Run Experiment

Deploy and run ML experiment: $ARGUMENTS

Workflow

Step 1: Detect Environment

Read the project's CLAUDE.md to determine the experiment environment:

• Local GPU (gpu: local): Look for local CUDA/MPS setup info
• Remote server (gpu: remote): Look for SSH alias, conda env, code directory
• Vast.ai (gpu: vast): Check for vast-instances.json at project root — if a running instance exists, use it. Also check CLAUDE.md for a ## Vast.ai section.
• Modal (gpu: modal): Serverless GPU via Modal. No SSH, no Docker, auto scale-to-zero. Delegate to /serverless-modal.

Modal detection: If CLAUDE.md has gpu: modal or a ## Modal section, the entire deployment is handled by /serverless-modal. Jump to Step 4: Deploy (Modal) — Steps 2-3 are not needed (Modal handles code sync and GPU allocation automatically).

Vast.ai detection priority:

1. If CLAUDE.md has gpu: vast or a ## Vast.ai section:
   • If vast-instances.json exists and has a running instance → use that instance
   • If no running instance → call /vast-gpu provision which analyzes the task, presents cost-optimized GPU options, and rents the user's choice
2. If no server info is found in CLAUDE.md, ask the user.

Step 2: Pre-flight Check

Check GPU availability on the target machine:

Remote (SSH):

ssh <server> nvidia-smi --query-gpu=index,memory.used,memory.total --format=csv,noheader

Remote (Vast.ai):

ssh -p <PORT> root@<HOST> nvidia-smi --query-gpu=index,memory.used,memory.total --format=csv,noheader

(Read ssh_host and ssh_port from vast-instances.json, or run vastai ssh-url <INSTANCE_ID> which returns ssh://root@HOST:PORT)

Local:

nvidia-smi --query-gpu=index,memory.used,memory.total --format=csv,noheader
# or for Mac MPS:
python -c "import torch; print('MPS available:', torch.backends.mps.is_available())"

Free GPU = memory.used < 500 MiB.

Step 3: Sync Code (Remote Only)

Check the project's CLAUDE.md for a code_sync setting. If not specified, default to rsync.

Option A: rsync (default)

Only sync necessary files — NOT data, checkpoints, or large files:

rsync -avz --include='*.py' --exclude='*' <local_src>/ <server>:<remote_dst>/

Option B: git (when code_sync: git is set in CLAUDE.md)

Push local changes to remote repo, then pull on the server:

# 1. Push from local
git add -A && git commit -m "sync: experiment deployment" && git push

# 2. Pull on server
ssh <server> "cd <remote_dst> && git pull"

Benefits: version-tracked, multi-server sync with one push, no rsync include/exclude rules needed.

Option C: Vast.ai instance

Sync code to the vast.ai instance (always rsync, code dir is /workspace/project/):

rsync -avz -e "ssh -p <PORT>" \
  --include='*.py' --include='*.yaml' --include='*.yml' --include='*.json' \
  --include='*.txt' --include='*.sh' --include='*/' \
  --exclude='*.pt' --exclude='*.pth' --exclude='*.ckpt' \
  --exclude='__pycache__' --exclude='.git' --exclude='data/' \
  --exclude='wandb/' --exclude='outputs/' \
  ./ root@<HOST>:/workspace/project/

If requirements.txt exists, install dependencies:

scp -P <PORT> requirements.txt root@<HOST>:/workspace/
ssh -p <PORT> root@<HOST> "pip install -q -r /workspace/requirements.txt"

Step 3.5: W&B Integration (when wandb: true in CLAUDE.md)

Skip this step entirely if wandb is not set or is false in CLAUDE.md.

Before deploying, ensure the experiment scripts have W&B logging:

1. Check if wandb is already in the script — look for import wandb or wandb.init. If present, skip to Step 4.
2. If not present, add W&B logging to the training script:

   import wandb
   wandb.init(project=WANDB_PROJECT, name=EXP_NAME, config={...hyperparams...})
   
   # Inside training loop:
   wandb.log({"train/loss": loss, "train/lr": lr, "step": step})
   
   # After eval:
   wandb.log({"eval/loss": eval_loss, "eval/ppl": ppl, "eval/accuracy": acc})
   
   # At end:
   wandb.finish()

3. Metrics to log (add whichever apply to the experiment):
   • train/loss — training loss per step
   • train/lr — learning rate
   • eval/loss, eval/ppl, eval/accuracy — eval metrics per epoch
   • gpu/memory_used — GPU memory (via torch.cuda.max_memory_allocated())
   • speed/samples_per_sec — throughput
   • Any custom metrics the experiment already computes
4. Verify wandb login on the target machine:

   ssh <server> "wandb status"  # should show logged in
   # If not logged in:
   ssh <server> "wandb login <WANDB_API_KEY>"

The W&B project name and API key come from CLAUDE.md (see example below). The experiment name is auto-generated from the script name + timestamp.

Step 4: Deploy

Remote (via SSH + screen)

For each experiment, create a dedicated screen session with GPU binding:

ssh <server> "screen -dmS <exp_name> bash -c '\
  eval \"\$(<conda_path>/conda shell.bash hook)\" && \
  conda activate <env> && \
  CUDA_VISIBLE_DEVICES=<gpu_id> python <script> <args> 2>&1 | tee <log_file>'"

Vast.ai instance

No conda needed — the Docker image has the environment. Use /workspace/project/ as working dir:

ssh -p <PORT> root@<HOST> "screen -dmS <exp_name> bash -c '\
  cd /workspace/project && \
  CUDA_VISIBLE_DEVICES=<gpu_id> python <script> <args> 2>&1 | tee /workspace/<log_file>'"

After launching, update the experiment field in vast-instances.json for this instance.

Modal (serverless)

When gpu: modal is detected, delegate to /serverless-modal:

1. Analyze task — determine VRAM needs, choose GPU, estimate cost
2. Generate launcher — create a modal_launcher.py that wraps the training script using modal.Mount.from_local_dir for code and modal.Volume for results
3. Run — modal run modal_launcher.py (runs locally, GPU executes remotely)
4. Collect results — results return via Volume or stdout, no manual download needed

Key Modal settings from CLAUDE.md:

• modal_gpu: GPU override (default: auto-select based on VRAM analysis)
• modal_timeout: Max seconds (default: 21600 = 6 hours)
• modal_volume: Named volume for persistent results

No SSH, no code sync, no screen sessions needed. Modal handles everything.

Local

# Linux with CUDA
CUDA_VISIBLE_DEVICES=<gpu_id> python <script> <args> 2>&1 | tee <log_file>

# Mac with MPS (PyTorch uses MPS automatically)
python <script> <args> 2>&1 | tee <log_file>

For local long-running jobs, use run_in_background: true to keep the conversation responsive.

Parallel runs in one milestone — nohup ... & wait:

mkdir -p runs/<run_id_A> runs/<run_id_B>
CUDA_VISIBLE_DEVICES=<gpu_A> nohup python <script> --seed 42  > runs/<run_id_A>/run.log 2>&1 &
CUDA_VISIBLE_DEVICES=<gpu_B> nohup python <script> --seed 200 > runs/<run_id_B>/run.log 2>&1 &
wait

Shell wait joins direct children only — do not use until ! pgrep -f '<script>.py'; do sleep N; done (it matches the polling bash's own cmdline → infinite loop). Caller invokes /monitor-experiment afterward to finalize each runs/<run-id>/cost.json.

Step 5: Verify Launch

Remote (SSH):

ssh <server> "screen -ls"

Remote (Vast.ai):

ssh -p <PORT> root@<HOST> "screen -ls"

Modal:

modal app list         # Check app is running
modal app logs <app>   # Stream logs

Local: Check process is running and GPU is allocated.

Step 5.5: Write initial cost manifest (mandatory, canonical)

Immediately after Step 5 verifies launch, write runs/<run-id>/cost.json with what is known at launch time. This file is the canonical, machine-readable cost record for the run — /monitor-experiment finalizes it on completion (Step 3.6 there), and downstream consumers (/auto-experiment Phase 5, /auto-iteration-loop Phase C, /auto-verify) read from this file rather than parsing free-form deploy logs.

Schema (write all fields; use null where the value isn't known yet):

{
  "run_id": "<run-id>",
  "status": "running",
  "started_at": "<ISO 8601 UTC, e.g. 2026-05-18T03:21:00Z>",
  "ended_at": null,
  "gpu_ids": [0, 1, 2, 3],
  "gpu_provider": "remote | local | vast | modal",
  "wall_clock_seconds": null,
  "gpu_seconds": null,
  "gpu_hours": null
}

gpu_ids is the effective CUDA_VISIBLE_DEVICES list (the leading positional arg parsed in Step 4, post-pinning). For Modal runs where the device list is auto-assigned, write the GPU tier as a single-element array (e.g., ["A100-80GB"]) and let the finalization step substitute actual numbers from the Modal API.

If runs/<run-id>/ doesn't exist yet, create it (mkdir -p).

Step 6: Auto-Destroy Vast.ai Instance (when gpu: vast and auto_destroy: true)

Skip this step if not using vast.ai or auto_destroy is false.

After the experiment completes (detected via /monitor-experiment or screen session ending):

1. Download results from the instance:

   rsync -avz -e "ssh -p <PORT>" root@<HOST>:/workspace/project/results/ ./results/

2. Download logs:

   scp -P <PORT> root@<HOST>:/workspace/*.log ./logs/

3. Destroy the instance to stop billing:

   vastai destroy instance <INSTANCE_ID>

4. Update vast-instances.json — mark status as destroyed.
5. Report cost:

   Vast.ai instance <ID> auto-destroyed.
   - Duration: ~X.X hours
   - Estimated cost: ~$X.XX
   - Results saved to: ./results/

This ensures users are never billed for idle instances. When auto_destroy: true (the default), the full lifecycle is automatic: rent → setup → run → collect → destroy.

Key Rules

• ALWAYS check GPU availability first — never blindly assign GPUs (except Modal, which manages allocation automatically)
• Each experiment gets its own screen session + GPU (remote) or background process (local)
• Use tee to save logs for later inspection
• Run deployment commands with run_in_background: true to keep conversation responsive
• Report back: which GPU, which screen/process, what command, estimated time
• If multiple experiments, launch them in parallel on different GPUs
• Vast.ai cost awareness: When using gpu: vast, always report the running cost. If auto_destroy: true, destroy the instance as soon as all experiments on it complete
• Modal cost awareness: Always estimate and display cost before running. Modal auto-scales to zero — no idle billing, no manual cleanup

CLAUDE.md Example

Users should add their server info to their project's CLAUDE.md:

## Remote Server
- gpu: remote               # use pre-configured SSH server
- SSH: `ssh my-gpu-server`
- GPU: 4x A100 (80GB each)
- Conda: `eval "$(/opt/conda/bin/conda shell.bash hook)" && conda activate research`
- Code dir: `/home/user/experiments/`
- code_sync: rsync          # default. Or set to "git" for git push/pull workflow
- wandb: false              # set to "true" to auto-add W&B logging to experiment scripts
- wandb_project: my-project # W&B project name (required if wandb: true)
- wandb_entity: my-team     # W&B team/user (optional, uses default if omitted)

## Vast.ai
- gpu: vast                  # rent on-demand GPU from vast.ai
- auto_destroy: true         # auto-destroy after experiment completes (default: true)
- max_budget: 5.00           # optional: max total $ to spend per experiment

## Modal
- gpu: modal                 # serverless GPU via Modal (no SSH, auto scale-to-zero)
- modal_gpu: A100-80GB       # optional: override GPU selection (default: auto-select)
- modal_timeout: 21600       # optional: max seconds (default: 6 hours)
- modal_volume: my-results   # optional: named volume for results persistence

## Local Environment
- gpu: local                 # use local GPU
- Mac MPS / Linux CUDA
- Conda env: `ml` (Python 3.10 + PyTorch)

Vast.ai setup: Run pip install vastai && vastai set api-key YOUR_KEY. Upload your SSH public key at https://cloud.vast.ai/manage-keys/. Set gpu: vast in your CLAUDE.md — /run-experiment will automatically rent an instance, run the experiment, and destroy it when done.

Modal setup: Run pip install modal && modal setup. Bind a payment method at https://modal.com/settings (NEVER through CLI) to unlock the full $30/month free tier (without card: $5/month only). Set a workspace spending limit to prevent accidental charges. Set gpu: modal in your CLAUDE.md — ideal for users without a local GPU who need to debug code or run small-scale tests.

W&B setup: Run wandb login on your server once (or set WANDB_API_KEY env var). The skill reads project/entity from CLAUDE.md and adds wandb.init() + wandb.log() to your training scripts automatically. Dashboard: https://wandb.ai/<entity>/<project>.