| #!/usr/bin/env bash |
| set -euo pipefail |
|
|
| if [[ $# -lt 2 || $# -gt 3 ]]; then |
| echo "usage: $0 <uniform|residual> <run-id> [proposal.json]" >&2 |
| exit 64 |
| fi |
|
|
| arm=$1 |
| run_id=$2 |
| proposal=${3:-} |
| project="${OPTR_PROJECT_ROOT:-$(cd "$(dirname "${BASH_SOURCE[0]}")/.." && pwd)}" |
| python="${OPTR_PYTHON:-python}" |
| config="${project}/configs/florence_bridge_20k_plus500.yaml" |
| accelerate_config="${project}/configs/deepspeed_florence_2gpu.yaml" |
|
|
| required_env=(OPTR_RUN_ROOT FLORENCE_MODEL_PATH BRIDGE_DATA_ROOT BRIDGE_STATISTICS_PATH BRIDGE_PROVENANCE_PATH BRIDGE_FINGERPRINT_PATH BRIDGE_MANIFEST_PATH COMMON_START_20K) |
| for name in "${required_env[@]}"; do |
| [[ -n "${!name:-}" ]] || { echo "missing environment variable: $name" >&2; exit 64; } |
| done |
| [[ -d "$COMMON_START_20K" ]] || { echo "common start not found: $COMMON_START_20K" >&2; exit 66; } |
| [[ -f "$COMMON_START_20K/checkpoint.json" ]] || { echo "checkpoint.json missing under common start" >&2; exit 66; } |
| [[ -f "$BRIDGE_STATISTICS_PATH" ]] || { echo "statistics file missing: $BRIDGE_STATISTICS_PATH" >&2; exit 66; } |
| [[ -f "$BRIDGE_PROVENANCE_PATH" ]] || { echo "provenance file missing: $BRIDGE_PROVENANCE_PATH" >&2; exit 66; } |
| [[ -f "$BRIDGE_FINGERPRINT_PATH" ]] || { echo "content fingerprint missing: $BRIDGE_FINGERPRINT_PATH" >&2; exit 66; } |
| [[ -f "$BRIDGE_MANIFEST_PATH" ]] || { echo "anchor manifest missing: $BRIDGE_MANIFEST_PATH" >&2; exit 66; } |
|
|
| case "$arm" in |
| uniform) |
| [[ -z "$proposal" ]] || { echo "uniform does not accept a proposal" >&2; exit 64; } |
| entry=starVLA/training/train_starvla.py |
| proposal_args=(--datasets.vla_data.temporal_proposal_path=null) |
| ;; |
| residual) |
| [[ -f "$proposal" ]] || { echo "proposal not found: $proposal" >&2; exit 66; } |
| entry=starVLA/training/train_starvla.py |
| proposal_args=(--datasets.vla_data.temporal_proposal_path="$proposal") |
| ;; |
| *) echo "unknown arm: $arm" >&2; exit 64 ;; |
| esac |
|
|
| cd "$project" |
| export PYTHONPATH="${project}:${PYTHONPATH:-}" |
| export WANDB_MODE="${WANDB_MODE:-offline}" |
| export PYTHONUNBUFFERED=1 |
| export PYTORCH_CUDA_ALLOC_CONF="${PYTORCH_CUDA_ALLOC_CONF:-expandable_segments:True}" |
| mkdir -p "$OPTR_RUN_ROOT" |
| "$python" -m accelerate.commands.launch \ |
| --config_file "$accelerate_config" \ |
| --num_processes 2 \ |
| "$entry" \ |
| --config_yaml "$config" \ |
| --run_id "$run_id" \ |
| --run_root_dir "$OPTR_RUN_ROOT" \ |
| "${proposal_args[@]}" |
|
|
