Datasets:

---

tarekmasryo

/

llm-system-ops-production-telemetry-sft-data

like 1

🧠 Synthetic production-style telemetry for LLM systems — built for ops analysis, monitoring, and supervision-style datasets.

📦 Dataset snapshot

• Structured records around prompts, responses, routing/tool behavior, and operational metadata
• Focus: reliability + cost/latency + quality slices

🔎 What’s inside

• Run-level traces (requests, responses, outcomes)
• Routing/tooling signals (where relevant)
• Ops telemetry: tokens, latency, estimated cost, failure/retry-style signals
• Supervision-style views (for SFT-like experiments), if provided

✅ Good for
📊 Monitoring dashboards (p50/p95 latency, token usage, cost per request, error rates)
🧪 Reliability analysis (failure modes, retries, fallback behavior)
🤖 Guardrail/risk policies (when to re-run, abstain, or route stronger)

⚠️ Notes

• Synthetic / privacy-safe → benchmarking + demos, not real user ground truth.
• Avoid leakage across aggregates computed after the final outcome.

🧾 Not for high-stakes decisions (clinical/legal/financial).