This is a concrete example of Microsoft 365 product fragmentation observed while trying to implement a simple business requirement.
contact@energration.com
AGENTS.md (Energration eudorus repository root) — follow this document for how to use the eudorus repo (agents, tools, GitHub policy, Kiro/Codex workflow, steering, ontology, credentials). Verbatim copy below.
Hand-maintained. Full list of agent guidance artifacts and how they are generated or maintained: docs/agent_guidance_inventory.md. Run
make agent-guidance-inventoryto display it. Quick-start operator contract:docs/how_to_work_with_eudorus_codex.md(one-page execution playbook).
On unified-memory systems (e.g. DGX Spark, UMA) or when repeatedly swapping models (7B ↔ 120B), Linux page cache can be reported as "used" memory by PyTorch/Ray/vLLM and similar runtimes. That causes false OOM or "free memory on device is less than desired" on startup even though the memory is reclaimable. This spec captures the mitigation as a first-class feature: drop page cache before model launch in the deploy flow so the runtime sees accurate free memory.
Endpoint: gx10 120B (Qwen 122B) at http://gx10-83fb.tail3dac72.ts.net:8002
Script: benchmark_120b_tokens_per_second.py
| What we measure | How |
| #!/usr/bin/env python3 | |
| """ | |
| Benchmark tokens per second and TTFT for the gx10 120B LLM endpoint (Qwen 122B). | |
| Calls POST /v1/chat/completions (non-streaming for throughput; optional streaming | |
| for TTFT). Supports multiple runs (mean ± std) and optional concurrent requests. | |
| Usage: | |
| python3 scripts/benchmark_120b_tokens_per_second.py [BASE_URL] | |
| python3 scripts/benchmark_120b_tokens_per_second.py --runs 5 --ttft --concurrent 2 |
Telemetry from the GX10-83FB host is exported to Prometheus and visualized in Grafana. Prometheus and Grafana run in the eudorus observatory stack in Docker on Zane. Access from any machine on the network (GX10, vonnegut, etc.) must use Zane’s Tailscale hostname, not localhost.
The observatory stack runs in Docker on Zane. Nginx is the router in front; deployment path on Zane:
/Users/lou/migration/rootfs/home/lou/observatory-deployment
Config: observatory/nginx/nginx.conf; compose: docker-compose.yml.
Interrogation date: 2026-03-13
How to check status and whether the model is coming up: See LLM_STATUS_AND_HEALTH.md.
| File | Purpose |
|---|
Host: gx10-83fb (Tailscale: gx10-83fb.tail3dac72.ts.net)
Port for 120B: 8002 (single active 120B service at a time)
Recommended stack: Qwen3.5 122B A10B + llama.cpp (Option D)
Authoritative artifacts to read first: docs/GX10_PORT_ASSIGNMENT.md, docs/evidence/gx10_runtime_baseline.json, ontology/configuration_management.ttl.
Required first step: Refresh runtime evidence with python3 scripts/capture_gx10_runtime_baseline.py, then run ./scripts/gx10_config_guard.py [--live] before any change. If it fails, fix the reported issues first.
Endpoint: gx10 120B (Qwen 122B) at http://gx10-83fb.tail3dac72.ts.net:8002
Script: scripts/benchmark_120b_tokens_per_second.py
| What we measure | How |