Skip to content

Suggestion: add WFGY (LLM robustness + RAG debugging framework) #1

@onestardao

Description

@onestardao

Hi, thanks for maintaining this collection of machine learning lists.

I would like to suggest one project that might fit under the LLM / RAG / evaluation or debugging related resources:

WFGY is a text-only framework that you feed into any strong LLM. It tries to treat LLM pipelines as systems that can be debugged, not just prompted.

A few parts that might be relevant to your readers:

  1. WFGY 2.0 ProblemMap (RAG and LLM failure modes)
    https://github.com/onestardao/WFGY/tree/main/ProblemMap

    • 16 canonical failure modes for RAG and LLM applications (hallucination, chunk drift, retrieval collapse, entropy collapse, bootstrap ordering, deployment deadlock, etc.).
    • Each problem has its own page with failure patterns, diagnosis procedure and proposed fixes.
    • Used as a checklist when people wonder “is this a vector store problem, a prompt problem, or a data problem”.
  2. WFGY 1.0 technical report (math + structure in prompts)
    PDF inside the repo with a more formal description of the math-in-prompt idea and quantitative benchmarks.

  3. WFGY 3.0 Singularity Demo (advanced, for researchers)
    A TXT pack with 131 “S-class” tension-based questions for long-horizon reasoning and evaluation. This is more experimental, but some researchers use it as a stress test for LLM reasoning.

If you feel this is in scope, I would be happy to open a PR and place it in the section you think fits best.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions