From b5030ffb5074219176a62e6edd773af03b09b6a2 Mon Sep 17 00:00:00 2001
From: jan <1760229+slomin@users.noreply.github.com>
Date: Wed, 8 Apr 2026 19:47:44 +0100
Subject: [PATCH] fix(runtime): use generic projector repo for non-ByteShape
 Qwen3.5-9B

Routes non-ByteShape Qwen3.5-9B models to the shared `unsloth/Qwen3.5-9B-GGUF`
projector repo instead of relying on publisher-specific projector rules.

Bumps the pinned `potato-inferno` dependency to the revision that applies
the generic 9B mapping and adds API coverage proving `_runtime_env`
publishes the expected projector repository for supported 9B models.

Closes #263
---
 requirements.txt                     |  2 +-
 tests/api/test_model_settings.py     |  6 ++--
 tests/api/test_projector_download.py |  2 +-
 tests/api/test_runtime_env.py        | 44 ++++++++++++++++++++++++++++
 4 files changed, 49 insertions(+), 5 deletions(-)

diff --git a/requirements.txt b/requirements.txt
index 009bdb2..6f0c9fd 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -3,4 +3,4 @@ uvicorn[standard]>=0.30.0,<1.0.0
 httpx>=0.27.0,<1.0.0
 psutil>=6.0.0,<7.0.0
 PyYAML>=6.0.2,<7.0.0
-potato-inferno @ git+https://github.com/potato-os/inferno.git@5f0b6a0
+potato-inferno @ git+https://github.com/potato-os/inferno.git@2785766
diff --git a/tests/api/test_model_settings.py b/tests/api/test_model_settings.py
index b16b3c0..49315d2 100644
--- a/tests/api/test_model_settings.py
+++ b/tests/api/test_model_settings.py
@@ -20,7 +20,7 @@ def test_update_model_settings_persists_per_model_chat_and_vision(runtime):
     with TestClient(app) as client:
         upload = client.post(
             "/internal/models/upload",
-            headers={"x-potato-filename": "Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf"},
+            headers={"x-potato-filename": "Qwen3.5-4B-Q4_K_M.gguf"},
             content=b"gguf",
         )
         assert upload.status_code == 200
@@ -74,7 +74,7 @@ def test_settings_document_yaml_round_trip_updates_active_model_and_model_settin
     with TestClient(app) as client:
         upload = client.post(
             "/internal/models/upload",
-            headers={"x-potato-filename": "Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf"},
+            headers={"x-potato-filename": "Qwen3.5-4B-Q4_K_M.gguf"},
             content=b"gguf",
         )
         assert upload.status_code == 200
@@ -155,7 +155,7 @@ def test_update_model_settings_rejects_invalid_numeric_value(runtime):
     with TestClient(app) as client:
         upload = client.post(
             "/internal/models/upload",
-            headers={"x-potato-filename": "Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf"},
+            headers={"x-potato-filename": "Qwen3.5-4B-Q4_K_M.gguf"},
             content=b"gguf",
         )
         assert upload.status_code == 200
diff --git a/tests/api/test_projector_download.py b/tests/api/test_projector_download.py
index d22fd61..294b619 100644
--- a/tests/api/test_projector_download.py
+++ b/tests/api/test_projector_download.py
@@ -28,7 +28,7 @@ def _fake_download(*, runtime, model_id: str):
     with TestClient(app) as client:
         upload = client.post(
             "/internal/models/upload",
-            headers={"x-potato-filename": "Qwen3.5-4B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf"},
+            headers={"x-potato-filename": "Qwen3.5-4B-Q4_K_M.gguf"},
             content=b"gguf",
         )
         assert upload.status_code == 200
diff --git a/tests/api/test_runtime_env.py b/tests/api/test_runtime_env.py
index f91bfef..68fad41 100644
--- a/tests/api/test_runtime_env.py
+++ b/tests/api/test_runtime_env.py
@@ -263,6 +263,50 @@ def test_runtime_env_enables_mmproj_auto_download_when_vision_on_and_no_mmproj(r
     assert "unsloth" in env["POTATO_HF_MMPROJ_REPO"] or "huggingface" in env["POTATO_HF_MMPROJ_REPO"].lower()
 
 
+def test_runtime_env_resolves_9b_non_byteshape_to_unsloth_repo(runtime):
+    """Non-ByteShape Qwen3.5-9B must resolve to unsloth/Qwen3.5-9B-GGUF."""
+    model_filename = "Qwen3.5-9B-Q4_K_M.gguf"
+    model_path = runtime.base_dir / "models" / model_filename
+    model_path.write_bytes(b"gguf")
+    runtime.model_path = model_path
+    runtime.models_state_path.write_text(
+        json.dumps(
+            {
+                "version": 1,
+                "countdown_enabled": True,
+                "default_model_downloaded_once": True,
+                "active_model_id": "default",
+                "default_model_id": "default",
+                "current_download_model_id": None,
+                "models": [
+                    {
+                        "id": "default",
+                        "filename": model_filename,
+                        "source_url": "https://example.com/qwen35-9b.gguf",
+                        "source_type": "url",
+                        "status": "ready",
+                        "error": None,
+                        "settings": {
+                            "vision": {
+                                "enabled": True,
+                                "projector_mode": "default",
+                                "projector_filename": None,
+                            }
+                        },
+                    }
+                ],
+            }
+        ),
+        encoding="utf-8",
+    )
+
+    env = _runtime_env(runtime)
+
+    assert env["POTATO_VISION_MODEL_NAME_PATTERN_QWEN35"] == "1"
+    assert env["POTATO_AUTO_DOWNLOAD_MMPROJ"] == "1"
+    assert env["POTATO_HF_MMPROJ_REPO"] == "unsloth/Qwen3.5-9B-GGUF"
+
+
 def test_projector_status_reports_generic_truthfully_for_offline_compat(runtime):
     """build_model_projector_status must report mmproj-F16.gguf as present even
     when model-specific candidates exist — preserves offline upgrades where the