🚀 New Features

Token Expiration Handling Improvements

• Switched from token reissue to fresh login when a token expires.
• Ensures more reliable and predictable authentication flow.

Automatic Token Update Post-Login

• After a successful login, self.tokens is now automatically updated to prevent stale token usage in subsequent requests.

UploadDataset Dataclass

• Added a new UploadDataset dataclass for clearer and more structured handling of dataset upload metadata.

Launcher Task Error Logging

• Improved error logging for task failures in the launcher, helping users debug configuration and runtime issues more easily.

API Client Info Logging

• API client now prints host and port information upon initialization to confirm connection settings.

🐞 Bug Fixes

• Fixed a bug where trainer model architecture configuration was incorrectly parsed.
• Resolved an issue with classification dataset settings not applying correctly in training.
• Added missing model_name definition in initialize_from_yaml() to avoid runtime errors during model setup.

🧠 Why these matter

• These updates improve reliability in authentication and API interaction, particularly in long-running sessions.
• Users benefit from better visibility into configuration issues and API usage context.
• Dataset and training configuration flows are now more stable and transparent, reducing the chance of runtime errors.

🚀 New Features

Structured Neuron-level Pruning (SNP)

• Added a new structured pruning method targeting neuron-level granularity.
• Allows more fine-tuned control over model size and latency during optimization.

Upload Progress Bar

• Introduced a progress bar display when uploading large models or datasets via the SDK.
• Helps users track upload status more clearly and improve user experience.

Model Name Standardization & Deprecation Notices

• Unified naming convention for preloaded model names.
• Added deprecation warnings to guide users toward supported configurations.

Environment Variable Configuration

• Refactored internal configuration to support HOST and PORT via environment variables.
• Simplifies deployment in containerized or cloud environments.

🐞 Bug Fixes

• Updated file-matching logic to support both *best.pt and *best_fx.pt patterns during model search.
• Fixed attribute reference from model.name to model_name to prevent runtime errors.

🧠 Why these matter

• The addition of Structured Neuron-level Pruning enables more granular and efficient model optimization, especially for edge deployment scenarios.
• Visual upload progress improves transparency and confidence during long upload operations.
• Unified naming and configuration approaches reduce user confusion and improve maintainability.
• Bug fixes ensure more reliable model handling and deployment flow.

🚀 New Features

Removal of Legacy v1 Module

• Eliminated the outdated v1 module to streamline the codebase and reduce maintenance overhead.

Simplification of Policy Options

• Removed the backward option from policy configurations, simplifying the policy setup process for users.

Elimination of Pydantic Dependency

• Removed the pydantic library from the project, reducing external dependencies and potential compatibility issues.

🐞 Bug Fixes

• Fixed an issue where identical comments were logged multiple times across different lines, improving log clarity.
• Resolved a problem with access token reissuance, enhancing authentication reliability.
• Modified the request body construction from using model_dump() to asdict(), ensuring compatibility with updated data models.
• Corrected compressor option configurations to prevent misconfigurations during model compression.

🧠 Why these matter

• Removing deprecated modules and simplifying configurations enhances the maintainability and usability of the SDK.
• Reducing external dependencies minimizes potential conflicts and streamlines the installation process.
• Addressing these bugs improves the overall stability and reliability of the platform, leading to a better user experience.

🚀 New Features

Introduction of v2 Clients and Modules

• Developed and integrated version 2 (v2) clients and modules, laying the groundwork for enhanced modularity and scalability in future developments.

netspresso_trainer Enhancements

• Updated netspresso_trainer to version 0.2.2, incorporating the latest features and improvements.
• Implemented lazy import for netspresso_trainer, optimizing resource usage and startup time.
• Added dataclass support for trainer configuration, simplifying the setup and management of training parameters.

🐞 Bug Fixes

• Adjusted the download_model function to ensure the downloaded model files have the correct .pt extension, resolving compatibility issues during model loading.

🧠 Why these matter

• The introduction of v2 clients and modules enhances the framework's flexibility, allowing for more efficient development and integration of new features.
• Updates to netspresso_trainer improve the user experience by providing more intuitive configuration options and reducing unnecessary resource consumption.
• Fixing the model download path extension ensures smoother workflows and prevents errors related to incorrect file formats.

🚀 New Features

AutoML Training with Bayesian & Hyperband

• Introduced automated machine learning training capabilities utilizing Bayesian optimization and Hyperband algorithms.
• Enables efficient hyperparameter tuning to achieve optimal model performance with minimal manual intervention.

New SDK Functions for Model Optimization

• Added get_prune_specs() and prune() functions to facilitate structured model pruning.
• Introduced get_retrain_specs() and retrain() functions to streamline the retraining process of pruned models.
• Implemented get_trt_engine_specs() and gen_trt_engine() functions for generating optimized TensorRT engines.
• Provided get_inference_specs() and inference() functions to simplify the inference process on optimized models.

Device Support Expansion

• Added support for the ARDUINO_NICLA_VISION device, enabling model deployment and benchmarking on this platform.

🐞 Bug Fixes

• Docker Installation Guide Update: Resolved issues in the INSTALLATION.md related to Docker usage:
  • Included the bash command in the Docker run instructions to ensure proper container initialization.
  • Added a step to install the package in editable mode (pip install -e .) to prevent ModuleNotFoundError for the netspresso module.

🧠 Why these matter

• The integration of AutoML capabilities significantly reduces the time and expertise required for model optimization, making advanced techniques more accessible.
• The new SDK functions provide a more modular and user-friendly approach to model pruning, retraining, engine generation, and inference, enhancing the overall developer experience.
• Expanding device support to include ARDUINO_NICLA_VISION broadens the deployment possibilities for users targeting edge devices.
• Updating the Docker installation guide ensures a smoother setup process, minimizing potential setup errors and improving usability.

🚀 New Features

🆕 NXP i.MX93 Device Support

• Added support for the NXP i.MX93 device in the converter and benchmarker modules.
• Users can now convert and benchmark models tailored for the NXP i.MX93 platform.

🧠 Why these matter

• Expanding device support to include NXP i.MX93 enables users to deploy optimized AI models on a broader range of hardware platforms.
• This enhancement facilitates seamless integration and performance evaluation on NXP's edge computing devices.

🚀 New Features

Training Result Visualization

• Added functions to plot training metrics and loss curves, enabling users to monitor model performance over time.

Profile Result Plotting

• Introduced plots for profiling results by step size and compression ratio, assisting in analyzing compression effects.

Unique Folder Creation

• Implemented automatic creation of unique output directories when downloading artifacts, preventing file overwrites.

Artifact Download Functionality

• Added the ability to download artifacts (e.g., trained models, logs) directly through the SDK.

Job Status Monitoring

• Users can now monitor the status of training, evaluation, and export jobs, providing better workflow management.

Pruning Configuration Enhancements

• Introduced options for step size, step operation, and reverse pruning, offering more control over the pruning process.

🐞 Bug Fixes

• Resolved issues related to the scope of Y-axis limits in plots, ensuring accurate data visualization.

• Fixed problems with output directory creation when downloading artifacts, enhancing file management reliability.

• Addressed inconsistencies in default step size values for pruning configurations.

🧠 Why these matter

• These updates enhance the user experience by providing better visualization tools and more control over model optimization processes.

• Improved artifact management and job monitoring streamline the workflow, making model development more efficient.

• The bug fixes ensure that the new features operate smoothly, contributing to a more stable and reliable platform.

🚀 New Features

Token Expiration Buffer

• Added a 60-second buffer to JWT token expiration logic to prevent unexpected session terminations during long-running processes.

Automatic Project Folder Creation

• The SDK now automatically creates project folders if they do not exist, ensuring a cleaner and more consistent file structure during runtime.

🐞 Bug Fixes

• Fixed issue where models using ONNX opset 17 failed to export correctly.
• Resolved misalignment between Studio and SDK behavior when running conversion tasks.
• Improved CLI error handling for invalid device names during export.

🧠 Why these matter

• These updates improve runtime stability and usability across long-running sessions.
• Users no longer need to manually create output folders or monitor token expiry in real time.
• Overall file handling and export flows are now more robust and consistent.

🚀 New Features

Updated Jupyter Notebooks

• Core examples have been updated to reflect the latest APIs and module usage.
• Cleaner structure, improved comments, and better alignment with actual workflows.

Google Colab Support

• Notebooks can now be run directly in Google Colab with one-click launch buttons.
• No need for local setup—just open and run in the cloud.

Improved Installation Guide

• New instructions added for Docker-based installation and isolated environment setup.
• Additional explanation of manual compression and quantization processes.

API Documentation Enhancements

• API reference updated with clearer function descriptions and usage examples.
• Aligns with recent structural changes in PyNetsPresso modules.

🐞 Bug Fixes

• Fixed folder creation issue when using custom paths in export pipelines.
• Improved error handling for invalid file inputs and missing arguments.
• Increased stability when running notebooks in memory-constrained environments like Google Colab.

🧠 Why these matter

• Makes it significantly easier for new users to get started with minimal setup.
• Reduces onboarding friction through improved documentation and interactive examples.
• Helps users catch environment and path-related issues early in the workflow.

🚀 New Features

Verification Option for Model Conversion

• You can now enable an optional verify argument when calling convert_model(), which runs automatic checks to validate that the converted model behaves correctly.

🐞 Bug Fixes

• Fixed a bug where incorrect verification results were not being surfaced properly after conversion.
• Improved error messaging when using invalid combinations of device/framework/software version in convert_model().

🧠 Why these matter

• This update improves model conversion reliability and helps users detect issues early in the deployment pipeline.
• It reduces time spent debugging export errors and gives users more confidence in their final output.