more working models

utils/constants.py

@@ -8,6 +8,18 @@ MODELS = [
         "is_thinking": True,
         "supports_reasoning_level": True,
     },
+    {
+        "name": "GPT-OSS-20B",
+        "id": "openai/gpt-oss-20b",
+        "is_thinking": True,
+        "supports_reasoning_level": True,
+    },
+    {
+        "name": "GPT-OSS-120B",
+        "id": "openai/gpt-oss-120b",
+        "is_thinking": True,
+        "supports_reasoning_level": True,
+    },
     {
         "name": "Qwen3-Next-80B-Instruct",
         "id": "Qwen/Qwen3-Next-80B-A3B-Instruct",
@@ -20,6 +32,24 @@ MODELS = [
         "is_thinking": True,
         "supports_reasoning_level": False,
     },
+    {
+        "name": "Qwen3-4B-Instruct",
+        "id": "Qwen/Qwen3-4B-Instruct-2507",
+        "is_thinking": False,
+        "supports_reasoning_level": False,
+    },
+    {
+        "name": "Gemma-3-12B-Instruct",
+        "id": "google/gemma-3-12b-it",
+        "is_thinking": False,
+        "supports_reasoning_level": False,
+    },
+    {
+        "name": "Gemma-3-27B-Instruct",
+        "id": "google/gemma-3-27b-it",
+        "is_thinking": False,
+        "supports_reasoning_level": False,
+    },
 ]
 
 UNUSED_MODELS = [

utils/model_interface.py

@@ -62,7 +62,7 @@ def get_default_system_prompt(model_id: str, reasoning_effort: str = "Low") -> s
 
 def make_messages(test: str, policy: str, model_id: str, reasoning_effort: str = "Low", system_prompt: str | None = None, response_format: str = RESPONSE_FORMAT) -> list[dict]:
     """Create messages based on model type."""
-    if is_gptoss_model(model_id):
+    if model_id.startswith("openai/gpt-oss-safeguard"):
         # GPT-OSS uses Harmony encoding
         enc = load_harmony_encoding(HarmonyEncodingName.HARMONY_GPT_OSS)
         system_content = SystemContent.new().with_reasoning_effort(reasoning_effort)
@@ -123,6 +123,7 @@ def run_test(
         temperature=temperature,
         top_p=top_p,
         stop=None,
+        # extra_headers={"X-HF-Bill-To": "roosttools"},
     )
 
     result = {"content": completion.choices[0].message.content}