docs/AMD_CREDIT_RUNBOOK.md

AMD Credit Runbook

Use this checklist as soon as AMD Developer Cloud credits are approved.

1. Create Instance

Target:

• AMD Developer Cloud
• AMD Instinct MI300X
• ROCm 6.x image if available
• Enough disk for videos, model cache, and rendered clips

2. Clone Repository

git clone https://github.com/JakgritB/ElevenClip.AI.git
cd ElevenClip.AI

3. Configure Environment

cp .env.example .env

Edit .env:

DEMO_MODE=false
HF_TOKEN=<your-hugging-face-token>
WHISPER_MODEL_ID=openai/whisper-large-v3
QWEN_TEXT_MODEL_ID=Qwen/Qwen2.5-7B-Instruct
QWEN_VL_MODEL_ID=Qwen/Qwen2-VL-7B-Instruct
FFMPEG_VIDEO_CODEC=h264_amf

4. Verify ROCm

rocminfo | head
rocm-smi

Verify PyTorch:

python - <<'PY'
import torch
print("cuda available:", torch.cuda.is_available())
print("device:", torch.cuda.get_device_name(0) if torch.cuda.is_available() else "none")
print("hip:", torch.version.hip)
PY

On ROCm, PyTorch still exposes AMD GPUs through the torch.cuda API.

5. Start Backend And Frontend

Docker path:

docker compose build --build-arg INSTALL_EXTRAS=.[ai,rocm-inference] backend
docker compose up

Manual backend path:

cd backend
python -m venv .venv
source .venv/bin/activate
pip install -e ".[ai,rocm-inference]"
uvicorn app.main:app --host 0.0.0.0 --port 8000

Manual frontend path:

cd frontend
npm install
npm run dev -- --host 0.0.0.0

6. Run Benchmark

CPU baseline:

DEMO_MODE=false HIP_VISIBLE_DEVICES= python scripts/benchmark.py --youtube-url "<demo-video-url>" --language Thai --style informative --niche education --clip-length 60

AMD GPU:

DEMO_MODE=false python scripts/benchmark.py --youtube-url "<demo-video-url>" --language Thai --style informative --niche education --clip-length 60

Save the JSON outputs into:

data/benchmarks/cpu.json
data/benchmarks/mi300x.json

7. Update Submission Materials

After the benchmark:

• Update README.md.
• Update docs/SUBMISSION.md.
• Update docs/PITCH_DECK.md.
• Update Hugging Face Space.
• Record the final demo video.