feat(agentx): display physical cache-tier token sources

[cquil11]

Summary

• ingest vllm:prompt_tokens_cached_by_source from AgentX server-metric artifacts
• combine fresh prefill with physical HBM, CPU, NVMe, and connector-defined cache-hit tiers without double counting
• retain the existing logical vLLM fallback and SGLang source mapping for older runs
• display tier-specific colors and English/Simplified Chinese labels on the point-detail chart
• bump the derived chart-series version so existing rows can be recomputed

Validation

• bun run typecheck
• bun run lint
• bun run fmt
• bun --cwd packages/db vitest run src/etl/compute-chart-series.test.ts
• bun --cwd packages/app vitest run src/components/inference/agentic-point/time-series-math.test.ts src/lib/api-route-catalog.test.ts
• bun --cwd packages/app vitest run src/lib/zh-copy.test.ts
• bun run --cwd packages/app test:e2e:component --spec cypress/component/agentic-token-source-chart.cy.tsx

Live-data plan

Run InferenceX AgentX sweep 33169766010 will be ingested into an isolated Neon branch cloned from production. This PR's Vercel preview will be scoped to that database and verified against its DRAM, NVMe, and tiered DRAM+NVMe points.

Compatibility

No relational migration is required. The existing JSONB chart_series payload is versioned from v15 to v16. Historical rows without the new vLLM metric keep their current source breakdown.

---

Note

Medium Risk
Changes ETL aggregation and chart-series versioning for prompt-token metrics, so incorrect tier merging or missed backfill would skew displayed breakdowns; sibling/API field additions are lower risk UI and read-path changes.

Overview
AgentX prompt-token source breakdown now uses vLLM’s physical prompt_tokens_cached_by_source metric when present: fresh prefill still comes from the logical prompt_tokens_by_source buckets (local_compute / miss), while HBM, CPU, NVMe, connector-defined tiers, and related labels replace the old logical local_cache_hit / external_kv_transfer series so stacked totals do not double-count. Chart-series derivation bumps to v16 (no SQL migration; stale JSONB rows need recomputation/backfill). Older vLLM blobs and SGLang’s existing mapping stay on the prior breakdown.

The point-detail UI adds tier-specific colors and English / Simplified Chinese labels on the stacked prefill-source chart (plus a Cypress component test). Sibling navigation loads kv_offloading, p90_intvty, and p90_ttft from the benchmark-siblings API (cache namespace v4), shows physical offload tiers on chips (DRAM, NVMe, combined) instead of generic off=ON, and rings chips that sit on both P90 interactivity and TTFT Pareto frontiers vs throughput per GPU.

Smaller fixes: eval ETL maps agentic ISL/OSL 0 to null; Pareto helpers in chart-utils are generic over {x,y} points; API docs and data-pipeline notes describe the new attribution rules.

^{Reviewed by Cursor Bugbot for commit 6452884. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
inferencemax-app	Ready	Preview	Sep 3, 2026 8:21pm UTC

[cquil11]

Preview test environment is ready.

• Preview: https://inferencemax-app-git-feat-cache-tier-toke-788211-semianalysisai.vercel.app
• Neon branch: pr-944-cache-tier-token-sources (br-divine-cake-ai4cq1m7), cloned from production and expiring 2026-09-04
• Ingested source run: https://github.com/SemiAnalysisAI/InferenceX/actions/runs/33169766010
• Ingest result: 25 benchmark points and 25 trace-replay sidecars; database verification passed

Representative points:

• NVMe / vllm-simple: https://inferencemax-app-git-feat-cache-tier-toke-788211-semianalysisai.vercel.app/inference/agentic/440534 — Fresh, HBM Cache Hit, NVMe Offload Cache Hit
• Native DRAM+NVMe: https://inferencemax-app-git-feat-cache-tier-toke-788211-semianalysisai.vercel.app/inference/agentic/440549 — Fresh, HBM Cache Hit, CPU Offload Cache Hit
• DRAM / Mooncake: https://inferencemax-app-git-feat-cache-tier-toke-788211-semianalysisai.vercel.app/inference/agentic/440541 — Fresh, HBM Cache Hit, External Cache Hit
• Chinese NVMe view: https://inferencemax-app-git-feat-cache-tier-toke-788211-semianalysisai.vercel.app/zh/inference/agentic/440534

Verified the deployed API response, English and Chinese page routes, browser rendering, and deployment runtime logs (zero 5xx responses during verification).

[cquil11]

Added in commit 68a8d4fd:

• Sibling chips now use physical KV tiers: DRAM, NVMe, or DRAM+NVMe; no-offload points have no suffix and the legacy off=ON text is gone.
• Points on both the P90 interactivity/throughput and P90 TTFT/throughput Pareto frontiers receive a double orange outline, with a legend below the selector.
• The deployed 25-point run marks three intersection points: NVMe c=12, DRAM c=9, and DRAM+NVMe c=7.

Verified on the deployed English and Chinese pages. All PR checks are green.

Preview: https://inferencemax-app-git-feat-cache-tier-toke-788211-semianalysisai.vercel.app/inference/agentic/440549