Symflower for LLMs

Symflower supports LLM development, benchmarking, LLM-based code generation workflows, and Retrieval-Augmented Generation (RAG):

Prompt engineering: Augment prompts with data from Symflower's analytics tools to provide the LLM with more context about the code.
Improving efficiency: Code repair makes LLM-generated code more useful by removing compile errors.
Training data curation: Using Symflower helps filter out projects with problematic code to provide higher-quality training data to the LLM.
LLM training, fine-tuning, RAG: Providing Symflower-generated tests or statically analyzed code to the LLM in a RAG scenario, during LLM training, or fine-tuning helps improve accuracy. Fine-tuning with Symflower showed an 8% improvement in a Java<-> Python transpiler.
Benchmarking: The DevQualityEval benchmark covers a variety of metrics to evaluate code quality and help find the most useful LLMs for the evaluated software development tasks.

Symflower's support for LLMs:

`symflower fix` is a static analysis tool to repair code generated by LLMs. Use it on local files to post-process the output of LLMs.

`symflower run` sends a main function to Symflower for symbolic execution and returns coverage as a result.

`symflower symbols` queries all functions and methods in a repository and outputs a list of all the symbols.

`symflower test` executes the tests in your project and outputs test results and coverage data.

`symflower test-runner` runs test impact analysis on the package level to identify and run tests affected by code changes in your query.

`symflower trace` provides visual insights into application execution behavior to analyze application performance.

DevQualityEval is a standardized LLM evaluation benchmark and framework to compare and improve code generation with LLMs.