A powerful Python library for enhancing LLM capabilities through advanced test-time compute and search algorithms, focusing on expert human-AI co-construction. This project is under consturction. Don't come with any expectations!
-
Test-Time Compute Algorithms
- Tree of Thoughts implementation
- Self-consistency sampling
- Dynamic prompt optimization
- Multi-perspective reasoning
-
Search Algorithms
- Beam search with adaptive width
- Best-first symbolic search
- Guided exploration with expert feedback
- Knowledge graph-based reasoning
-
Expert Human-AI Co-Construction
- Interactive reasoning frameworks
- Expert feedback integration
- Knowledge distillation from expert interactions
- Adaptive learning from human guidance
pip install superllmfrom superllm import ThoughtTree, ExpertFeedback
from superllm.search import AdaptiveBeamSearch
# Initialize a thought tree for complex reasoning
thought_tree = ThoughtTree()
# Create an expert feedback system
expert_system = ExpertFeedback()
# Configure adaptive beam search
search = AdaptiveBeamSearch(
beam_width=5,
max_steps=10
)
# Run enhanced inference
result = thought_tree.solve(
prompt="Your complex question here",
search_algorithm=search,
expert_system=expert_system
)For detailed documentation, visit docs.superllm.ai
- Thought Trees: Implement tree-based reasoning strategies for complex problem-solving
- Search Algorithms: Advanced search techniques for exploring solution spaces
- Expert Systems: Tools for integrating human expert knowledge and feedback
- Evaluation Metrics: Comprehensive metrics for assessing solution quality
We welcome contributions! Please check our Contributing Guidelines for details.
This project is licensed under the MIT License - see the LICENSE file for details.
If you use SuperLLM in your research, please cite:
@software{superllm2024,
title = {SuperLLM: Enhanced LLM Capabilities through Test-Time Compute},
author = {SuperLLM Contributors},
year = {2024},
url = {https://github.com/to314as/superllm}
}