Skip to content

CHANGELOG.md

Latest commit

 

History

History
142 lines (120 loc) · 6.73 KB

CHANGELOG.md

File metadata and controls

142 lines (120 loc) · 6.73 KB

Changelog

All notable changes to this project will be documented in this file.

The format is based on Keep a Changelog, and this project adheres to Semantic Versioning.

[Unreleased]

Nothing yet.

[0.5.1] - 2025-01-11

Removed

  • Removed redundant create_structured_completion() function from core.client module
    • Function was unused in the codebase; structured outputs are implemented directly in judge.py
    • Cleaned up unused imports and test cases

[0.5.0] - 2025-01-11

Added

  • Hugging Face Hub integration for direct dataset uploads
    • push_to_hub() function in new hf_hub module to upload datasets to HF Hub
    • Uploads JSONL files (train.jsonl, val.jsonl), manifest.json, and auto-generated README.md
    • CLI flags: --push-to-hub, --repo-id, --hf-token, --private
    • Support for both public and private repositories
    • Auto-generated dataset cards with dataset statistics, model info, usage examples, and citation
  • Optional dependency: huggingface_hub>=0.20.0 (install with pip install toolsgen[hf])
  • Example in examples/hf_hub_upload/ with dotenv configuration
  • Test suite for HF Hub functionality in tests/test_hf_hub.py
  • push_to_hub exported from main toolsgen package for easier imports

[0.4.0] - 2025-01-10

Added

  • Quality tagging system for generated records
    • generate_quality_tags() method in JudgeResponse to automatically tag samples based on judge scores
    • Tags include overall quality levels (high/medium/low_quality) and dimension-specific tags (excellent/poor tool selection, arguments, clarity)
    • Configurable thresholds for quality classification
    • quality_tags field automatically populated in generated records

[0.3.0] - 2025-01-10

Added

  • Hugging Face dataset integration utilities in examples/nano_tool_calling_v1/
    • dataset_to_tools() function to load tools from Hugging Face datasets
    • validate_json_schema() for OpenAI tool schema validation with recursive array type checking
    • push_to_hf.py script for uploading generated datasets to Hugging Face Hub
  • Complete example workflow for Nano Tool Calling v1 dataset generation
    • Configuration, generation, validation, and publishing pipeline
    • Analysis utilities for function inspection
    • Comprehensive README with dataset card format

Changed

  • Enhanced batch sampling progress bar display for better user feedback
  • Improved parallel processing record ordering and ID assignment

[0.2.2] - 2025-01-09

Changed

  • Records are now written to JSONL file immediately as they complete in parallel mode, rather than waiting for all generation to finish
  • Improved memory efficiency by removing records from buffer after writing to disk

[0.2.1] - 2025-01-09

Fixed

  • Fixed integration tests to work with refactored module structure

[0.2.0] - 2025-01-09

Added

  • Parallel generation support with multiprocessing via --workers and --worker-batch-size CLI flags
  • num_workers and worker_batch_size configuration options in GenerationConfig
  • Parallel generation example in examples/parallel/

Fixed

  • Fixed tool subset diversity preservation in parallel mode by sorting records by original sample index before assigning final IDs

[0.1.4] - 2025-11-09

Changed

  • Made max_tokens optional across all chat completion helpers and dataset flows so callers can rely on model defaults unless a limit is explicitly set.

[0.1.3] - 2025-11-08

Added

  • Batching controls (batch_size, shuffle_tools) in GenerationConfig, CLI flags, and docs to opt into chunked sampling.
  • Deterministic chunk-based sampling path that reuses batches in a wrap-around manner when generating many subsets.

Changed

  • CLI now forwards batching parameters so dataset generation can reuse the refactored sampling logic end-to-end.

[0.1.2] - 2025-11-08

Fixed

  • Restored toolsgen version output by sourcing __version__ from package metadata when running the CLI

[0.1.1] - 2025-11-08

Added

  • Official support declarations for Python 3.12, 3.13, and 3.14 in project metadata

[0.1.0] - 2025-11-08

Added

  • Project skeleton with package, CLI, tests, and basic schemas
  • Core dataset generation functionality
    • LLM-based user request generation
    • Tool call generation with OpenAI-compatible APIs
    • Record creation and serialization
  • Complete CLI implementation with model configuration options
  • Comprehensive test suite for schemas, sampling, config, and generator
  • Public API exports in __init__.py
  • Support for param_aware and semantic sampling strategies
  • LLM-as-a-judge scoring system
    • Rubric-based evaluation (tool relevance, argument quality, clarity)
    • Structured outputs using pydantic models for reliable parsing
  • JSONL output format with train/val split support
  • Prompts module with centralized prompt templates
  • Semantic sampling strategy based on keyword similarity

Changed

  • [BREAKING] generate_dataset() function signature updated:
    • New signature: generate_dataset(output_dir, gen_config, model_config, tools_path=None, tools=None)
    • Old signature: generate_dataset(tools_path, output_dir, gen_config, model_config)
    • Now supports passing tools list directly via tools parameter as alternative to tools_path
    • Exactly one of tools_path or tools must be provided
  • [BREAKING] Reduced dependencies (removed typer, python-dotenv from core dependencies)
  • [BREAKING] Flattened module structure - removed nested folders (io/, judge/, providers/)
  • [BREAKING] CLI rewritten using argparse instead of typer (stdlib only)
  • [BREAKING] Module reorganization:
    • generator.py, config.py, io/writer.py merged into core.py
    • judge/scorer.py moved to judge.py
    • providers/openai_compat.py removed - using OpenAI SDK directly
  • [BREAKING] Import paths updated:
    • from toolsgen.config → from toolsgen.core
    • from toolsgen.generator → from toolsgen.core
    • from toolsgen.judge.scorer → from toolsgen.judge
  • Simplified semantic sampling algorithm (reduced complexity)
  • Environment variable loading: direct os.environ instead of python-dotenv for core functionality
  • Enhanced error handling and user feedback in CLI
  • Output format uses JSONL for datasets
  • Judge system uses OpenAI SDK structured outputs directly
  • Prompts extracted from inline strings to dedicated module
  • Updated README with simplified installation and new import paths

Removed

  • Dependency on typer (replaced with stdlib argparse)
  • Dependency on python-dotenv from core (moved to dev dependencies, used optionally in examples)
  • OpenAI wrapper abstraction layer (use SDK directly)
  • Nested folder structure (io/, judge/, providers/)

Fixed

  • Missing dependencies declaration in pyproject.toml
  • Reduced code complexity and improved maintainability