tests : use reasoning instead of reasoning_budget in server tests (ggml-org#20432)

Author: pwilkin

tools/server/tests/unit/test_template.py

@@ -11,6 +11,7 @@
 
 import datetime
 from utils import *
+from typing import Literal
 
 server: ServerProcess
 
@@ -23,24 +24,24 @@ def create_server():
 
 
 @pytest.mark.parametrize("tools", [None, [], [TEST_TOOL]])
-@pytest.mark.parametrize("template_name,reasoning_budget,expected_end", [
-    ("deepseek-ai-DeepSeek-R1-Distill-Qwen-32B", None, "<think>\n"),
-    ("deepseek-ai-DeepSeek-R1-Distill-Qwen-32B",   -1, "<think>\n"),
-    ("deepseek-ai-DeepSeek-R1-Distill-Qwen-32B",    0, "<think>\n</think>"),
+@pytest.mark.parametrize("template_name,reasoning,expected_end", [
+    ("deepseek-ai-DeepSeek-R1-Distill-Qwen-32B",  "on", "<think>\n"),
+    ("deepseek-ai-DeepSeek-R1-Distill-Qwen-32B","auto", "<think>\n"),
+    ("deepseek-ai-DeepSeek-R1-Distill-Qwen-32B", "off", "<think>\n</think>"),
 
-    ("Qwen-Qwen3-0.6B", -1, "<|im_start|>assistant\n"),
-    ("Qwen-Qwen3-0.6B",  0, "<|im_start|>assistant\n<think>\n\n</think>\n\n"),
+    ("Qwen-Qwen3-0.6B","auto", "<|im_start|>assistant\n"),
+    ("Qwen-Qwen3-0.6B", "off", "<|im_start|>assistant\n<think>\n\n</think>\n\n"),
 
-    ("Qwen-QwQ-32B", -1, "<|im_start|>assistant\n<think>\n"),
-    ("Qwen-QwQ-32B",  0, "<|im_start|>assistant\n<think>\n</think>"),
+    ("Qwen-QwQ-32B","auto", "<|im_start|>assistant\n<think>\n"),
+    ("Qwen-QwQ-32B", "off", "<|im_start|>assistant\n<think>\n</think>"),
 
-    ("CohereForAI-c4ai-command-r7b-12-2024-tool_use", -1, "<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>"),
-    ("CohereForAI-c4ai-command-r7b-12-2024-tool_use",  0, "<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|><|START_THINKING|><|END_THINKING|>"),
+    ("CohereForAI-c4ai-command-r7b-12-2024-tool_use","auto", "<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|>"),
+    ("CohereForAI-c4ai-command-r7b-12-2024-tool_use", "off", "<|START_OF_TURN_TOKEN|><|CHATBOT_TOKEN|><|START_THINKING|><|END_THINKING|>"),
 ])
-def test_reasoning_budget(template_name: str, reasoning_budget: int | None, expected_end: str, tools: list[dict]):
+def test_reasoning(template_name: str, reasoning: Literal['on', 'off', 'auto'] | None, expected_end: str, tools: list[dict]):
     global server
     server.jinja = True
-    server.reasoning_budget = reasoning_budget
+    server.reasoning = reasoning
     server.chat_template_file = f'../../../models/templates/{template_name}.jinja'
     server.start()
 

tools/server/tests/utils.py

@@ -95,7 +95,7 @@ class ServerProcess:
     no_webui: bool | None = None
     jinja: bool | None = None
     reasoning_format: Literal['deepseek', 'none', 'nothink'] | None = None
-    reasoning_budget: int | None = None
+    reasoning: Literal['on', 'off', 'auto'] | None = None
     chat_template: str | None = None
     chat_template_file: str | None = None
     server_path: str | None = None
@@ -225,8 +225,8 @@ def start(self, timeout_seconds: int | None = DEFAULT_HTTP_TIMEOUT) -> None:
             server_args.append("--no-jinja")
         if self.reasoning_format is not None:
             server_args.extend(("--reasoning-format", self.reasoning_format))
-        if self.reasoning_budget is not None:
-            server_args.extend(("--reasoning-budget", self.reasoning_budget))
+        if self.reasoning is not None:
+            server_args.extend(("--reasoning", self.reasoning))
         if self.chat_template:
             server_args.extend(["--chat-template", self.chat_template])
         if self.chat_template_file: