v1.2.0 — Training, LoRA/PEFT, DVC, experiment tracking, LangGraph advisor

[1.2.0] — 2026-07-27

v1.2.0 close-out: advanced GPU packaging (validate-only), cost-per-token, training backends,
LoRA/PEFT, DVC for deterministic exports, pluggable experiment tracking, and a tool-grounded
LangGraph advisor with non-fabrication CI. Tag: v1.2.0.
Feature commits:

Phase commits (16–22)

Phase	Commit	Summary
16	96cb7e1	Advanced GPU serving — GPTQ/AWQ/FP8 + TensorRT-LLM templates
17	164ef1a	Cost-per-token + GPU utilization tracking
18	179a1cc	Training backends + training-job contract
19	21d590b	LoRA/PEFT adapter fine-tuning + transparent serving
20	baa2ba4	DVC for deterministic model exports
21	3e81029	Pluggable experiment tracking (MLflow + W&B offline)
22	c61e16c	LangGraph advisor — tool-grounded, non-fabrication

Phase 22 — LangGraph advisor (tool-grounded, non-fabrication)

• advisor/: LangGraph with query_prometheus, read_benchmark_results, query_routing_history, recommend (ADR-014)
• Synthesis via template + optional pinned reference-tiny-llm; no paid LLM API in CI
• CI asserts every numeric/backend claim in the answer appears in that run’s tool evidence (extends ADR-007)

Phase 21 — Pluggable experiment tracking (MLflow + W&B offline)

Feature commit 3e81029; artifact-serve fix 62510a7.

• training/common/tracking.py: ExperimentTracker interface; MlflowTracker + WandbTracker (offline-only); default VULCAN_TRACKER_BACKEND=none (ADR-013)
• FSDP/DDP + LoRA report existing loss/throughput through the interface (no recomputation)
• Compose MLflow on host port 9014; CI asserts MLflow API metrics + W&B offline-run-* dir (never wandb.ai)

Phase 20 — DVC data versioning (deterministic model exports)

Feature commit baa2ba4.

• dvc.yaml wraps existing export_llm.py / export_vision.py; local filesystem remote only in CI (ADR-012)
• Cross-check: DVC-tracked primary outs' SHA256 must match sha256sums.txt / MANIFEST (does not replace MANIFEST)
• Scope: deterministic reference exports only — no training/adapter content-hash tracking (extends ADR-009 / ADR-011)
• CI dvc-repro job: dvc repro + clean dvc status + unchanged dvc.lock

Phase 19 — LoRA / PEFT (adapter fine-tuning + transparent serving)

Feature commit 21d590b.

• LoraFineTuneSpec / LoraFineTuneResult in contracts/training-job-contract/ (ADR-011; extends ADR-010)
• training/fsdp-ddp/lora/: CPU PEFT fine-tune on pinned reference-tiny-llm; structural adapter verify (no SHA256 pin — ADR-009)
• BentoML optionally serves reference-tiny-llm-lora-demo through unchanged /v1/infer (ADR-001)
• CI: logits delta proof (base vs base+adapter) on a fixed prompt

Phase 18 — Training backends (v1.2.0 track)

Feature commit 179a1cc.

• contracts/training-job-contract/: TrainingJobSpec / TrainingJobResult OpenAPI + JSON Schema (ADR-010)
• training/{ray-train,fsdp-ddp,deepspeed}/: CPU-simulated distributed training (gloo, world_size=2; ADR-009)
• FSDP/DDP SIGTERM → checkpoint → resume test; DeepSpeed ZeRO-1/2 CPU path + GPU runbook
• Cost-exporter: vulcan_estimated_cost_usd_per_training_step from training results × ADR-008 $/GPU-hour
• Host ports 9011–9013 (compose profile training); ADR-009 / ADR-010

Phase 17 — Cost-per-token and GPU utilization tracking

Feature commit 164ef1a.

• observability/gpu-metrics/: real DCGM-exporter Helm values + cluster Prometheus scrape for phase-7 pools; synthetic DCGM-shaped exporter for compose/CI
• Cost-exporter: vulcan_estimated_cost_usd_per_token from Bedrock pricing-reference.json and ADR-008 $/GPU-hour × benchmark throughput (phase-7 instance types only)
• Grafana: cost-per-token + LIVE-SYNTHETIC GPU util panels; LIVE vs PLACEHOLDER table updated
• ADR-008; KNOWN_GAPS #2 closed; CI smoke asserts cost-per-token + synthetic DCGM

Phase 16 — Advanced GPU serving (v1.1.0 track)

Feature commit 96cb7e1.

• Extended serving/vllm/docs/gpu-mode.md: continuous batching, PagedAttention vs CPU KV, speculative decoding pairs
• serving/vllm/gpu-variants/{gptq,awq,fp8}/: schema-valid resource manifests (supports_quantization, declared VRAM envelopes)
• serving/triton/tensorrt-llm/: TensorRT-LLM config.pbtxt template + Dockerfile + runbook
• ADR-007; adr-gate; CI validate-only (no GPU, no invented numbers)

v1.0.0 — Security hardening, docs site, release close-out

[1.0.0] — 2026-07-26

Release hardening: security scans, docs site, coverage extensions, demo/case-study honesty docs. Tag: v1.0.0.

Phase 15 — Security hardening, docs site, and v1.0.0 release

Feature commit cba0197; tag v1.0.0 includes the follow-up changelog hash fix.

• Trivy (CRITICAL, ignore-unfixed) + Syft SBOM on built images: bentoml, ray-serve, triton, triton-engine, vllm, vllm-engine, gateway, cost-exporter
• Semgrep (p/python, p/golang) across serving/gateway/pipelines Python + Go
• Coverage gate ≥65% extended to gateway/internal and observability/cost-exporter; exemptions documented in CONTRIBUTING + KNOWN_GAPS
• MkDocs Material site wiring existing READMEs + ADRs 001–006 (make docs-serve)
• docs/DEMO_SCRIPT.md, docs/CASE_STUDY.md, docs/KNOWN_GAPS.md (commands verified against the live CPU stack)

Phase commits (0–15)

Phase	Commit	Summary
0	bd92ee4	Foundations, model contract, CI skeleton
1	e8d1e88	Reference models and benchmark harness
2	c7f7a02	BentoML adapter
3	181c72c	Ray Serve adapter
4	3e35d86	Triton adapter
5	54c5bb0	vLLM adapter
6	372dc19	KServe adapter
7	c5c4fdf	GPU Operator, device plugin, MIG
8	ba466fc	Kueue multi-tenant GPU scheduling
9	856277f	Karpenter GPU autoscaling and checkpoint-resume
10	559203d	SageMaker Pipelines, Endpoints, Model Registry
11	eece3f6	Bedrock-aware gateway adapter
12	cf0c480	Kubeflow Pipelines + Training Operator → KServe
13	3d83fe3	Routing gateway and benchmark-driven selection
14	04beccb	Observability — tracing, metrics, cost dashboards
15	cba0197	Security hardening, docs site, v1.0.0 release

Phase 14 — Observability (tracing, metrics, cost dashboards) — 04beccb

• observability/: Prometheus (:9008), Grafana (:9009), Tempo (:9010), OTel collector, Alertmanager, blackbox
• OTel on gateway + bentoml/ray-serve/triton/vllm; scrapes existing phase-0 /metrics
• Cost-exporter reuses benchmark + Bedrock pricing-reference; LIVE vs PLACEHOLDER GPU panels (ADR-002)

Phase 13 — Routing gateway — 3d83fe3

• gateway/ on :9007; ADR-006; explainable fallback; SageMaker/KServe excluded without recorded data

Phase 12 — Kubeflow → KServe — cf0c480

• KFP train→eval→register; Training Operator + KServe handoff; validate-only CI

Phase 11 — Bedrock adapter — eece3f6

• bedrock-gateway/ + static pricing-reference.json; moto CI

Phase 10 — SageMaker — 559203d

• Pipelines / Registry / Endpoint helpers; moto CI; manual runbook

Phase 9 — Karpenter + checkpointing — 856277f

• Spot NodePools; SIGTERM checkpoint library; ADR-005

Phase 8 — Kueue — ba466fc

• Multi-tenant GPU queues; ADR-004

Phase 7 — GPU Operator / MIG — c5c4fdf

• Operator values, MIG profiles, terraform GPU EKS plan-only; ADR-003

Phase 6 — KServe — 372dc19

• Helm InferenceServices wrapping Triton/vLLM; validate-only

Phase 5 — vLLM — 54c5bb0

• Host :9004; LLM-only shim; CPU k6 artifact

Phase 4 — Triton — 3e35d86

• Host :9003 shim + internal engine; ONNX model repo

Phase 3 — Ray Serve — 181c72c

• Host :9002; both reference models

Phase 2 — BentoML — c7f7a02

• Host ports 9000–9099 convention; :9000 adapter

Phase 1 — Models + benchmark — e8d1e88

• GPT-2 + ResNet-18 pins; k6 harness; serving/common client + conformance

Phase 0 — Foundations — bd92ee4

• Contract OpenAPI/JSON Schema; ADR-001/002; CI skeleton; CPU compose placeholder