Leave GPU region unpinned by default

docs/deployment-configs.md

@@ -110,12 +110,12 @@ Validation loads the recipes and checks that they can share one frontend. Backen
 The checked-in [deploy_models.sh](../scripts/deploy_models.sh) lists the complete active config set. Add a config path there, then run it. The deployment command owns frontend selection and trainer/inference updates:
 
 ~~~bash
-./scripts/deploy_models.sh --app my-spindle --env dev --region us-west
+./scripts/deploy_models.sh --app my-spindle --env dev
 ./scripts/deploy_models.sh --refresh-trainer qwen35-9b-lora-16k
 ./scripts/deploy_models.sh --refresh-inference qwen35-9b-lora-16k
 ~~~
 
-Frontend, region, environment, secret names, and volume names default in `BaseConfig.platform`. Set them in the same Python config using dotted overrides:
+Frontend, region, environment, secret names, and volume names default in `BaseConfig.platform`. `platform.modal.region` defaults to `None`, which leaves GPU placement unpinned (Modal schedules trainers and inference pools in any region); set it only to restrict placement. Set them in the same Python config using dotted overrides:
 
 ```python
 from spindle.configs.gpt_oss_20b_lora_64k import Config as Parent

src/spindle/configuration.py

@@ -15,7 +15,7 @@ def _apply_overrides(values, overrides):
 class BaseConfig:
     platform = {
         "frontend": "spindle",
-        "modal": {"environment": None, "region": "us-west"},
+        "modal": {"environment": None, "region": None},
         "secrets": {
             "api": "spindle-api",
             "sampler_proxy": "spindle-proxy",

src/spindle/providers/modal/deployment_apps.py

@@ -82,6 +82,12 @@ def secrets_for(platform, *, training=False):
     return result
 
 
+def server_placement(region):
+    if region is None:
+        return {}
+    return {"routing_region": region, "compute_region": region}
+
+
 def deployment_env(values):
     """Keep user environment overrides separate from Spindle's deployment wiring."""
     if any(key.startswith("SPINDLE_") for key in values):
@@ -224,8 +230,7 @@ def build_rollout_app(deployment, pool, platform=None, *, image=None):
         startup_timeout=recipe.inference_startup_timeout_s,
         exit_grace_period=300,
         port=8000,
-        routing_region=platform["modal"]["region"],
-        compute_region=platform["modal"]["region"],
+        **server_placement(platform["modal"]["region"]),
     )
     class Server:
         @modal.enter()

src/spindle/providers/modal/scoped.py

@@ -93,7 +93,6 @@ def register_sampler(
         startup_timeout=1800,
         exit_grace_period=120,
         port=8000,
-        routing_region="us-west",
     )
     class Sampler:
         sidecar = None
@@ -417,7 +416,6 @@ async def gateway():
         timeout=1200,
         secrets=[*telemetry_secrets, api_secret],
         volumes={"/checkpoints": checkpoints},
-        routing_region="us-west",
     )
     @modal.concurrent(max_inputs=128)
     @modal.asgi_app(requires_proxy_auth=False)

tests/providers/test_deployment_apps.py

@@ -73,7 +73,7 @@ def test_trainer_declaration_and_executor_configuration(
     app, trainer = deployment_apps.build_trainer_app(row, platform, image=image)
     declaration, _ = app.functions["trainer"]
     assert declaration["gpu"] == "H100:4"
-    assert declaration["region"] == "us-west"
+    assert declaration["region"] is None
     assert declaration["max_containers"] is None
     assert declaration["single_use_containers"] is True
     assert declaration["image"] is image
@@ -127,7 +127,8 @@ def test_pool_starts_native_server_and_correct_sidecar(builders, monkeypatch, ki
     )
     assert settings["min_containers"] == 0
     assert settings["target_concurrency"] == 16
-    assert settings["compute_region"] == "us-west"
+    assert "compute_region" not in settings
+    assert "routing_region" not in settings
 
     calls, commands, stops = [], [], []
     process = object()