gitadmin/foxhunt
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity
Files
46fea9a0e39fe82db8a7eda4cc43bdfd92cc299a
foxhunt/BINARY_UPLOAD_QUICK_REF.md

7.2 KiB
Raw Blame History

Binary Upload Quick Reference

Script: scripts/upload_binary.py Purpose: Quick binary uploads to RunPod S3 for hyperparameter optimization workflows Status: ✅ PRODUCTION READY Last Updated: 2025-10-30

Quick Start

# 1. Activate .venv
source .venv/bin/activate

# 2. Upload binary (auto-finds in target/release/examples/)
python3 scripts/upload_binary.py --binary-name hyperopt_mamba2_demo

# Output:
# ✅ Uploaded to: s3://se3zdnb5o4/binaries/hyperopt_mamba2_demo_cuda_20251030_001234
# Next: /runpod-volume/binaries/hyperopt_mamba2_demo_cuda_20251030_001234

Common Use Cases

1. Upload Latest Hyperopt Binary (Default)

python3 scripts/upload_binary.py --binary-name hyperopt_mamba2_demo
# → binaries/hyperopt_mamba2_demo_cuda_20251030_120534

2. Force Overwrite (Skip Checksum)

python3 scripts/upload_binary.py --binary-name hyperopt_tft_demo --force
# Uploads even if MD5 matches

3. Upload Without Timestamp (Static Name)

python3 scripts/upload_binary.py --binary-name hyperopt_dqn_demo --no-timestamp
# → binaries/hyperopt_dqn_demo_cuda
# ⚠️ Overwrites existing file!

4. Upload Non-CUDA Binary

python3 scripts/upload_binary.py --binary-name custom_tool --no-cuda
# → binaries/custom_tool_20251030_120534

5. Dry Run (Validation Only)

python3 scripts/upload_binary.py --binary-name hyperopt_ppo_demo --dry-run
# Validates binary but doesn't upload

6. Upload Custom Path

python3 scripts/upload_binary.py --binary-path ./my_custom_binary --force
# Upload from anywhere

Features

Automatic Binary Location

  • Searches target/release/examples/ by name
  • Handles build hashes (e.g., hyperopt_mamba2_demo-84b145a77f64618b)
  • Selects most recent if multiple matches

Validation

  • ✅ Checks file exists and is executable
  • ✅ Validates size (warns if < 100KB)
  • ✅ MD5 checksum comparison (skips upload if unchanged)

S3 Organization

s3://se3zdnb5o4/binaries/
├── hyperopt_mamba2_demo_cuda_20251030_120000
├── hyperopt_tft_demo_cuda_20251030_143000
├── hyperopt_dqn_demo_cuda_20251030_150000
└── hyperopt_ppo_demo_cuda_20251030_163000

Progress Tracking

Uploading hyperopt_mamba2_demo ━━━━━━━━━━ 100% • 14.2 MB • 45.3 MB/s • 0:00:00
✅ Upload complete!
  S3 URI: s3://se3zdnb5o4/binaries/hyperopt_mamba2_demo_cuda_20251030_120534

Integration with Deployment

Step 1: Build Binary

cargo build --release --example hyperopt_mamba2_demo --features cuda

Step 2: Upload to S3

python3 scripts/upload_binary.py --binary-name hyperopt_mamba2_demo
# → /runpod-volume/binaries/hyperopt_mamba2_demo_cuda_20251030_120534

Step 3: Deploy to RunPod

python3 scripts/runpod_deploy.py \
  --gpu-type "RTX A4000" \
  --command "/runpod-volume/binaries/hyperopt_mamba2_demo_cuda_20251030_120534 \
    --parquet-file /runpod-volume/test_data/ES_FUT_180d.parquet \
    --trials 50 \
    --timeout 2h \
    --s3-bucket se3zdnb5o4 \
    --s3-prefix hyperopt_runs/mamba2/"

Options Reference

Option Description Example
--binary-name Binary name (auto-finds) hyperopt_mamba2_demo
--binary-path Direct path to binary ./custom_binary
--force Force upload (skip checksum) --force
--no-timestamp Static name (overwrites) --no-timestamp
--no-cuda Omit _cuda suffix --no-cuda
--dry-run Validate only --dry-run

Requirements

Environment

# Activate .venv (REQUIRED)
source .venv/bin/activate

Configuration (.env.runpod)

RUNPOD_S3_ACCESS_KEY=<access_key>
RUNPOD_S3_SECRET=<secret_key>
RUNPOD_VOLUME_ID=se3zdnb5o4
RUNPOD_S3_ENDPOINT=https://s3api-eur-is-1.runpod.io
RUNPOD_S3_REGION=eur-is-1

Dependencies

pip install -r foxhunt_runpod/foxhunt_runpod/requirements.txt
# Installs: boto3, rich, pydantic-settings, python-dotenv

Troubleshooting

"Binary not found"

# Build binary first
cargo build --release --example hyperopt_mamba2_demo --features cuda

# Verify location
ls -lh target/release/examples/hyperopt_mamba2_demo

"Not running in virtual environment"

source .venv/bin/activate
python3 scripts/upload_binary.py --help

"Configuration error"

# Verify .env.runpod exists
cat .env.runpod | grep RUNPOD_S3

# Check S3 credentials
aws s3 ls s3://se3zdnb5o4/binaries/ \
  --profile runpod \
  --endpoint-url https://s3api-eur-is-1.runpod.io

"Binary suspiciously small"

# Check binary was built with --release
cargo build --release --example <name> --features cuda

# Debug build produces smaller, unoptimized binaries

Workflow Examples

Example 1: MAMBA-2 Hyperopt Iteration

# 1. Update code
vim ml/examples/hyperopt_mamba2_demo.rs

# 2. Rebuild
cargo build --release --example hyperopt_mamba2_demo --features cuda

# 3. Upload
python3 scripts/upload_binary.py --binary-name hyperopt_mamba2_demo
# → binaries/hyperopt_mamba2_demo_cuda_20251030_153400

# 4. Deploy
python3 scripts/runpod_deploy.py \
  --command "/runpod-volume/binaries/hyperopt_mamba2_demo_cuda_20251030_153400 \
    --trials 50 --timeout 2h"

Example 2: Quick Overwrite (Same Binary Name)

# Fast iteration: overwrite with static name
cargo build --release --example hyperopt_dqn_demo --features cuda
python3 scripts/upload_binary.py \
  --binary-name hyperopt_dqn_demo \
  --no-timestamp \
  --force

# Deploy always uses same path
python3 scripts/runpod_deploy.py \
  --command "/runpod-volume/binaries/hyperopt_dqn_demo_cuda"

Example 3: Multi-Binary Upload

# Upload all hyperopt binaries at once
for binary in hyperopt_mamba2_demo hyperopt_tft_demo hyperopt_dqn_demo hyperopt_ppo_demo; do
  python3 scripts/upload_binary.py --binary-name $binary
done

Performance Notes

Binary Size Upload Time (50 Mbps) Typical Use
hyperopt_mamba2_demo 14.2 MB ~2.3s MAMBA-2 hyperopt
hyperopt_tft_demo 21.1 MB ~3.4s TFT hyperopt
hyperopt_dqn_demo 13.3 MB ~2.1s DQN hyperopt
hyperopt_ppo_demo 13.0 MB ~2.1s PPO hyperopt

MD5 Checksum: If file unchanged, upload skipped (0s)

Best Practices

  1. Always Use .venv: Ensures correct dependencies
  2. Use Timestamps: Allows version history (default)
  3. Force Only When Needed: Saves bandwidth
  4. Dry Run First: Validate before uploading large files
  5. Static Names for Stable Workflows: Use --no-timestamp for production deployments

Related Documentation

  • RUNPOD_VOLUME_MOUNT_ARCHITECTURE.md: Volume mount system design
  • RUNPOD_DEPLOY_SCRIPT_UPDATE.md: Full deployment workflow
  • ML_TRAINING_PARQUET_GUIDE.md: Training binary usage
  • HYPEROPT_DEPLOYMENT_COMPLETE.md: Hyperopt system architecture

Version History

v1.0.0 (2025-10-30)

  • ✅ Auto-locates binaries in target/release/examples/
  • ✅ MD5 checksum validation (skips unchanged)
  • ✅ Timestamped naming for version control
  • ✅ Progress bar with transfer speed
  • ✅ Dry run mode for validation
  • ✅ Integration with foxhunt_runpod.S3Client
Reference in New Issue View Git Blame Copy Permalink