fix: keep TRUST_REMOTE_CODE=True for built-in models

The constant is the default for built-in models with known model cards,
which are trusted. Custom models added via API already default to
trust_remote_code=False in ModelCard.fetch_from_hf(). The CLI flag
overrides custom models only.

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

bench/exo_bench.py

@@ -75,7 +75,7 @@ def load_tokenizer_for_bench(model_id: str) -> Any:
         model_path = Path(
             snapshot_download(
                 model_id,
-                allow_patterns=["*.json", "*.py", "*.tiktoken"],
+                allow_patterns=["*.json", "*.py", "*.tiktoken", "*.model"],
             )
         )
 

dashboard/src/lib/components/HeaderNav.svelte

@@ -95,7 +95,7 @@
     {#if showHome}
       <button
         onclick={handleHome}
-        class="text-sm text-exo-light-gray hover:text-exo-yellow transition-colors tracking-wider uppercase flex items-center gap-2 cursor-pointer"
+        class="text-sm text-white/70 hover:text-exo-yellow transition-colors tracking-wider uppercase flex items-center gap-2 cursor-pointer"
         title="Back to topology view"
       >
         <svg
@@ -116,7 +116,7 @@
     {/if}
     <a
       href="/#/downloads"
-      class="text-sm text-exo-light-gray hover:text-exo-yellow transition-colors tracking-wider uppercase flex items-center gap-2 cursor-pointer"
+      class="text-sm text-white/70 hover:text-exo-yellow transition-colors tracking-wider uppercase flex items-center gap-2 cursor-pointer"
       title="View downloads overview"
     >
       {#if downloadProgress}

dashboard/src/routes/downloads/+page.svelte

@@ -412,7 +412,7 @@
                   <div>{col.label}</div>
                   {#if col.diskAvailable != null}
                     <div
-                      class="text-[9px] text-exo-light-gray/60 normal-case tracking-normal mt-0.5"
+                      class="text-[9px] text-white/70 normal-case tracking-normal mt-0.5"
                     >
                       {formatBytes(col.diskAvailable)} free
                     </div>
@@ -436,7 +436,7 @@
                       </div>
                       {#if row.prettyName}
                         <div
-                          class="text-[10px] text-exo-light-gray/60"
+                          class="text-[10px] text-white/60"
                           title={row.modelId}
                         >
                           {row.modelId}
@@ -450,7 +450,7 @@
                       title="View model details"
                     >
                       <svg
-                        class="w-4 h-4 text-white/30 hover:text-white/60"
+                        class="w-4 h-4 text-white/60 hover:text-white/80"
                         viewBox="0 0 24 24"
                         fill="currentColor"
                       >
@@ -483,12 +483,12 @@
                             clip-rule="evenodd"
                           ></path>
                         </svg>
-                        <span class="text-xs text-exo-light-gray/70"
+                        <span class="text-xs text-white/70"
                           >{formatBytes(cell.totalBytes)}</span
                         >
                         <button
                           type="button"
-                          class="text-exo-light-gray/40 hover:text-red-400 transition-colors mt-0.5 cursor-pointer"
+                          class="text-white/50 hover:text-red-400 transition-colors mt-0.5 cursor-pointer"
                           onclick={() =>
                             deleteDownload(col.nodeId, row.modelId)}
                           title="Delete from this node"
@@ -530,7 +530,7 @@
                             ).toFixed(1)}%"
                           ></div>
                         </div>
-                        <span class="text-[10px] text-exo-light-gray/60"
+                        <span class="text-[10px] text-white/70"
                           >{formatSpeed(cell.speed)}</span
                         >
                       </div>
@@ -542,7 +542,7 @@
                           : "Download pending"}
                       >
                         {#if cell.downloaded > 0 && cell.total > 0}
-                          <span class="text-exo-light-gray/70 text-xs"
+                          <span class="text-white/70 text-xs"
                             >{formatBytes(cell.downloaded)} / {formatBytes(
                               cell.total,
                             )}</span
@@ -561,7 +561,7 @@
                           {#if row.shardMetadata}
                             <button
                               type="button"
-                              class="text-exo-light-gray/50 hover:text-exo-yellow transition-colors cursor-pointer"
+                              class="text-white/50 hover:text-exo-yellow transition-colors cursor-pointer"
                               onclick={() =>
                                 startDownload(col.nodeId, row.shardMetadata!)}
                               title="Resume download on this node"
@@ -581,14 +581,13 @@
                               </svg>
                             </button>
                           {:else}
-                            <span class="text-exo-light-gray/40 text-[10px]"
-                              >paused</span
+                            <span class="text-white/50 text-[10px]">paused</span
                             >
                           {/if}
                         {:else if row.shardMetadata}
                           <button
                             type="button"
-                            class="text-exo-light-gray/30 hover:text-exo-yellow transition-colors cursor-pointer"
+                            class="text-white/50 hover:text-exo-yellow transition-colors cursor-pointer"
                             onclick={() =>
                               startDownload(col.nodeId, row.shardMetadata!)}
                             title="Start download on this node"
@@ -608,8 +607,7 @@
                             </svg>
                           </button>
                         {:else}
-                          <span class="text-exo-light-gray/50 text-sm">...</span
-                          >
+                          <span class="text-white/40 text-sm">...</span>
                         {/if}
                       </div>
                     {:else if cell.kind === "failed"}
@@ -631,7 +629,7 @@
                         {#if row.shardMetadata}
                           <button
                             type="button"
-                            class="text-exo-light-gray/40 hover:text-exo-yellow transition-colors cursor-pointer"
+                            class="text-white/50 hover:text-exo-yellow transition-colors cursor-pointer"
                             onclick={() =>
                               startDownload(col.nodeId, row.shardMetadata!)}
                             title="Retry download on this node"
@@ -663,7 +661,7 @@
                         {#if row.shardMetadata}
                           <button
                             type="button"
-                            class="text-exo-light-gray/30 hover:text-exo-yellow transition-colors mt-0.5 opacity-0 group-hover:opacity-100 cursor-pointer"
+                            class="text-white/50 hover:text-exo-yellow transition-colors mt-0.5 opacity-0 group-hover:opacity-100 cursor-pointer"
                             onclick={() =>
                               startDownload(col.nodeId, row.shardMetadata!)}
                             title="Download to this node"

src/exo/main.py

@@ -261,6 +261,13 @@ def main():
     if args.offline:
         logger.info("Running in OFFLINE mode — no internet checks, local models only")
 
+    # Set trust_remote_code override env var for runner subprocesses
+    if args.trust_remote_code:
+        os.environ["EXO_TRUST_REMOTE_CODE"] = "1"
+        logger.warning(
+            "--trust-remote-code enabled: models may execute arbitrary code during loading"
+        )
+
     # Set FAST_SYNCH override env var for runner subprocesses
     if args.fast_synch is True:
         os.environ["EXO_FAST_SYNCH"] = "on"
@@ -285,6 +292,7 @@ class Args(CamelCaseModel):
     no_downloads: bool = False
     offline: bool = False
     fast_synch: bool | None = None  # None = auto, True = force on, False = force off
+    trust_remote_code: bool = False
 
     @classmethod
     def parse(cls) -> Self:
@@ -336,6 +344,11 @@ class Args(CamelCaseModel):
             action="store_true",
             help="Run in offline/air-gapped mode: skip internet checks, use only pre-staged local models",
         )
+        parser.add_argument(
+            "--trust-remote-code",
+            action="store_true",
+            help="Allow models to execute custom code during tokenizer loading (security-sensitive, CLI-only)",
+        )
         fast_synch_group = parser.add_mutually_exclusive_group()
         fast_synch_group.add_argument(
             "--fast-synch",

src/exo/master/tests/test_placement.py

@@ -14,10 +14,12 @@ from exo.shared.models.model_cards import ModelCard, ModelId, ModelTask
 from exo.shared.topology import Topology
 from exo.shared.types.commands import PlaceInstance
 from exo.shared.types.common import CommandId, NodeId
-from exo.shared.types.events import InstanceCreated, InstanceDeleted
+from exo.shared.types.events import InstanceCreated, InstanceDeleted, TaskStatusUpdated
 from exo.shared.types.memory import Memory
 from exo.shared.types.multiaddr import Multiaddr
 from exo.shared.types.profiling import NetworkInterfaceInfo, NodeNetworkInfo
+from exo.shared.types.tasks import TaskId, TaskStatus, TextGeneration
+from exo.shared.types.text_generation import InputMessage, TextGenerationTaskParams
 from exo.shared.types.topology import Connection, SocketConnection
 from exo.shared.types.worker.instances import (
     Instance,
@@ -456,3 +458,117 @@ def test_tensor_rdma_backend_connectivity_matrix(
         else:
             ip_part = coordinator.split(":")[0]
             assert len(ip_part.split(".")) == 4
+
+
+def _make_task(
+    instance_id: InstanceId,
+    status: TaskStatus = TaskStatus.Running,
+) -> TextGeneration:
+    return TextGeneration(
+        task_id=TaskId(),
+        task_status=status,
+        instance_id=instance_id,
+        command_id=CommandId(),
+        task_params=TextGenerationTaskParams(
+            model=ModelId("test-model"),
+            input=[InputMessage(role="user", content="hello")],
+        ),
+    )
+
+
+def test_get_transition_events_delete_instance_cancels_running_tasks(
+    instance: Instance,
+):
+    # arrange
+    instance_id = InstanceId()
+    current_instances: dict[InstanceId, Instance] = {instance_id: instance}
+    target_instances: dict[InstanceId, Instance] = {}
+    task = _make_task(instance_id, TaskStatus.Running)
+    tasks = {task.task_id: task}
+
+    # act
+    events = get_transition_events(current_instances, target_instances, tasks)
+
+    # assert – cancellation event should come before the deletion event
+    assert len(events) == 2
+    assert isinstance(events[0], TaskStatusUpdated)
+    assert events[0].task_id == task.task_id
+    assert events[0].task_status == TaskStatus.Cancelled
+    assert isinstance(events[1], InstanceDeleted)
+    assert events[1].instance_id == instance_id
+
+
+def test_get_transition_events_delete_instance_cancels_pending_tasks(
+    instance: Instance,
+):
+    # arrange
+    instance_id = InstanceId()
+    current_instances: dict[InstanceId, Instance] = {instance_id: instance}
+    target_instances: dict[InstanceId, Instance] = {}
+    task = _make_task(instance_id, TaskStatus.Pending)
+    tasks = {task.task_id: task}
+
+    # act
+    events = get_transition_events(current_instances, target_instances, tasks)
+
+    # assert
+    assert len(events) == 2
+    assert isinstance(events[0], TaskStatusUpdated)
+    assert events[0].task_id == task.task_id
+    assert events[0].task_status == TaskStatus.Cancelled
+    assert isinstance(events[1], InstanceDeleted)
+
+
+def test_get_transition_events_delete_instance_ignores_completed_tasks(
+    instance: Instance,
+):
+    # arrange
+    instance_id = InstanceId()
+    current_instances: dict[InstanceId, Instance] = {instance_id: instance}
+    target_instances: dict[InstanceId, Instance] = {}
+    tasks = {
+        t.task_id: t
+        for t in [
+            _make_task(instance_id, TaskStatus.Complete),
+            _make_task(instance_id, TaskStatus.Failed),
+            _make_task(instance_id, TaskStatus.TimedOut),
+            _make_task(instance_id, TaskStatus.Cancelled),
+        ]
+    }
+
+    # act
+    events = get_transition_events(current_instances, target_instances, tasks)
+
+    # assert – only the InstanceDeleted event, no cancellations
+    assert len(events) == 1
+    assert isinstance(events[0], InstanceDeleted)
+
+
+def test_get_transition_events_delete_instance_cancels_only_matching_tasks(
+    instance: Instance,
+):
+    # arrange
+    instance_id_a = InstanceId()
+    instance_id_b = InstanceId()
+    current_instances: dict[InstanceId, Instance] = {
+        instance_id_a: instance,
+        instance_id_b: instance,
+    }
+    # only delete instance A, keep instance B
+    target_instances: dict[InstanceId, Instance] = {instance_id_b: instance}
+
+    task_a = _make_task(instance_id_a, TaskStatus.Running)
+    task_b = _make_task(instance_id_b, TaskStatus.Running)
+    tasks = {task_a.task_id: task_a, task_b.task_id: task_b}
+
+    # act
+    events = get_transition_events(current_instances, target_instances, tasks)
+
+    # assert – only task_a should be cancelled
+    cancel_events = [e for e in events if isinstance(e, TaskStatusUpdated)]
+    delete_events = [e for e in events if isinstance(e, InstanceDeleted)]
+    assert len(cancel_events) == 1
+    assert cancel_events[0].task_id == task_a.task_id
+    assert cancel_events[0].task_status == TaskStatus.Cancelled
+    assert len(delete_events) == 1
+    assert delete_events[0].instance_id == instance_id_a

src/exo/worker/engines/mlx/auto_parallel.py

@@ -852,6 +852,8 @@ class QwenShardingStrategy(TensorParallelShardingStrategy):
                 layer.self_attn.o_proj = self.sharded_to_all_linear(
                     layer.self_attn.o_proj
                 )
+                layer.self_attn.n_heads //= self.N
+                layer.self_attn.n_kv_heads //= self.N
             else:
                 assert isinstance(layer, Qwen3NextDecoderLayer)
                 if hasattr(layer, "linear_attn"):

src/exo/worker/engines/mlx/constants.py

@@ -13,5 +13,6 @@ KV_CACHE_BITS: int | None = None
 
 DEFAULT_TOP_LOGPROBS: int = 5
 
-# TODO: We should really make this opt-in, but Kimi requires trust_remote_code=True
+# True for built-in models with known model cards; custom models added via API default to False
+# and can be overridden with the --trust-remote-code CLI flag.
 TRUST_REMOTE_CODE: bool = True

src/exo/worker/engines/mlx/utils_mlx.py

@@ -291,10 +291,14 @@ def shard_and_load(
 
 def get_tokenizer(model_path: Path, shard_metadata: ShardMetadata) -> TokenizerWrapper:
     """Load tokenizer for a model shard. Delegates to load_tokenizer_for_model_id."""
+    trust_remote_code = (
+        shard_metadata.model_card.trust_remote_code
+        or os.environ.get("EXO_TRUST_REMOTE_CODE") == "1"
+    )
     return load_tokenizer_for_model_id(
         shard_metadata.model_card.model_id,
         model_path,
-        trust_remote_code=shard_metadata.model_card.trust_remote_code,
+        trust_remote_code=trust_remote_code,
     )
 
 

tests/get_all_models_on_cluster.py

@@ -8,7 +8,7 @@ from urllib.request import urlopen
 
 h = sys.argv[1] if len(sys.argv) > 1 else sys.exit(f"USAGE: {sys.argv[0]} host")
 ts = subprocess.run(
-    ["/Applications/Tailscale.app/Contents/MacOS/Tailscale", "status"], check=True, text=True, capture_output=True
+    ["tailscale", "status"], check=True, text=True, capture_output=True
 ).stdout.splitlines()
 ip = next(
     (sl[0] for line in ts if len(sl := line.split()) >= 2 if sl[1] == h), None

tests/start_distributed_test.py

@@ -15,7 +15,7 @@ if not (args := sys.argv[1:]):
 kind = args[0] if args[0] in ("jaccl", "ring") else "both"
 hosts = args[1:] if kind != "both" else args
 ts = subprocess.run(
-    ["/Applications/Tailscale.app/Contents/MacOS/Tailscale", "status"], check=True, text=True, capture_output=True
+    ["tailscale", "status"], check=True, text=True, capture_output=True
 ).stdout.splitlines()
 ip = {sl[1]: sl[0] for line in ts if len(sl := line.split()) >= 2}
 ips = [ip[h] for h in hosts]