mirror of
https://github.com/tiennm99/goclaw.git
synced 2026-10-13 20:14:50 +00:00
viettranx
86d564967f
Gemini 2.5/3 default to high thinking via OpenAI-compat, consuming the entire max_tokens budget and truncating titles to a single word (and adding latency). Title generation is a trivial task that does not benefit from reasoning, so set OptThinkingLevel="off" and bump max_tokens from 50 to 256 for headroom across reasoning-capable providers. Also update mapGeminiReasoningEffort to forward "off" as "low" (the minimum effort all Gemini models accept via OpenAI-compat), since not forwarding causes the server to fall back to "high".
97 lines
3.7 KiB
Go
97 lines
3.7 KiB
Go
package providers
|
|
|
|
import (
|
|
"testing"
|
|
)
|
|
|
|
// TestBuildRequestBody_GeminiForwardsReasoningEffort verifies that
|
|
// `OptThinkingLevel` maps to `reasoning_effort` on the wire for Gemini routes.
|
|
// Without this forwarding, Gemini 3 defaults to its highest thinking budget and
|
|
// consumes the full max_tokens allocation, leaving no tokens for tool args.
|
|
// Trace: 019d8f33-2de1-7ab2-9a32-9df92cd610dd.
|
|
func TestBuildRequestBody_GeminiForwardsReasoningEffort(t *testing.T) {
|
|
cases := []struct {
|
|
name string
|
|
level string
|
|
wantValue string
|
|
wantExists bool
|
|
}{
|
|
{"low_verbatim", "low", "low", true},
|
|
{"minimal_verbatim", "minimal", "minimal", true},
|
|
{"high_verbatim", "high", "high", true},
|
|
{"medium_maps_to_high", "medium", "high", true},
|
|
{"off_maps_to_low", "off", "low", true},
|
|
{"empty_omitted", "", "", false},
|
|
{"unknown_omitted", "garbage", "", false},
|
|
}
|
|
|
|
for _, tc := range cases {
|
|
t.Run(tc.name, func(t *testing.T) {
|
|
p := NewOpenAIProvider("test-gemini", "key",
|
|
"https://generativelanguage.googleapis.com/v1beta/openai", "gemini-3-flash-preview")
|
|
req := ChatRequest{
|
|
Messages: []Message{{Role: "user", Content: "hi"}},
|
|
Options: map[string]any{OptThinkingLevel: tc.level},
|
|
}
|
|
body := p.buildRequestBody("gemini-3-flash-preview", req, false)
|
|
got, exists := body[OptReasoningEffort]
|
|
if exists != tc.wantExists {
|
|
t.Fatalf("level=%q: reasoning_effort presence = %v, want %v (body=%v)", tc.level, exists, tc.wantExists, body)
|
|
}
|
|
if exists {
|
|
if str, ok := got.(string); !ok || str != tc.wantValue {
|
|
t.Fatalf("level=%q: reasoning_effort = %v, want %q", tc.level, got, tc.wantValue)
|
|
}
|
|
}
|
|
})
|
|
}
|
|
}
|
|
|
|
// TestBuildRequestBody_GeminiDetectionByModelName verifies that gemini routing
|
|
// detection works via model substring even when the apiBase is a proxy
|
|
// (OpenRouter, LiteLLM), and that non-gemini models on the same proxy don't
|
|
// receive `reasoning_effort`.
|
|
func TestBuildRequestBody_GeminiDetectionByModelName(t *testing.T) {
|
|
t.Run("openrouter_gemini_forwards", func(t *testing.T) {
|
|
p := NewOpenAIProvider("openrouter", "key",
|
|
"https://openrouter.ai/api/v1", "google/gemini-2.5-pro")
|
|
req := ChatRequest{
|
|
Messages: []Message{{Role: "user", Content: "hi"}},
|
|
Options: map[string]any{OptThinkingLevel: "low"},
|
|
}
|
|
body := p.buildRequestBody("google/gemini-2.5-pro", req, false)
|
|
if got, ok := body[OptReasoningEffort].(string); !ok || got != "low" {
|
|
t.Fatalf("openrouter gemini must forward reasoning_effort=low; got %v (exists=%v)", got, ok)
|
|
}
|
|
})
|
|
|
|
t.Run("openrouter_claude_omits", func(t *testing.T) {
|
|
p := NewOpenAIProvider("openrouter", "key",
|
|
"https://openrouter.ai/api/v1", "anthropic/claude-sonnet-4")
|
|
req := ChatRequest{
|
|
Messages: []Message{{Role: "user", Content: "hi"}},
|
|
Options: map[string]any{OptThinkingLevel: "low"},
|
|
}
|
|
body := p.buildRequestBody("anthropic/claude-sonnet-4", req, false)
|
|
if _, exists := body[OptReasoningEffort]; exists {
|
|
t.Fatalf("non-gemini route must NOT receive reasoning_effort (body=%v)", body)
|
|
}
|
|
})
|
|
}
|
|
|
|
// TestBuildRequestBody_NonGeminiUnaffected verifies that vanilla OpenAI-compat
|
|
// hosts (Together, Groq, vLLM, etc.) are not affected by the gemini branch —
|
|
// they should continue to reject `reasoning_effort` via the existing gate.
|
|
func TestBuildRequestBody_NonGeminiUnaffected(t *testing.T) {
|
|
p := NewOpenAIProvider("together", "key",
|
|
"https://api.together.xyz/v1", "Qwen/Qwen2.5-72B-Instruct-Turbo")
|
|
req := ChatRequest{
|
|
Messages: []Message{{Role: "user", Content: "hi"}},
|
|
Options: map[string]any{OptThinkingLevel: "low"},
|
|
}
|
|
body := p.buildRequestBody("Qwen/Qwen2.5-72B-Instruct-Turbo", req, false)
|
|
if _, exists := body[OptReasoningEffort]; exists {
|
|
t.Fatalf("together/qwen must NOT receive reasoning_effort; body=%v", body)
|
|
}
|
|
}
|