nezumi/9router
1
0
Fork 0
mirror of https://github.com/Nezumi-2711/9router.git synced 2026-09-22 20:00:47 +00:00
Code Issues Packages Projects Releases Wiki Activity

fix(gemini-cli): raise output floor for thinking and add validated toolConfig (#2486)

Browse Source 
Gemini CLI requests with small max_tokens spend the whole output budget on
thoughts after reasoning_effort maps to thinkingConfig, returning blank
content or finish=length. Raise maxOutputTokens floors per thinking level/
budget (clamped to caps.maxOutput). Also emit toolConfig
functionCallingConfig.mode=VALIDATED for Gemini CLI tool requests to avoid
MALFORMED_FUNCTION_CALL.

Co-authored-by: Cursor <cursoragent@cursor.com>
This commit is contained in:
DOMANHDUC
2026-07-10 11:32:11 +07:00
committed by decolua
co-authored by Cursor
parent 0d4d4bc261
commit 7610f28f42
4 changed files with 107 additions and 12 deletions
Download Patch File Download Diff File Expand all files Collapse all files
tests/translator/thinking-unified.test.js
+16
View File
@@ -78,11 +78,27 @@ describe("applyThinking per provider format", () => {
const out = apply("gemini", "gemini-3-pro", { reasoning_effort: "auto" }, "gemini");
expect(out.generationConfig.thinkingConfig.thinkingLevel).toBe("high");
});
it("gemini-3 high thinking raises too-small maxOutputTokens", () => {
const out = apply("gemini-cli", "gemini-3.1-pro-preview", {
request: { generationConfig: { maxOutputTokens: 128 } },
reasoning_effort: "high",
}, "gemini-cli");
expect(out.request.generationConfig.thinkingConfig).toEqual({ thinkingLevel: "high", includeThoughts: true });
expect(out.request.generationConfig.maxOutputTokens).toBe(65535);
});
it("gemini-2.5 → thinkingBudget", () => {
const out = apply("gemini", "gemini-2.5-flash", { reasoning_effort: "high" }, "gemini");
expect(out.generationConfig.thinkingConfig.thinkingBudget).toBe(24576);
expect(out.generationConfig.thinkingConfig.thinkingLevel).toBeUndefined();
});
it("gemini-2.5 budget thinking keeps enough room for answer tokens", () => {
const out = apply("gemini-cli", "gemini-2.5-pro", {
request: { generationConfig: { maxOutputTokens: 1024 } },
reasoning_effort: "high",
}, "gemini-cli");
expect(out.request.generationConfig.thinkingConfig).toEqual({ thinkingBudget: 24576, includeThoughts: true });
expect(out.request.generationConfig.maxOutputTokens).toBe(32768);
});
it("GLM off → enable_thinking:false (not thinking.disabled)", () => {
const out = apply("openai", "glm-4.6", { reasoning_effort: "none" }, "glm");
expect(out.enable_thinking).toBe(false);