v1.2.1

Fixes

• OpenHosta now can't be used with Python 3.13 due to the incompatibility with pytorch

v2.0-beta1

GENERAL REFACTO

v1.2.0

v1.2.0 - 10/14/2024

New Features

• predict Function
  The predict function is now available, allowing you to create internal models—currently supporting linear regression—based on user-provided training data. This simplifies model generation without relying on external APIs. Key functionalities include:

  • .retrain: Retrain models with specified parameters.
  • .continue_train: Continue training with existing model weights.
  • .emulate: Run predictions through an LLM or create a model directly using internal linear regression based on training data.

• TrainingSet Management
  Manage training datasets effortlessly with new tools:

  • .visualize: Inspect current data visually.
  • .add: Add new examples.

Enhancements

• Expanded Dataset Support:
  load_training_example (previously load_examples) supports JSON, JSONL, and CSV formats for easier integration.

• Verbose Mode in predict:
  Track detailed model training and define target losses with get_loss.

v1.1.2

• Fixes
  • Remove debug prints

v1.2-rc1

v1.2rc-1 10/10/2024

New Features

1. TrainingSet Management
   Manage training datasets effortlessly with new tools:

   • .visualize: Inspect current data visually.
   • .add: Add new examples.

2. Enhanced predict Attributes
   New functionalities for predict:

   • .retrain: Retrain models with specified parameters.
   • .continue_train: Continue training with existing weights.
   • .emulate: Run predictions through an LLM.

Enhancements

• Expanded Dataset Support: load_training_example (previously load_examples) supports JSON, JSONL, and CSV formats for easier integration.

• Verbose Mode in predict: Track detailed model training and set target losses with get_loss.

Fixes

• CUDA Compatibility: predict now works with CUDA-enabled GPUs (device ID selection pending).

v1.1.1

OpenHosta v1.1.1 - 10/07/24

This patch focuses on performance enhancements and bug fixes related to prompt handling and output typing.

Features

• Introduced the _last_request attribute to the Model object for enhanced tracking.

Optimization

• Streamlined the emulate user prompt by limiting it to the function call string and moving additional information to the system prompt.
• Enhanced the structure of the emulate prompt with markdown headers.
• Eliminated unnecessary sentences such as confidence levels and transitions for a more concise output.

Internal

• Refactored the prompt-building function within emulate for improved maintainability.
• Removed the validate function in the LLM call response handler to simplify the codebase.

Fixes

• Added code block syntax for clearer function definitions.
• Re-added the diagramm attribute (now deprecated) for backward compatibility.
• Explicitly included a neutral response (None) in the emulate prompt to handle edge cases.

Instructions for Update

To update to this version, use the following command:

pip install --upgrade OpenHosta

Remember to delete __hostacache__ folders to avoid version conflicts.

v1.2-beta4

v1.2beta-4

Features

• TrainingSet
  The TrainingSet feature is now available! It allows you to manage datasets for hosta_injected functions with the following tools:

  • .visualize: Visualize the dataset (examples and data) for a specific function and get the list of them.
  • .add: Add new data into ho_data.
  • .generate: Create data by calling an LLM.

• New Attributes in Predict Functions
  The Predict function now includes these new attributes:

  • .retrain: Retrain the model.
  • .continue_train: Continue training using the previous weights.
  • .emulate: Run the prediction function in emulation mode with a model.

Enhancements

• load_examples has been renamed to load_training_example and now supports datasets in JSON, JSONL, and CSV formats for storage in hosta_cache.
• The Predict function now supports a verbose mode for clearer model operations and includes a get_loss attribute to specify a loss to obtain !

Full Changelog: v1.2-beta3...v1.2-beta4

v1.2-beta3

v1.2beta-3

• Fixes

  • predict is now CUDA compatible (currently, CUDA(id) cannot be chosen).

• Enhancements

  • mergenow merge with 1.1

What's Changed

• Merge v1.1 to v1.2beta-2 by @ramosleandre in #126
• Update Cuda compatibility by @ramosleandre

v1.1.0

OpenHosta v1.1.0 - 09/27/2024

This release introduces significant new features, enhancements, optimizations, and bug fixes to improve the functionality and performance of OpenHosta.

New Features

• example Function: Adds examples for a specified hosta-injected function and caches them.
• save_examples Function: Saves all examples of a hosta-injected function in a JSONL file.
• load_examples Function: Loads an example file into a cache for a hosta-injected function.
• set_prompt in PromptManager: Automatically changes a prompt in "prompt.json".
• suggest function: Works the same as the __suggest__ attributs but in a function

Changes

• suggest attribute: Fixed diagramm to diagram.

Enhancements

• thought Function: Automatically guesses its return type with LLM.
• suggest Function: Returns data from LLM in a dictionary.
• emulate Function:
  • Now works for class methods.
  • Integrates local variables of the emulated function into the LLM prompt.
  • Includes a verification output to check and convert the output to the specified type if necessary. Supported types include Pydantic models, types from the typing module (List, Dict, Tuple, etc.), and built-in types (dict, int, float, etc.).
  • Added _last_request attribute to access the prompt sent to the LLM and _last_response to acces the raw message received.
• Support for Typing Module: Allows specific return types from the typing module, including List, Dict, Tuple, Set, FrozenSet, Deque, Iterable, Sequence, Mapping, Union, Optional, Literal.
• Return Type Prediction: The return type predicted by thought is now attached as an attribute (_return_type).

Optimizations

• emulate Prompt: Removed confidence level, resulting in ~20% speed improvement.
• Enhanced Prompts: Added code blocks, separators, and references to the processed function name for emulate, enhance, thought, and estimate.

Bug Fixes

• emulate Function:
  • Now decorator-resistant and works in nested functions.
  • now works when emulated function is called inside another one
  • works now with lru_cache decorator from functools module
• config.set_default_model: Works correctly with config.set_default_apiKey.
• thought Function: Now accepts multiple arguments.
• suggest and analytics Functions: Now call LLM with Model class (ai_call).
• Flask Compatibility: Added compatibility with Flask.
• Added a loop to find the frame in _extend_scope

Internal Improvements

• Functional Tests: Added for each library feature.
• Custom Exception Classes: Added for request and frame errors.
• Makefile: Added for cleaning, packaging, and testing.
• GitHub Workflows: Added for linting, formatting, and testing on push to dev and main branches.

Documentation

• Google Cobal Link: Integrated into the documentation.
• Inconsistencies and Errors: Corrected various documentation issues.

Performance Metrics

• emulate Function: Execution time decreased by ~20% despite a slight performance drop since version v1.0.2.
• thought Function: Execution time increased significantly due to the type prediction system, resulting in a double LLM call.
• Caching System: No impact on the execution time of OpenHosta functions.

Instructions for Update

To update to this version, use the following command:

pip install --upgrade OpenHosta

Remember to delete __hostacache__ folders to avoid version conflicts.

Acknowledgements

Thank you to all contributors for your valuable feedback and contributions: @ramosleandre, @MerlinDEVILLARD, @WilliamJlvt, and @battmanux.

---

v1.1-rc4

v1.1-rc4

• Feature

  • Added suggest function. Works the same as the __suggest__ attributs but in a function

• Doc

  • Many inconsistencies and errors corrected.

• Internal

  • Added Github workflows for linting, formating and testing when pushing to dev and main

• Fixes

  • suggest attribute diagramm is now diagram

What's Changed

• Fix/#111 @Merlin devillard by @MerlinDEVILLARD in #112
• Fix/#110 @Merlin devillard by @MerlinDEVILLARD in #118
• Dev by @MerlinDEVILLARD in #119

Full Changelog: v1.1-rc3...v1.1-rc4