README.md

metadata

license: apache-2.0
library_name: docker
tags:
  - finc2e
  - humai
  - governance
  - mlops
  - decision-intelligence
  - dual-metrics
  - docker
  - runtime
  - might-hub

FinC2E Dual-Metrics Runtime (Preview) — v1.0.0

A lightweight, container-first runtime for FinC2E that introduces Dual Metrics as a foundation for governance-grade AI systems.

This image is part of the shared Hugging Face → NVIDIA NGC release flow and is designed to work alongside FinC2E-Governance.

---

What is Dual Metrics?

Dual Metrics means measuring two dimensions simultaneously:

1) Runtime / System Metrics

• Service health
• Startup telemetry
• Adapter pull timing
• Deployment readiness

2) Decision-System Readiness

• Foundations for audit-friendly execution
• Governance-aligned decision trace hooks
• Separation of runtime logic and private model weights

This runtime does not expose private weights.
It securely pulls them at startup using gated access.

---

Hugging Face Repositories

• Runtime (this repository):
  MightHubHumAI/FinC2E_DualMetrics_Runtime

• Adapter (private):
  MightHubHumAI/FinC2E_Llama33_70B_Adapter

• Model Preview:
  MightHubHumAI/FinC2E

• Governance Space:
  FinC2E-Governance / HumAI FinC2E HQ

---

Image Tags

• Stable (immutable): 1.0.0
• Moving: latest

For reproducible tests and mirrors (NGC / Brev), use 1.0.0.

---

Pull & Run

Pull image

docker pull registry.huggingface.co/MightHubHumAI/FinC2E_DualMetrics_Runtime:1.0.0
Run container
docker run --rm \
  -e HF_TOKEN=YOUR_HF_TOKEN \
  -e ADAPTER_REPO=MightHubHumAI/FinC2E_Llama33_70B_Adapter \
  -p 8000:8000 \
  registry.huggingface.co/MightHubHumAI/FinC2E_DualMetrics_Runtime:1.0.0


HF_TOKEN must have read access to the private adapter repository.

Endpoints
Health
curl http://localhost:8000/health


Example response:

{"status":"ok","service":"FinC2E Runtime"}

Dual Metrics
curl http://localhost:8000/metrics


Example response:

{
  "runtime": "FinC2E_DualMetrics_Runtime",
  "adapter_repo": "MightHubHumAI/FinC2E_Llama33_70B_Adapter",
  "download_seconds": 4.242,
  "timestamp": "2025-09-16T14:30:00Z"
}

Environment Variables

HF_TOKEN (required)
Hugging Face token with permission to access the private adapter repository.

ADAPTER_REPO (optional)
Defaults to MightHubHumAI/FinC2E_Llama33_70B_Adapter.

PORT (optional)
Defaults to 8000.

Security & IP Model

Adapter weights are private and never embedded in the image.

Runtime image is public, auditable, and mirrorable.

This separation enables:

controlled access

governance and audit readiness

Hugging Face → NVIDIA NGC parity

Brev and enterprise deployment flows

Release Notes — v1.0.0

First public runtime release.

Secure adapter pull via HF_TOKEN.

/health and /metrics endpoints available.

Immutable 1.0.0 tag + rolling latest.

HF → NGC → Brev Shared Release Discipline

Hugging Face tag 1.0.0 == NVIDIA NGC tag 1.0.0.

No breaking changes under the same tag.

All future changes increment semver (1.0.1, 1.1.0, 2.0.0).

This ensures deterministic behavior across registries and deployment platforms.

Roadmap

Upcoming iterations include:

Decision trace endpoints (JSONL, audit-ready).

Model routing and orchestration hooks (Galaxy container foundations).

NVIDIA NGC packaging and listing.

Brev deployment profiles for GPU-backed environments.

License

Apache-2.0 applies to runtime code and container packaging.
Model adapters and weights are governed separately under a private license.