcription alignment, audio diarization, VAD. by @KevKibe in #98
• Chore: Update DOCS/DEPLOYMENT.MD and comments, add faster-whisper dependencies to src/deployments/requirements.txt.`. by @KevKibe in #100
• Bump rustls from 0.22.3 to 0.22.4 in /src/deployment/rustinference by @dependabot in #99
• Refactors: Change filenames for easier identification of peft code and non-peft. by @KevKibe in #102
• Fix: 'argparse` argument name, update version number for new release by @KevKibe in #107
• Fix: whisper_arch parameter in load_asr_model in convert_model_to_optimized_format method. by @KevKibe in #110
• Chore: Add tests for deployment.peft_speech_inference module, fix src/deployment/speech_inference_cli.py. by @KevKibe in #111
• Feat: Add num_samples parameter to training.load_data module to load a section of the training data. by @KevKibe in #112
• Fix: src/deployment/faster_whisper/mel_filters.npz file in build. by @KevKibe in #113

Full Changelog: v0.9.0...v0.10.0

v0.9.1

Release Highlights

This release updates the deployment module by separating inference for PEFT finetuned models and fully finetuned models.

For more information, refer to deployment.md

Improvements for Inference Pipeline for Fully Finetuned Models

• Model Optimization: Uses CTranslate2 for optimized model performance.
• Backend: Leverages the faster-whisper backend (previously based on the Transformers Pipeline) for faster processing.
• Alignment: Uses Wav2vec models to ensure precise audio alignment.
• Multi-Speaker Diarization: Integrates pyannote-audio for speaker identification and labeling.
• Voice Activity Detection (VAD): Employs VAD preprocessing to reduce hallucination and improve batching without compromising Word Error Rate (WER).

Full Changelog: v0.9.0...v0.9.1

v0.9.0

Release Highlights

• This update introduces Prometheus and Grafana monitoring to the inference API located in src/deployment/main dubbed speech_inference.
• speech_inference's function is to either translate or transcribe an audiofile.

What's Changed

• Add Prometheus and Grafana monitoring to inference API in src/deployment/main by @KevKibe in #90
• Update setup.cfg version number by @KevKibe in #91
• Delete datasource uid by @KevKibe in #92

Full Changelog: v0.8.0...v0.9.0

v0.8.0

Release Highlights

• wandb version update to 0.16.6 .
• uvicorn and fastapi update to build.
• Tests for training.load_data module.

What's Changed

• build(deps): bump wandb from 0.16.5 to 0.16.6 by @dependabot in #68
• Chore: Tests for training.load_data module. by @KevKibe in #86
• Update setup.cfg to include uvicorn and fastapi dependencies by @KevKibe in #87
• Fix: update workflows to run on pull request only by @KevKibe in #88

Full Changelog: v0.7.0...v0.8.0

v0.7.0

Bug Fix

• Fixes Pydantic bug in inference pipeline.

What's Changed

• Revert inference.py back to loading model without AutoModelForSpeechSeq2Seq and Au… by @KevKibe in #84
• Update package version to 0.7.0 by @KevKibe in #85

Full Changelog: v0.6.0...v0.7.0

v0.6.0

What's Changed

• fix: disable_progress_bar and disable_metadata parametersby @KevKibe in #83

Full Changelog: v0.5.0...v0.6.0

v0.5.0

Release Highlights:

• Introduction feature to train model on a dataset of more than one language.
• Fixes logging dependency issue.

What's Changed

• Feat: Train the model on different language dataset. by @KevKibe in #82

Full Changelog: v0.4.0...v0.5.0

v0.4.0

Release Highlights:

• Introduction of a logging feature for the Inference endpoint.
• Reduced Docker Image size for better containerization of API endpoint.
• Updated instructions in the README.

What's Changed

• Chore: Update README.md, model.config generation parameters, pytorch_model.bin file save to huggingface by @KevKibe in #73
• Chore: Update README.md, load multiple datasets feature in beta by @KevKibe in #74
• Chore: Add README.md Diagram by @KevKibe in #76
• add PARAMETERS.md, and update deployment.md , RUN specifications in Dockerfile by @KevKibe in #77
• Update README.md deployment.md link by @KevKibe in #78
• Feat: Add logging to inference endpoint src/deployment/main.py, fix model config and load_data module. by @KevKibe in #79
• update: add model save in pytorch_model.bin format by @KevKibe in #81

Full Changelog: v0.3.0...v0.4.0

v0.3.0

What's Changed

• Updates README.md Usage section by @KevKibe in #65
• Fix: Dependency version bug by @KevKibe in #66
• Update: Ignore warnings by @KevKibe in #67

Full Changelog: v0.2.9...v0.3.0