WFGY/ProblemMap/GlobalFixMap/LLM_Providers/README.md
2025-08-26 16:55:38 +08:00

10 KiB
Raw Blame History

LLM Providers: Guardrails and Fix Patterns

A compact hub to stabilize provider-specific failures without changing your infra. Use this when symptoms look “model problem” but root cause is actually schema, retrieval, or orchestration.

Open these first

Core acceptance

  • ΔS(question, retrieved) ≤ 0.45
  • Coverage ≥ 0.70 for the target section
  • λ remains convergent across three paraphrases and two seeds
  • E_resonance stays flat through long windows

Typical provider symptoms → exact fix

Symptom Likely cause Open this
JSON mode breaks, invalid objects schema too loose or nested tool calls Data Contracts, Logic Collapse
Tool calls loop or stall agent role drift, missing timeouts Multi-Agent Problems, role-drift deep dive
High similarity yet wrong snippet metric mismatch or fragmented store Embedding ≠ Semantic, Vectorstore Fragmentation
Answers flip between runs prompt headers reorder and λ flips Context Drift, Retrieval Traceability
Hybrid retrievers worse than single query parsing split, mis-weighted rerank Query Parsing Split, Rerankers
Jailbreaks or bluffing overconfidence and missing fences Bluffing Controls, Retrieval Traceability

Fix in 60 seconds

  1. Measure ΔS
    Compute ΔS(question, retrieved) and ΔS(retrieved, expected anchor).
    Stable < 0.40, transitional 0.400.60, risk ≥ 0.60.

  2. Probe λ_observe
    Vary k and prompt headers. If λ flips, lock the schema and apply a BBAM variance clamp.

  3. Apply the module

  • Retrieval drift → BBMC + Data Contracts
  • Reasoning collapse → BBCR bridge + BBAM
  • Dead ends in long runs → BBPF alternate paths
  1. Verify
    Coverage ≥ 0.70 on three paraphrases. λ convergent on two seeds.

Provider-level gotchas checklist

  • Truncation. Confirm token accounting for system + tools + hidden preambles. If truncated, compress citations through Data Contracts.
  • Streaming chunk boundaries. Do not parse partial JSON while λ is unstable. Buffer until BBAM settles.
  • Temperature and top-p. If ΔS is already high, reduce entropy. If retrieval is sparse, raise recall through rerankers instead of temperature.
  • Multi-model routing. Keep traceability stable when swapping GPT, Claude, Gemini, Mistral. Use the same snippet schema and citation header across providers.
  • Rate limits and retries. Backoff with idempotent ops. Never rebuild indexes inside retry loops.
  • Eval parity. Run the same acceptance on all providers to avoid overfitting a single model.

Quick routes to per-provider pages


🔗 Quick-Start Downloads (60 sec)

Tool Link 3-Step Setup
WFGY 1.0 PDF Engine Paper 1 Download · 2 Upload to your LLM · 3 Ask “Answer using WFGY + <your question>”
TXT OS (plain-text OS) TXTOS.txt 1 Download · 2 Paste into any LLM chat · 3 Type “hello world” — OS boots instantly

🧭 Explore More

Module Description Link
WFGY Core WFGY 2.0 engine is live: full symbolic reasoning architecture and math stack View →
Problem Map 1.0 Initial 16-mode diagnostic and symbolic fix framework View →
Problem Map 2.0 RAG-focused failure tree, modular fixes, and pipelines View →
Semantic Clinic Index Expanded failure catalog: prompt injection, memory bugs, logic drift View →
Semantic Blueprint Layer-based symbolic reasoning & semantic modulations View →
Benchmark vs GPT-5 Stress test GPT-5 with full WFGY reasoning suite View →
🧙‍♂️ Starter Village 🏡 New here? Lost in symbols? Click here and let the wizard guide you through Start →

👑 Early Stargazers: See the Hall of Fame
Engineers, hackers, and open source builders who supported WFGY from day one.

GitHub stars WFGY Engine 2.0 is already unlocked. Star the repo to help others discover it and unlock more on the Unlock Board.

WFGY Main   TXT OS   Blah   Blot   Bloc   Blur   Blow  

say “GO” and Ill do the first provider page. I suggest ProblemMap/GlobalFixMap/LLM_Providers/openai.md next.