update

finmuse-deliverables/finmuse-radeon-source/README.md

@@ -0,0 +1,165 @@
+# FinMuse Radeon
+
+FinMuse Radeon is a Track 1 multimodal content creation demo for financial customer product promotion and introduction services. It helps banks, securities firms, insurers and wealth management teams transform one product brief into scenario-based marketing copy, visual prompts, SVG posters, a landing page, voiceover text and compliance-aware metadata.
+
+The project is designed to demonstrate how AMD Radeon GPU / ROCm can support local, stable and reproducible financial content generation workflows. The baseline code uses a standard-library rule engine so judges can run it anywhere; on an AMD ROCm environment the same commands automatically record the GPU detection path.
+
+## 1. Project background
+
+Financial institutions need to introduce products across many daily-life scenarios:
+
+- Wealth management for commute, family planning and retirement review.
+- Insurance product education for family health, accident protection and travel.
+- Credit card and banking services for mobile payment, shopping and business use.
+- Fund and investment education for portfolio review and risk awareness.
+- Branch screens, mobile apps, relationship-manager presentations and social videos.
+
+Traditional production is fragmented: product teams write briefs, marketing teams create copy, design teams create visuals, compliance teams review wording, and channel teams adapt the same idea repeatedly. FinMuse Radeon compresses this into a repeatable generation pipeline.
+
+## 2. What the demo generates
+
+For each financial product brief, the CLI generates:
+
+- `variants.json`: scenario-based content variants.
+- `poster-01.svg`, `poster-02.svg`, `poster-03.svg`: high-resolution 1280x720 SVG promotional cards.
+- `landing-page.html`: product introduction landing page.
+- `gpu-status.json`: AMD Radeon / ROCm detection result.
+- `run-summary.json`: runtime, clarity score, stability score and diversity tags.
+- `demo-report.md`: cross-run report summarizing clarity, stability and diversity.
+
+## 3. Repository structure
+
+```text
+finmuse-radeon/
+  finmuse/
+    __init__.py
+    cli.py            # command line interface
+    generator.py      # financial content generation logic
+    gpu.py            # AMD Radeon / ROCm detection helpers
+  examples/
+    wealth-card.json
+    insurance-family.json
+  outputs/
+    wealth-card/      # generated sample output
+    insurance-family/ # generated sample output
+    demo-report.md
+    gpu-status.json
+  scripts/
+    run_demo.sh
+  requirements.txt
+  README.md
+```
+
+## 4. Quick start
+
+> The baseline demo has no mandatory third-party dependency.
+
+```bash
+cd finmuse-radeon
+python -m finmuse.cli check-gpu
+python -m finmuse.cli generate --brief examples/wealth-card.json --variants 3 --out outputs/wealth-card
+python -m finmuse.cli generate --brief examples/insurance-family.json --variants 3 --out outputs/insurance-family
+python -m finmuse.cli report --run outputs/wealth-card --run outputs/insurance-family --out outputs/demo-report.md
+```
+
+On this packaging machine, the recorded status is CPU fallback because `rocm-smi` and `rocminfo` are not available. The demo does not fake an AMD GPU result.
+
+## 5. AMD Radeon GPU / ROCm validation commands
+
+On a Linux machine with supported AMD Radeon GPU and ROCm installed, run:
+
+```bash
+rocm-smi
+rocminfo | head -80
+python -m finmuse.cli check-gpu --out outputs/gpu-status.json
+python -m finmuse.cli generate --brief examples/wealth-card.json --variants 6 --out outputs/wealth-card-rocm
+python -m finmuse.cli report --run outputs/wealth-card-rocm --out outputs/demo-report-rocm.md
+```
+
+Expected evidence in `gpu-status.json`:
+
+```json
+{
+  "rocm_smi_found": true,
+  "rocminfo_found": true,
+  "mode": "amd_rocm_gpu",
+  "devices": ["AMD Radeon ..."]
+}
+```
+
+If PyTorch ROCm is installed, `gpu.py` also checks whether the HIP backend is available. The current baseline generator is lightweight; a production extension can replace the rule engine with LLM, diffusion, TTS and video models running through ROCm-enabled runtimes.
+
+## 6. System architecture
+
+```text
+Financial Brief JSON
+      |
+      v
+Brief Parser ----> Compliance Guardrails
+      |                    |
+      v                    v
+Scenario Planner --> Variant Generator --> Quality Metrics
+      |                    |
+      v                    v
+Visual Prompt Builder  Copy / Voiceover Builder
+      |                    |
+      +-------> Asset Renderer: SVG posters + HTML landing page
+                              |
+                              v
+                  Metadata + GPU Runtime Recorder
+```
+
+Key design choices:
+
+- Financial-scenario planner maps products to life moments.
+- Compliance guardrails avoid unrealistic return promises.
+- Variant generator produces diverse scenes, tone words and channels.
+- Quality metrics record clarity, stability and diversity.
+- GPU recorder captures AMD Radeon / ROCm runtime evidence.
+
+## 7. Model and algorithm design
+
+The hackathon baseline uses deterministic generation so judges can reproduce results exactly:
+
+- Seed creation from product brief hash.
+- Scenario library by product type: wealth, insurance, credit, loan and fund.
+- Tone library: premium, warm, youth and stable.
+- Rule-based headline, subtitle, caption and voiceover templates.
+- SVG/HTML rendering for visible multimodal assets.
+- Clarity score and stability score for demo evaluation.
+- Diversity tags to prove multi-scenario output coverage.
+
+Production extension path:
+
+- LLM for product-brief understanding and compliance rewriting.
+- Diffusion model for financial-scene key visuals.
+- TTS for relationship-manager narration.
+- Video compositor for branch-screen and social-video formats.
+- ROCm/HIP acceleration for local GPU inference.
+
+## 8. Demo evidence generated in this package
+
+The included sample run generated two product campaigns:
+
+1. `Aurora Smart Wealth Card`
+2. `Family Shield Plan`
+
+Each campaign includes three content variants and reports:
+
+- Average clarity score.
+- Average stability score.
+- Scenario and channel diversity tags.
+- Runtime and hardware mode.
+
+## 9. Limitations and honest GPU statement
+
+This packaged environment does not expose AMD Radeon GPU or ROCm tools, so the included execution result is `cpu_fallback`. The source code is designed to detect and record AMD ROCm evidence when run on supported hardware. The demo video therefore shows:
+
+- Actual command-line run in the current environment.
+- Actual generated outputs from the packaged code.
+- The exact AMD ROCm commands required for GPU validation.
+- A clear note that no GPU result is fabricated in this environment.
+
+## 10. License and compliance note
+
+This is a hackathon prototype. Financial copy is for demonstration only and must be reviewed by qualified compliance and legal teams before real customer use.

finmuse-deliverables/finmuse-radeon-source/examples/insurance-family.json

@@ -0,0 +1,14 @@
+{
+  "institution": "Radeon Trust Insurance",
+  "product_name": "Family Shield Plan",
+  "product_type": "insurance",
+  "target_audience": "young families and urban professionals",
+  "tone": "warm",
+  "channels": ["branch screen", "mobile app", "social video"],
+  "benefits": [
+    "life-stage protection explanation",
+    "family health and accident coverage story",
+    "advisor-assisted policy comparison"
+  ],
+  "disclaimer": "Insurance coverage is subject to official policy terms and underwriting review."
+}

finmuse-deliverables/finmuse-radeon-source/examples/wealth-card.json

@@ -0,0 +1,14 @@
+{
+  "institution": "Aurora Bank",
+  "product_name": "Aurora Smart Wealth Card",
+  "product_type": "wealth",
+  "target_audience": "young professionals and mass affluent families",
+  "tone": "premium",
+  "channels": ["mobile app", "branch screen", "social short video"],
+  "benefits": [
+    "goal-based asset allocation introduction",
+    "one-stop banking and wealth management service",
+    "risk-aware portfolio education"
+  ],
+  "disclaimer": "For demonstration only. Investment involves risk and requires suitability assessment."
+}

finmuse-deliverables/finmuse-radeon-source/finmuse/__init__.py

@@ -0,0 +1 @@
+__version__ = "0.2.0"

finmuse-deliverables/finmuse-radeon-source/finmuse/cli.py

@@ -0,0 +1,45 @@
+from __future__ import annotations
+
+import argparse
+import json
+from pathlib import Path
+
+from .generator import build_report, generate_package
+from .gpu import write_status
+
+
+def _print_json(data):
+    print(json.dumps(data, ensure_ascii=False, indent=2))
+
+
+def main(argv=None):
+    parser = argparse.ArgumentParser(prog="finmuse", description="FinMuse Radeon financial product multimodal demo")
+    sub = parser.add_subparsers(dest="command", required=True)
+
+    p_gpu = sub.add_parser("check-gpu", help="Detect AMD Radeon / ROCm runtime and write gpu-status.json")
+    p_gpu.add_argument("--out", default="outputs/gpu-status.json")
+
+    p_gen = sub.add_parser("generate", help="Generate financial product promotional assets")
+    p_gen.add_argument("--brief", required=True, help="Path to product brief JSON")
+    p_gen.add_argument("--variants", type=int, default=3)
+    p_gen.add_argument("--out", required=True, help="Output directory")
+
+    p_report = sub.add_parser("report", help="Build markdown report from one or more run directories")
+    p_report.add_argument("--run", action="append", required=True)
+    p_report.add_argument("--out", default="outputs/demo-report.md")
+
+    args = parser.parse_args(argv)
+    if args.command == "check-gpu":
+        Path(args.out).parent.mkdir(parents=True, exist_ok=True)
+        data = write_status(args.out)
+        _print_json(data)
+    elif args.command == "generate":
+        data = generate_package(args.brief, args.out, args.variants)
+        _print_json(data)
+    elif args.command == "report":
+        text = build_report(args.run, args.out)
+        print(text)
+
+
+if __name__ == "__main__":
+    main()