From 2fbec025f6f8583737e89d8f62ac2a177c646dc1 Mon Sep 17 00:00:00 2001
From: flujo-app <300233937+flujo-app@users.noreply.github.com>
Date: Thu, 12 Nov 2026 07:17:18 -0500
Subject: [PATCH] Self-hosted models with remote code: code and tokenizer from
 the weights' pinned revision; release 0.7.3

---
 package-lock.json               | 4 ++--
 package.json                    | 2 +-
 selfhost/modal/settings.py      | 4 +++-
 selfhost/modal/test_settings.py | 4 ++++
 4 files changed, 10 insertions(+), 4 deletions(-)

diff --git a/package-lock.json b/package-lock.json
index e686ac0..3f0054a 100644
--- a/package-lock.json
+++ b/package-lock.json
@@ -1,12 +1,12 @@
 {
   "name": "flujo-factory",
-  "version": "0.7.2",
+  "version": "0.7.3",
   "lockfileVersion": 3,
   "requires": true,
   "packages": {
     "": {
       "name": "flujo-factory",
-      "version": "0.7.2",
+      "version": "0.7.3",
       "license": "MIT",
       "dependencies": {
         "@modelcontextprotocol/sdk": "1.32.1"
diff --git a/package.json b/package.json
index 7fb1cb5..6c7c9bc 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,6 @@
 {
   "name": "flujo-factory",
-  "version": "0.7.2",
+  "version": "0.7.3",
   "description": "Durable FLUJO swarm engine and Factory SDK, CLI, and MCP server",
   "license": "MIT",
   "repository": {
diff --git a/selfhost/modal/settings.py b/selfhost/modal/settings.py
index bcdc915..49b5549 100644
--- a/selfhost/modal/settings.py
+++ b/selfhost/modal/settings.py
@@ -167,7 +167,9 @@ def vllm_command(self):
                    '--reasoning-parser', self.reasoning_parser,
                    '--default-chat-template-kwargs', json.dumps(self.chat_template_kwargs())]
         if self.trust_remote_code:
-            command += ['--trust-remote-code']
+            # The model's own code and tokenizer come from the same pinned revision as its weights; without these flags
+            # the Hub serves the default branch's code files (transformers warns "A new version ... was downloaded").
+            command += ['--trust-remote-code', '--code-revision', self.revision, '--tokenizer-revision', self.revision]
         if self.gpu_count > 1:
             command += ['--tensor-parallel-size', str(self.gpu_count)]
         return command + list(self.extra_args) + self.variant_args()
diff --git a/selfhost/modal/test_settings.py b/selfhost/modal/test_settings.py
index cc26bf9..ebd003b 100644
--- a/selfhost/modal/test_settings.py
+++ b/selfhost/modal/test_settings.py
@@ -52,6 +52,10 @@ def test_nemotron_command_uses_the_release_image_mtp_and_hybrid_prefix_caching(s
             self.assertEqual(command[command.index(pair[0]) + 1], pair[1])
         for flag in ('--trust-remote-code', '--language-model-only', '--async-scheduling', '--enable-mamba-shared-prefix-checkpoint'):
             self.assertIn(flag, command)
+        # remote code and tokenizer from the same pinned revision as the weights
+        for flag in ('--revision', '--code-revision', '--tokenizer-revision'):
+            self.assertEqual(command[command.index(flag) + 1], s.revision, flag)
+        self.assertNotIn('--code-revision', resolve(QWEN, 'qwen').vllm_command(), 'no remote code, nothing to pin')
         self.assertEqual(json.loads(command[command.index('--speculative-config') + 1]), {'method': 'mtp', 'num_speculative_tokens': 1})
         self.assertEqual(s.env, {'OMP_NUM_THREADS': '1'})
         self.assertEqual(s.input_modalities, ('text',))