chore: refactor for sweeps and IP configs

docker-compose.yml

@@ -10,9 +10,13 @@ services:
       - HF_TOKEN=${HF_TOKEN}
       - WANDB_API_KEY=${WANDB_API_KEY}
       - GITHUB_TOKEN=${GITHUB_TOKEN}
+      - GOOGLE_APPLICATION_CREDENTIALS=/secrets/gcp-sa.json
+      - GCP_ACCOUNT=${GCP_ACCOUNT:-}
+      - WATCHDOG_CONFIG_PATTERN=${WATCHDOG_CONFIG_PATTERN:-v6e_*.conf}
       - CLOUDSDK_CONFIG=/.config/gcloud
     volumes:
       - ~/.config/gcloud:/.config/gcloud:rw
+      - ./secrets/gcp-sa.json:/secrets/gcp-sa.json:ro
 
   tensorboard-rl:
     image: tensorflow/tensorflow:latest

docker/TPUWatchdog.dockerfile

@@ -36,24 +36,54 @@ if [ -n "$GOOGLE_APPLICATION_CREDENTIALS" ] && [ -f "$GOOGLE_APPLICATION_CREDENT
         echo "Authenticating gcloud using service account key..."
         gcloud auth activate-service-account --key-file="$GOOGLE_APPLICATION_CREDENTIALS"
 
-        # Extract project ID from the key file
+        if [ -z "$PROJECT_ID" ]; then
-        PROJECT_ID=$(jq -r '.project_id' "$GOOGLE_APPLICATION_CREDENTIALS")
+            PROJECT_ID=$(jq -r '.project_id // empty' "$GOOGLE_APPLICATION_CREDENTIALS")
-        if [ -n "$PROJECT_ID" ] && [ "$PROJECT_ID" != "null" ]; then
-            gcloud config set project "$PROJECT_ID"
-            echo "Set project to $PROJECT_ID"
         fi
+    elif [ "$CRED_TYPE" = "authorized_user" ]; then
+        echo "Authenticating gcloud using authorized_user refresh token..."
+
+        AUTH_ACCOUNT="$GCP_ACCOUNT"
+        if [ -z "$AUTH_ACCOUNT" ]; then
+            AUTH_ACCOUNT=$(jq -r '.account // empty' "$GOOGLE_APPLICATION_CREDENTIALS")
+        fi
+        if [ -z "$AUTH_ACCOUNT" ]; then
+            AUTH_ACCOUNT=$(gcloud config get-value account 2>/dev/null || true)
+        fi
+
+        REFRESH_TOKEN=$(jq -r '.refresh_token // empty' "$GOOGLE_APPLICATION_CREDENTIALS")
+        if [ -z "$AUTH_ACCOUNT" ] || [ -z "$REFRESH_TOKEN" ]; then
+            echo "Error: authorized_user credentials require GCP_ACCOUNT (or embedded account) and refresh_token."
+            exit 1
+        fi
+
+        gcloud auth activate-refresh-token "$AUTH_ACCOUNT" "$REFRESH_TOKEN"
     else
-        echo "Note: Using application default credentials or mounted gcloud config..."
+        echo "Warning: unsupported credential file type '$CRED_TYPE'. Falling back to mounted gcloud config."
     fi
 else
     echo "Note: Assuming gcloud config is mounted from host."
 fi
 
+if [ -n "$PROJECT_ID" ]; then
+    gcloud config set project "$PROJECT_ID"
+    echo "Set project to $PROJECT_ID"
+fi
+
 # Run the watchdogs in the background using bash instead of tmux
 # Tmux needs a TTY to attach properly which we might not have in docker
 # Stagger startups by 15s to prevent simultaneous TPU creation quota hits
+CONFIG_PATTERN=${WATCHDOG_CONFIG_PATTERN:-"*.conf"}
+shopt -s nullglob
+CONFIGS=(/app/tpu_orchestration/configs/$CONFIG_PATTERN)
+
+if [ ${#CONFIGS[@]} -eq 0 ]; then
+    echo "Error: no watchdog configs matched pattern '$CONFIG_PATTERN'."
+    exit 1
+fi
+
+echo "Using watchdog config pattern: $CONFIG_PATTERN"
 DELAY=0
-for conf in /app/tpu_orchestration/configs/*.conf; do
+for conf in "${CONFIGS[@]}"; do
     echo "Starting watchdog for $(basename "$conf" .conf) (delay: ${DELAY}s)"
     (sleep $DELAY && /app/tpu_orchestration/watchdog.sh "$conf") &
     DELAY=$((DELAY + 15))

scripts/ray_distributed_train.py

@@ -1,14 +1,18 @@
 from __future__ import annotations
 
 import argparse
+import contextlib
+import concurrent.futures
 import os
 import shlex
 import subprocess
 import sys
+import threading
 import time
 from pathlib import Path
 
 import ray
+from ray.util.scheduling_strategies import NodeAffinitySchedulingStrategy
 
 
 def _has_flag(tokens: list[str], name: str) -> bool:
@@ -24,18 +28,301 @@ def _entry_tokens(run_kind: str, entry_args: str) -> list[str]:
     return tokens
 
 
-def _alive_node_ips() -> list[str]:
+def _get_flag_value(tokens: list[str], name: str, default: str = "") -> str:
+    for idx, tok in enumerate(tokens):
+        if tok == name and idx + 1 < len(tokens):
+            return str(tokens[idx + 1])
+        if tok.startswith(f"{name}="):
+            return str(tok.split("=", 1)[1])
+    return str(default)
+
+
+def _set_flag_value(tokens: list[str], name: str, value: str) -> list[str]:
+    updated: list[str] = []
+    replaced = False
+    idx = 0
+    while idx < len(tokens):
+        tok = tokens[idx]
+        if tok == name:
+            replaced = True
+            updated.extend([name, str(value)])
+            idx += 2
+            continue
+        if tok.startswith(f"{name}="):
+            replaced = True
+            updated.append(f"{name}={value}")
+            idx += 1
+            continue
+        updated.append(tok)
+        idx += 1
+    if not replaced:
+        updated.extend([name, str(value)])
+    return updated
+
+
+def _remove_flag(tokens: list[str], name: str) -> list[str]:
+    updated: list[str] = []
+    idx = 0
+    while idx < len(tokens):
+        tok = tokens[idx]
+        if tok == name:
+            idx += 1
+            continue
+        if tok.startswith(f"{name}="):
+            idx += 1
+            continue
+        updated.append(tok)
+        idx += 1
+    return updated
+
+
+def _csv_values(raw: str) -> list[str]:
+    return [piece.strip() for piece in str(raw).split(",") if piece.strip()]
+
+
+def _alpha_token(alpha: str) -> str:
+    return str(alpha).replace(".", "p").replace("-", "m")
+
+
+def _truthy(value: str | bool | None) -> bool:
+    if isinstance(value, bool):
+        return value
+    if value is None:
+        return False
+    return str(value).strip().lower() in {"1", "true", "yes", "on"}
+
+
+def _alive_nodes() -> list[tuple[str, str]]:
     seen: set[str] = set()
-    ips: list[str] = []
+    nodes: list[tuple[str, str]] = []
     for node in ray.nodes():
         if not bool(node.get("Alive", False)):
             continue
+        node_id = str(node.get("NodeID", "")).strip()
         ip = str(node.get("NodeManagerAddress", "")).strip()
-        if not ip or ip in seen:
+        if not node_id or not ip or node_id in seen:
             continue
-        seen.add(ip)
+        seen.add(node_id)
-        ips.append(ip)
+        nodes.append((node_id, ip))
-    return sorted(ips)
+    return sorted(nodes, key=lambda item: (item[1], item[0]))
+
+
+def _benchmark_cells(
+    tokens: list[str], *, compare_robust: bool
+) -> list[tuple[str, str, str, bool]]:
+    tiers = _csv_values(
+        _get_flag_value(tokens, "--tiers", "static,surge,linear,qtable,ppo")
+    )
+    alphas = _csv_values(_get_flag_value(tokens, "--alpha-values", "0.0,0.3,0.6"))
+    base_no_robust = _has_flag(tokens, "--no-robust")
+    if compare_robust:
+        modes = [("robust", False), ("no_robust", True)]
+    else:
+        modes = [("no_robust", True)] if base_no_robust else [("robust", False)]
+    return [
+        (tier, alpha, mode_label, no_robust)
+        for tier in tiers
+        for alpha in alphas
+        for mode_label, no_robust in modes
+    ]
+
+
+def _thread_limited_env(env: dict[str, str], threads: int) -> dict[str, str]:
+    bounded = dict(env)
+    n = str(max(1, int(threads)))
+    for key in (
+        "OMP_NUM_THREADS",
+        "MKL_NUM_THREADS",
+        "OPENBLAS_NUM_THREADS",
+        "NUMEXPR_NUM_THREADS",
+        "VECLIB_MAXIMUM_THREADS",
+        "BLIS_NUM_THREADS",
+    ):
+        bounded[key] = n
+    return bounded
+
+
+@contextlib.contextmanager
+def _semaphore_guard(semaphore: threading.Semaphore | None):
+    if semaphore is None:
+        yield
+        return
+    semaphore.acquire()
+    try:
+        yield
+    finally:
+        semaphore.release()
+
+
+def _run_benchmark_cells_parallel(
+    *,
+    root: str,
+    env: dict[str, str],
+    base_tokens: list[str],
+    compare_robust: bool,
+    inner_workers: int,
+    inner_threads: int,
+    max_heavy_workers: int,
+    rank: int,
+) -> int:
+    cells = _benchmark_cells(base_tokens, compare_robust=compare_robust)
+    if not cells:
+        return 0
+
+    cwd = str(Path(root))
+    base_out = _get_flag_value(base_tokens, "--output-dir", "engine/studies/results")
+    max_workers = max(1, min(int(inner_workers), len(cells)))
+    heavy_tiers = {"ppo", "a2c", "dqn"}
+    heavy_limit = max(1, int(max_heavy_workers))
+    heavy_sem = threading.Semaphore(heavy_limit)
+    print(
+        {
+            "rank": int(rank),
+            "benchmark_cells": len(cells),
+            "inner_workers": int(max_workers),
+            "inner_threads": int(max(1, int(inner_threads))),
+            "heavy_limit": int(heavy_limit),
+        }
+    )
+
+    def _run_cell(
+        index: int,
+        total: int,
+        tier: str,
+        alpha: str,
+        mode_label: str,
+        no_robust: bool,
+    ) -> tuple[str, str, str, int]:
+        tokens = list(base_tokens)
+        tokens = _set_flag_value(tokens, "--tiers", tier)
+        tokens = _set_flag_value(tokens, "--alpha-values", alpha)
+        if no_robust:
+            if not _has_flag(tokens, "--no-robust"):
+                tokens.append("--no-robust")
+        else:
+            tokens = _remove_flag(tokens, "--no-robust")
+
+        cell_out = (
+            Path(base_out)
+            / f"tier_{tier}"
+            / f"mode_{mode_label}"
+            / f"alpha_{_alpha_token(alpha)}"
+        )
+        tokens = _set_flag_value(tokens, "--output-dir", str(cell_out))
+        cmd = [sys.executable, "-m", "engine.train", *tokens]
+        cell_env = _thread_limited_env(env, int(inner_threads))
+        cell_env["PHANTOM_BENCHMARK_COMPARE_ROBUST"] = "0"
+        print(
+            {
+                "rank": int(rank),
+                "cell": f"{index}/{total}",
+                "tier": tier,
+                "mode": mode_label,
+                "alpha": alpha,
+                "command": " ".join(cmd),
+            }
+        )
+        heavy_guard = heavy_sem if str(tier).lower() in heavy_tiers else None
+        with _semaphore_guard(heavy_guard):
+            proc = subprocess.run(cmd, cwd=cwd, env=cell_env)
+        return tier, alpha, mode_label, int(proc.returncode)
+
+    failures: list[tuple[str, str, str, int]] = []
+    with concurrent.futures.ThreadPoolExecutor(max_workers=max_workers) as pool:
+        futures = [
+            pool.submit(_run_cell, idx, len(cells), tier, alpha, mode_label, no_robust)
+            for idx, (tier, alpha, mode_label, no_robust) in enumerate(cells, start=1)
+        ]
+        for fut in concurrent.futures.as_completed(futures):
+            tier, alpha, mode_label, code = fut.result()
+            if code != 0:
+                failures.append((tier, alpha, mode_label, code))
+
+    if failures:
+        print({"rank": int(rank), "benchmark_failures": failures})
+        return 1
+    return 0
+
+
+def _run_sweep_agents_parallel(
+    *,
+    root: str,
+    env: dict[str, str],
+    base_tokens: list[str],
+    run_kind: str,
+    rank: int,
+    agents_per_node: int,
+    agent_count: int,
+    inner_threads: int,
+    tpu_agent_slots: int,
+) -> int:
+    total = max(1, int(agents_per_node))
+    cwd = str(Path(root))
+    wants_tpu = str(env.get("JAX_PLATFORMS", "")).strip().lower() == "tpu"
+    tpu_slots = max(0, int(tpu_agent_slots))
+    print(
+        {
+            "rank": int(rank),
+            "sweep_agents": int(total),
+            "agent_count": int(agent_count),
+            "inner_threads": int(max(1, int(inner_threads))),
+            "jax_platform": str(env.get("JAX_PLATFORMS", "")),
+            "tpu_agent_slots": int(tpu_slots),
+        }
+    )
+
+    def _run_agent(slot: int) -> int:
+        tokens = list(base_tokens)
+        if int(agent_count) > 0 and not _has_flag(tokens, "--count"):
+            tokens.extend(["--count", str(int(agent_count))])
+
+        if _has_flag(tokens, "--group"):
+            base_group = _get_flag_value(tokens, "--group", "ray-sweep")
+            tokens = _set_flag_value(tokens, "--group", f"{base_group}-a{slot}")
+
+        if run_kind == "benchmark":
+            out_dir = _get_flag_value(tokens, "--output-dir", "engine/studies/results")
+            tokens = _set_flag_value(
+                tokens, "--output-dir", str(Path(out_dir) / f"agent_{slot}")
+            )
+        if run_kind == "train":
+            model_dir = _get_flag_value(tokens, "--model-dir", "engine/models")
+            tokens = _set_flag_value(
+                tokens, "--model-dir", str(Path(model_dir) / f"agent_{slot}")
+            )
+
+        cmd = [sys.executable, "-m", "engine.train", *tokens]
+        agent_env = _thread_limited_env(env, int(inner_threads))
+        if wants_tpu and tpu_slots > 0 and int(slot) > tpu_slots:
+            agent_env["JAX_PLATFORMS"] = "cpu"
+            agent_env["JAX_PLATFORM_NAME"] = "cpu"
+        agent_env["PHANTOM_SWEEP_AGENT_SLOT"] = str(int(slot))
+        print(
+            {
+                "rank": int(rank),
+                "agent_slot": int(slot),
+                "jax_platform": str(agent_env.get("JAX_PLATFORMS", "")),
+                "command": " ".join(cmd),
+            }
+        )
+        proc = subprocess.run(cmd, cwd=cwd, env=agent_env)
+        return int(proc.returncode)
+
+    failures: list[tuple[int, int]] = []
+    with concurrent.futures.ThreadPoolExecutor(max_workers=total) as pool:
+        future_map = {
+            pool.submit(_run_agent, slot): slot for slot in range(1, total + 1)
+        }
+        for future in concurrent.futures.as_completed(future_map):
+            slot = int(future_map[future])
+            code = int(future.result())
+            if code != 0:
+                failures.append((slot, code))
+
+    if failures:
+        print({"rank": int(rank), "sweep_failures": failures})
+        return 1
+    return 0
 
 
 @ray.remote(max_retries=0)
@@ -44,6 +331,8 @@ def _train_on_node(
     root: str,
     run_kind: str,
     entry_args: str,
+    node_id: str,
+    node_ip: str,
     rank: int,
     world_size: int,
     coordinator_ip: str,
@@ -54,17 +343,32 @@ def _train_on_node(
     output_root: str,
     wandb_entity: str,
     wandb_project: str,
+    agents_per_node: int,
+    agent_count: int,
+    inner_workers: int,
+    inner_threads: int,
+    max_heavy_workers: int,
     sync_jax: bool,
 ) -> int:
     env = dict(os.environ)
     env["PYTHONUNBUFFERED"] = "1"
     requested_platform = str(env.get("PHANTOM_JAX_PLATFORM", "tpu")).strip().lower()
-    if world_size > 1 and requested_platform == "tpu":
+    allow_multi_node_tpu = _truthy(env.get("PHANTOM_ALLOW_MULTI_NODE_TPU"))
+    if world_size > 1 and requested_platform == "tpu" and not allow_multi_node_tpu:
         requested_platform = "cpu"
         print(
             "PHANTOM_DISTRIBUTED_NOTE: forcing JAX_PLATFORMS=cpu for multi-node SB3 runs "
+            "(set PHANTOM_ALLOW_MULTI_NODE_TPU=1 to keep TPU for JAX workloads)"
+        )
+    elif world_size > 1 and requested_platform == "tpu" and allow_multi_node_tpu:
+        print(
+            "PHANTOM_DISTRIBUTED_NOTE: keeping JAX_PLATFORMS=tpu in multi-node mixed mode"
         )
     env["JAX_PLATFORMS"] = requested_platform
+    if requested_platform == "cpu":
+        env["JAX_PLATFORM_NAME"] = "cpu"
+    else:
+        env.pop("JAX_PLATFORM_NAME", None)
     # Keep each train process in single-host mode to avoid accidental global stalls.
     env["CLOUD_TPU_TASK_ID"] = "0"
     if run_kind == "benchmark":
@@ -96,13 +400,29 @@ def _train_on_node(
         )
 
     tokens = _entry_tokens(run_kind, entry_args)
+    is_sweep_agent = _has_flag(tokens, "--sweep-agent")
     seed = int(base_seed + rank)
-    if not _has_flag(tokens, "--seed"):
+    if not is_sweep_agent and not _has_flag(tokens, "--seed"):
         tokens.extend(["--seed", str(seed)])
 
     if run_kind == "train" and not _has_flag(tokens, "--group"):
         tokens.extend(["--group", run_group])
 
+    if is_sweep_agent and int(agent_count) > 0 and not _has_flag(tokens, "--count"):
+        tokens.extend(["--count", str(int(agent_count))])
+
+    try:
+        tpu_agent_slots = int(
+            str(
+                env.get(
+                    "PHANTOM_TPU_AGENT_SLOTS",
+                    "1" if requested_platform == "tpu" else "0",
+                )
+            ).strip()
+        )
+    except ValueError:
+        tpu_agent_slots = 1 if requested_platform == "tpu" else 0
+
     if (
         run_kind == "benchmark"
         and output_root
@@ -112,9 +432,36 @@ def _train_on_node(
         out_dir.parent.mkdir(parents=True, exist_ok=True)
         tokens.extend(["--output-dir", str(out_dir)])
 
+    if is_sweep_agent and int(agents_per_node) > 1:
+        return _run_sweep_agents_parallel(
+            root=root,
+            env=env,
+            base_tokens=tokens,
+            run_kind=run_kind,
+            rank=rank,
+            agents_per_node=int(agents_per_node),
+            agent_count=int(agent_count),
+            inner_threads=int(inner_threads),
+            tpu_agent_slots=int(max(0, tpu_agent_slots)),
+        )
+
+    if run_kind == "benchmark" and int(inner_workers) > 1 and not is_sweep_agent:
+        return _run_benchmark_cells_parallel(
+            root=root,
+            env=env,
+            base_tokens=tokens,
+            compare_robust=bool(compare_robust),
+            inner_workers=int(inner_workers),
+            inner_threads=int(inner_threads),
+            max_heavy_workers=int(max_heavy_workers),
+            rank=rank,
+        )
+
     cmd = [sys.executable, "-m", "engine.train", *tokens]
     print(
         {
+            "node_id": node_id,
+            "node_ip": node_ip,
             "rank": int(rank),
             "run_kind": run_kind,
             "seed": int(seed),
@@ -124,7 +471,9 @@ def _train_on_node(
             "command": " ".join(cmd),
         }
     )
-    proc = subprocess.run(cmd, cwd=cwd, env=env)
+    proc = subprocess.run(
+        cmd, cwd=cwd, env=_thread_limited_env(env, int(inner_threads))
+    )
     return int(proc.returncode)
 
 
@@ -145,6 +494,12 @@ def main() -> None:
     parser.add_argument("--output-root", type=str, default="")
     parser.add_argument("--wandb-entity", type=str, default="")
     parser.add_argument("--wandb-project", type=str, default="")
+    parser.add_argument("--agents-per-node", type=int, default=1)
+    parser.add_argument("--agent-count", type=int, default=0)
+    parser.add_argument("--inner-workers", type=int, default=1)
+    parser.add_argument("--inner-threads", type=int, default=1)
+    parser.add_argument("--max-heavy-workers", type=int, default=2)
+    parser.add_argument("--worker-cpus", type=float, default=1.0)
     args = parser.parse_args()
 
     entry_args = str(args.entry_args or args.train_args).strip()
@@ -153,21 +508,24 @@ def main() -> None:
 
     ray.init(address="auto")
 
-    node_ips = _alive_node_ips()
+    node_entries = _alive_nodes()
-    if not node_ips:
+    if not node_entries:
         raise RuntimeError("No alive Ray nodes found")
 
     requested = int(args.num_nodes)
     if requested > 0:
-        node_ips = node_ips[:requested]
+        node_entries = node_entries[:requested]
 
-    world_size = len(node_ips)
+    world_size = len(node_entries)
-    coordinator_ip = node_ips[0]
+    coordinator_ip = node_entries[0][1]
     run_group = args.run_group or f"ray-dist-{int(time.time())}"
 
     print(
         {
-            "nodes": node_ips,
+            "nodes": [
+                {"node_id": node_id, "node_ip": node_ip}
+                for node_id, node_ip in node_entries
+            ],
             "world_size": world_size,
             "coordinator": f"{coordinator_ip}:{int(args.coordinator_port)}",
             "run_kind": str(args.run_kind),
@@ -175,18 +533,35 @@ def main() -> None:
             "run_group": run_group,
             "compare_robust": bool(args.compare_robust),
             "output_root": str(args.output_root),
+            "agents_per_node": int(args.agents_per_node),
+            "agent_count": int(args.agent_count),
+            "inner_workers": int(args.inner_workers),
+            "inner_threads": int(args.inner_threads),
+            "max_heavy_workers": int(args.max_heavy_workers),
         }
     )
 
     futures = []
     root = str(Path(__file__).resolve().parents[1])
-    for rank, node_ip in enumerate(node_ips):
+    for rank, (node_id, node_ip) in enumerate(node_entries):
-        resources = {f"node:{node_ip}": 0.01, "TPU": float(args.tpu_per_task)}
+        resources: dict[str, float] = {}
+        tpu_per_task = float(args.tpu_per_task)
+        if tpu_per_task > 0.0:
+            resources["TPU"] = tpu_per_task
         futures.append(
-            _train_on_node.options(resources=resources).remote(
+            _train_on_node.options(
+                resources=resources,
+                num_cpus=float(args.worker_cpus),
+                scheduling_strategy=NodeAffinitySchedulingStrategy(
+                    node_id=node_id,
+                    soft=False,
+                ),
+            ).remote(
                 root=root,
                 run_kind=str(args.run_kind),
                 entry_args=entry_args,
+                node_id=node_id,
+                node_ip=node_ip,
                 rank=rank,
                 world_size=world_size,
                 coordinator_ip=coordinator_ip,
@@ -197,6 +572,11 @@ def main() -> None:
                 output_root=str(args.output_root),
                 wandb_entity=str(args.wandb_entity),
                 wandb_project=str(args.wandb_project),
+                agents_per_node=int(args.agents_per_node),
+                agent_count=int(args.agent_count),
+                inner_workers=int(args.inner_workers),
+                inner_threads=int(args.inner_threads),
+                max_heavy_workers=int(args.max_heavy_workers),
                 sync_jax=bool(args.sync_jax and str(args.run_kind) == "train"),
             )
         )

submit_ray_job.sh

@@ -4,6 +4,7 @@
 #   RAY_MODE=single       -> one run (default)
 #   RAY_MODE=distributed  -> one run per TPU node (experimental)
 #   RAY_MODE=benchmark    -> one benchmark run per TPU node (overnight)
+#   RAY_MODE=sweep        -> distributed W&B sweep agents
 
 set -euo pipefail
 
@@ -28,7 +29,14 @@ env = dotenv_values(".env")
 # Filter out empty/None values
 env_vars = {k: v for k, v in env.items() if v}
 env_vars.setdefault("CLOUD_TPU_TASK_ID", os.getenv("CLOUD_TPU_TASK_ID", "0"))
-for k in ("WANDB_ENTITY", "WANDB_PROJECT", "PHANTOM_BENCHMARK_COMPARE_ROBUST"):
+for k in (
+    "WANDB_ENTITY",
+    "WANDB_PROJECT",
+    "PHANTOM_BENCHMARK_COMPARE_ROBUST",
+    "PHANTOM_JAX_PLATFORM",
+    "PHANTOM_ALLOW_MULTI_NODE_TPU",
+    "PHANTOM_TPU_AGENT_SLOTS",
+):
     if os.getenv(k):
         env_vars[k] = os.getenv(k)
 
@@ -52,6 +60,15 @@ print(json.dumps({
 RAY_MODE="${RAY_MODE:-single}"
 TRAIN_ARGS="${TRAIN_ARGS:---algo ppo --total-timesteps 1000000}"
 BENCHMARK_ARGS="${BENCHMARK_ARGS:---project capstone_tpu --tiers static,surge,linear,qtable,ppo --alpha-values 0.0,0.1,0.25,0.4,0.6,0.8 --episodes 12 --total-timesteps 30000 --max-steps 100 --robust-radius 0.2 --robust-points 7 --robust-rollouts 1 --lambda-coi 0.2 --eta-ux 0.5 --reward-profit-weight 1.0 --device cpu}"
+INNER_WORKERS="${INNER_WORKERS:-16}"
+INNER_THREADS="${INNER_THREADS:-1}"
+MAX_HEAVY_WORKERS="${MAX_HEAVY_WORKERS:-3}"
+WORKER_CPUS="${WORKER_CPUS:-$((INNER_WORKERS * INNER_THREADS))}"
+SWEEP_KIND="${SWEEP_KIND:-benchmark}"
+SWEEP_METHOD="${SWEEP_METHOD:-random}"
+SWEEP_RUN_CAP="${SWEEP_RUN_CAP:-0}"
+AGENTS_PER_NODE="${AGENTS_PER_NODE:-16}"
+AGENT_COUNT="${AGENT_COUNT:-0}"
 
 SUBMIT_ARGS=()
 if [ "${RAY_NO_WAIT:-0}" = "1" ]; then
@@ -104,6 +121,10 @@ if [ "$RAY_MODE" = "benchmark" ]; then
     --output-root "${OUTPUT_ROOT:-engine/studies/results/overnight}"
     --wandb-entity "${WANDB_ENTITY:-lusiana}"
     --wandb-project "${WANDB_PROJECT:-capstone_tpu}"
+    --inner-workers "${INNER_WORKERS}"
+    --inner-threads "${INNER_THREADS}"
+    --max-heavy-workers "${MAX_HEAVY_WORKERS}"
+    --worker-cpus "${WORKER_CPUS}"
   )
   if [ "${COMPARE_ROBUST:-1}" = "1" ]; then
     DIST_ARGS+=(--compare-robust)
@@ -112,5 +133,97 @@ if [ "$RAY_MODE" = "benchmark" ]; then
   exit 0
 fi
 
-echo "Unsupported RAY_MODE='$RAY_MODE' (expected 'single', 'distributed', or 'benchmark')." >&2
+if [ "$RAY_MODE" = "sweep" ]; then
+  SWEEP_PROJECT="${WANDB_PROJECT:-capstone_tpu}"
+  SWEEP_ENTITY="${WANDB_ENTITY:-lusiana}"
+  SWEEP_ID_VALUE="${SWEEP_ID:-}"
+  SWEEP_NUM_NODES="${NUM_NODES:-5}"
+  PY_SWEEP_BIN="${PY_SWEEP_BIN:-}"
+  if [ -z "$PY_SWEEP_BIN" ]; then
+    for cand in "$ROOT/.venv/bin/python" "$ROOT/.venv-ray/bin/python" python3 python; do
+      if [ "$cand" = "python3" ] || [ "$cand" = "python" ]; then
+        command -v "$cand" >/dev/null 2>&1 || continue
+      elif [ ! -x "$cand" ]; then
+        continue
+      fi
+      if "$cand" - <<'PY' >/dev/null 2>&1
+import sys
+from pathlib import Path
+cwd = str(Path.cwd())
+sys.path = [p for p in sys.path if p not in {'', cwd}]
+import wandb
+print(wandb.__name__)
+PY
+      then
+        PY_SWEEP_BIN="$cand"
+        break
+      fi
+    done
+  fi
+  if [ -z "$PY_SWEEP_BIN" ]; then
+    echo "No python interpreter with wandb is available for sweep creation." >&2
+    exit 1
+  fi
+
+  if [ -z "$SWEEP_ID_VALUE" ]; then
+    if [ -z "${WANDB_API_KEY:-}" ]; then
+      export WANDB_API_KEY
+      WANDB_API_KEY="$($PY_SWEEP_BIN - <<'PY'
+from dotenv import dotenv_values
+print(dotenv_values('.env').get('WANDB_API_KEY', '').strip())
+PY
+)"
+    fi
+    if [ -z "${WANDB_API_KEY:-}" ]; then
+      echo "WANDB_API_KEY is required to create a sweep." >&2
+      exit 1
+    fi
+    SWEEP_ID_VALUE="$($PY_SWEEP_BIN "$ROOT/scripts/wandb_create_sweep.py" \
+      --kind "$SWEEP_KIND" \
+      --project "$SWEEP_PROJECT" \
+      --entity "$SWEEP_ENTITY" \
+      --method "$SWEEP_METHOD" \
+      --run-cap "$SWEEP_RUN_CAP")"
+  fi
+
+  SWEEP_ENTRY_ARGS="${SWEEP_ENTRY_ARGS:-}"
+  if [ -z "$SWEEP_ENTRY_ARGS" ]; then
+    SWEEP_ENTRY_ARGS="--sweep-agent --sweep-id $SWEEP_ID_VALUE --project $SWEEP_PROJECT --device cpu"
+  fi
+
+  if [ "$AGENT_COUNT" = "0" ] && [ "${SWEEP_RUN_CAP:-0}" -gt 0 ]; then
+    TOTAL_AGENTS=$((SWEEP_NUM_NODES * AGENTS_PER_NODE))
+    if [ "$TOTAL_AGENTS" -gt 0 ]; then
+      AGENT_COUNT=$(((SWEEP_RUN_CAP + TOTAL_AGENTS - 1) / TOTAL_AGENTS))
+      echo "Derived AGENT_COUNT=$AGENT_COUNT from SWEEP_RUN_CAP=$SWEEP_RUN_CAP across $TOTAL_AGENTS agents"
+    fi
+  fi
+
+  DIST_ARGS=(
+    python
+    scripts/ray_distributed_train.py
+    --run-kind "$SWEEP_KIND"
+    --entry-args "$SWEEP_ENTRY_ARGS"
+    --num-nodes "${SWEEP_NUM_NODES}"
+    --tpu-per-task "${TPU_PER_TASK:-0}"
+    --base-seed "${BASE_SEED:-42}"
+    --wandb-entity "$SWEEP_ENTITY"
+    --wandb-project "$SWEEP_PROJECT"
+    --agents-per-node "$AGENTS_PER_NODE"
+    --agent-count "$AGENT_COUNT"
+    --inner-threads "$INNER_THREADS"
+    --worker-cpus "${WORKER_CPUS:-$((AGENTS_PER_NODE * INNER_THREADS))}"
+  )
+  if [ "$SWEEP_KIND" = "benchmark" ]; then
+    DIST_ARGS+=(--output-root "${OUTPUT_ROOT:-engine/studies/results/sweeps}")
+  fi
+  if [ "${COMPARE_ROBUST:-0}" = "1" ]; then
+    DIST_ARGS+=(--compare-robust)
+  fi
+  echo "SWEEP_ID=$SWEEP_ID_VALUE"
+  "$RAY_BIN" "${COMMON_ARGS[@]}" "${DIST_ARGS[@]}"
+  exit 0
+fi
+
+echo "Unsupported RAY_MODE='$RAY_MODE' (expected 'single', 'distributed', 'benchmark', or 'sweep')." >&2
 exit 1

tpu_orchestration/watchdog.sh

@@ -97,6 +97,8 @@ while true; do
     # Determine runtime version
     RT_VERSION=${RUNTIME_VERSION:-"tpu-ubuntu2204-base"}
 
+    CREATE_LOG="/tmp/tpu_create_${QR_NAME}.log"
+
     gcloud compute tpus queued-resources create $QR_NAME \
       --project=$PROJECT_ID \
       --node-id=$QR_NAME \
@@ -104,20 +106,23 @@ while true; do
       --accelerator-type=$ACCEL_TYPE \
       --runtime-version=$RT_VERSION \
       $SPOT_FLAG \
+      --internal-ips \
       --metadata-from-file startup-script=$(dirname $0)/tpu_startup.sh \
-      --metadata "$METADATA" 2>&1 | tee /tmp/tpu_create_${QR_NAME}.log
+      --metadata "$METADATA" 2>&1 | tee "$CREATE_LOG"
 
-    if [ $? -eq 0 ]; then
+    CREATE_EXIT=${PIPESTATUS[0]}
+      
+    if [ $CREATE_EXIT -eq 0 ]; then
         echo "[$(date)] Successfully queued $QR_NAME."
         RETRY_DELAY=60
-    elif grep -q "IN_USE_ADDRESSES" /tmp/tpu_create_${QR_NAME}.log 2>/dev/null; then
+    elif grep -q "IN_USE_ADDRESSES" "$CREATE_LOG" 2>/dev/null; then
         echo "[$(date)] IP quota hit - backing off ${RETRY_DELAY}s"
         sleep $RETRY_DELAY
         RETRY_DELAY=$((RETRY_DELAY * 2))
         [ $RETRY_DELAY -gt $MAX_RETRY_DELAY ] && RETRY_DELAY=$MAX_RETRY_DELAY
         continue
     else
-        echo "[$(date)] Failed to queue $QR_NAME."
+        echo "[$(date)] Failed to queue $QR_NAME (exit=$CREATE_EXIT)."
         RETRY_DELAY=60
     fi
   else