From 690ad46d88c0444dfcc5766d403a8c383b09ca27fd90697cba1e0cf5915a37f7 Mon Sep 17 00:00:00 2001
From: tlg <thomas.langer@destengs.com>
Date: Sat, 4 Apr 2026 07:30:13 +0200
Subject: [PATCH] feat: config loading for models.yaml and api_keys.yaml

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
---
 kischdle/llmux/llmux/config.py      | 79 +++++++++++++++++++++++++++++
 kischdle/llmux/tests/test_config.py | 56 ++++++++++++++++++++
 2 files changed, 135 insertions(+)
 create mode 100644 kischdle/llmux/llmux/config.py
 create mode 100644 kischdle/llmux/tests/test_config.py

diff --git a/kischdle/llmux/llmux/config.py b/kischdle/llmux/llmux/config.py
new file mode 100644
index 0000000..e42a066
--- /dev/null
+++ b/kischdle/llmux/llmux/config.py
@@ -0,0 +1,79 @@
+import os
+from dataclasses import dataclass, field
+from pathlib import Path
+
+import yaml
+
+
+def _config_dir() -> Path:
+    return Path(os.environ.get("LLMUX_CONFIG_DIR", "/config"))
+
+
+@dataclass
+class PhysicalModel:
+    type: str  # "llm", "asr", "tts"
+    backend: str  # "transformers", "llamacpp", "chatterbox"
+    estimated_vram_gb: float
+    model_id: str = ""
+    model_file: str = ""
+    mmproj_file: str = ""
+    supports_vision: bool = False
+    supports_tools: bool = False
+    default_language: str = ""
+    variant: str = ""
+
+
+@dataclass
+class VirtualModel:
+    physical: str
+    params: dict = field(default_factory=dict)
+
+
+@dataclass
+class ApiKey:
+    key: str
+    name: str
+
+
+def load_models_config(
+    config_path: Path | None = None,
+) -> tuple[dict[str, PhysicalModel], dict[str, VirtualModel]]:
+    if config_path is None:
+        config_path = _config_dir() / "models.yaml"
+
+    with open(config_path) as f:
+        raw = yaml.safe_load(f)
+
+    physical: dict[str, PhysicalModel] = {}
+    for model_id, attrs in raw["physical_models"].items():
+        physical[model_id] = PhysicalModel(
+            type=attrs["type"],
+            backend=attrs["backend"],
+            estimated_vram_gb=attrs["estimated_vram_gb"],
+            model_id=attrs.get("model_id", ""),
+            model_file=attrs.get("model_file", ""),
+            mmproj_file=attrs.get("mmproj_file", ""),
+            supports_vision=attrs.get("supports_vision", False),
+            supports_tools=attrs.get("supports_tools", False),
+            default_language=attrs.get("default_language", ""),
+            variant=attrs.get("variant", ""),
+        )
+
+    virtual: dict[str, VirtualModel] = {}
+    for model_name, attrs in raw["virtual_models"].items():
+        virtual[model_name] = VirtualModel(
+            physical=attrs["physical"],
+            params=attrs.get("params", {}),
+        )
+
+    return physical, virtual
+
+
+def load_api_keys(config_path: Path | None = None) -> list[ApiKey]:
+    if config_path is None:
+        config_path = _config_dir() / "api_keys.yaml"
+
+    with open(config_path) as f:
+        raw = yaml.safe_load(f)
+
+    return [ApiKey(key=entry["key"], name=entry["name"]) for entry in raw["api_keys"]]
diff --git a/kischdle/llmux/tests/test_config.py b/kischdle/llmux/tests/test_config.py
new file mode 100644
index 0000000..ab807e7
--- /dev/null
+++ b/kischdle/llmux/tests/test_config.py
@@ -0,0 +1,56 @@
+from llmux.config import load_models_config, load_api_keys, PhysicalModel, VirtualModel
+
+
+def test_load_models_config_returns_physical_and_virtual():
+    physical, virtual = load_models_config()
+    assert isinstance(physical, dict)
+    assert isinstance(virtual, dict)
+    assert len(physical) == 9
+    assert len(virtual) == 16
+
+
+def test_physical_model_has_required_fields():
+    physical, _ = load_models_config()
+    qwen = physical["qwen3.5-9b-fp8"]
+    assert qwen.type == "llm"
+    assert qwen.backend == "transformers"
+    assert qwen.model_id == "lovedheart/Qwen3.5-9B-FP8"
+    assert qwen.estimated_vram_gb == 9
+    assert qwen.supports_vision is True
+    assert qwen.supports_tools is True
+
+
+def test_physical_model_llamacpp_has_gguf_fields():
+    physical, _ = load_models_config()
+    uncensored = physical["qwen3.5-9b-fp8-uncensored"]
+    assert uncensored.backend == "llamacpp"
+    assert uncensored.model_file == "Qwen3.5-9B-Uncensored-HauhauCS-Aggressive-Q8_0.gguf"
+    assert uncensored.mmproj_file == "mmproj-Qwen3.5-9B-Uncensored-HauhauCS-Aggressive-BF16.gguf"
+
+
+def test_virtual_model_maps_to_physical():
+    _, virtual = load_models_config()
+    thinking = virtual["Qwen3.5-9B-FP8-Thinking"]
+    assert thinking.physical == "qwen3.5-9b-fp8"
+    assert thinking.params == {"enable_thinking": True}
+
+
+def test_virtual_model_gpt_oss_has_system_prompt():
+    _, virtual = load_models_config()
+    low = virtual["GPT-OSS-20B-Low"]
+    assert low.physical == "gpt-oss-20b"
+    assert low.params == {"system_prompt_prefix": "Reasoning: low"}
+
+
+def test_virtual_model_without_params():
+    _, virtual = load_models_config()
+    ct = virtual["cohere-transcribe"]
+    assert ct.physical == "cohere-transcribe"
+    assert ct.params == {}
+
+
+def test_load_api_keys():
+    keys = load_api_keys()
+    assert len(keys) == 3
+    assert all(k.key.startswith("sk-llmux-") for k in keys)
+    assert {k.name for k in keys} == {"Open WebUI", "Remote Whisper clients", "OpenCode"}