Disable Nemotron thinking because it uses way too many tokens to think

atdrendel · atdrendel · commit 877cf575c3aa · 2026-03-12T22:46:59.000+01:00
diff --git a/Tests/MLXLMIntegrationTests/ToolCallIntegrationTests.swift b/Tests/MLXLMIntegrationTests/ToolCallIntegrationTests.swift
@@ -380,7 +380,8 @@ public class ToolCallIntegrationTests: XCTestCase {
                     "What's the weather in Tokyo and what time is it there?"
                 ),
             ],
-            tools: multiToolSchema
+            tools: multiToolSchema,
+            additionalContext: ["enable_thinking": false]
         )
 
         let (result, toolCalls) = try await generateWithTools(

Original file line number	Diff line number	Diff line change
`@@ -380,7 +380,8 @@ public class ToolCallIntegrationTests: XCTestCase {`
`380`	`380`	`"What's the weather in Tokyo and what time is it there?"`
`381`	`381`	`),`
`382`	`382`	`],`
`383`		`- tools: multiToolSchema`
	`383`	`+ tools: multiToolSchema,`
	`384`	`+ additionalContext: ["enable_thinking": false]`
`384`	`385`	`)`
`385`	`386`
`386`	`387`	`let (result, toolCalls) = try await generateWithTools(`