feat(ratio-setting): add compatibility for gemma-4 model suffix handling

[sxjeru]

根据 AIS 和相关文档，Gemma 4 已支持通过 thinkingLevel 控制思考开关。

即使 newapi 支持 gemma-4-26b-a4b-it-minimal 。

Summary by CodeRabbit

• Bug Fixes
  • Improved normalization of gemma-4 model names to ensure consistent formatting across different model variants and configurations.

[coderabbitai]

Walkthrough

A new NormalizeGemma4EffortModelName function is added to normalize gemma-4* model names by trimming effort suffixes when the Gemini thinking adapter setting is enabled. The existing FormatMatchingModelName function is extended to invoke this normalization as an additional processing step.

Changes

Cohort / File(s)	Summary
Model Ratio Normalization setting/ratio_setting/model_ratio.go	Added NormalizeGemma4EffortModelName exported function to trim effort suffixes from gemma-4* models using reasoning.TrimEffortSuffix. Extended FormatMatchingModelName to invoke this normalization after gemini-2.5-* thinking-budget processing and before GPT/gizmo prefix mappings.

Estimated code review effort

🎯 2 (Simple) | ⏱️ ~10 minutes

Possibly related PRs

• feat: gemini 3 thinking level gemini-3-pro-preview-high #2344: Modifies the same file to add/extend Gemini model name normalization by trimming effort suffixes via reasoning.TrimEffortSuffix.

Suggested reviewers

• seefs001

Poem

🐰 With whiskers twitching, effort suffixes fall,
Gemma-4 models trimmed down, normalized for all,
No -minimal, no -high to clutter the way,
Just clean model names to brighten the day! ✨

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 66.67% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately reflects the main change: adding normalization support for gemma-4 model names with effort suffixes (-minimal/-high) when the thinking adapter setting is enabled.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Review rate limit: 7/8 reviews remaining, refill in 7 minutes and 30 seconds.}

_{Comment @coderabbitai help to get the list of available commands and usage tips.}

[Copilot]

Pull request overview

This PR extends the model-name normalization logic used by ratio/price lookups to handle Gemma 4 “reasoning effort” suffixes (e.g., -minimal, -high) by trimming them to the base model name, improving compatibility with upstream model naming variants.

Changes:

• Add setting/reasoning dependency to reuse the shared effort-suffix parsing logic.
• Update FormatMatchingModelName to trim Gemma 4 effort suffixes (via reasoning.TrimEffortSuffix) so ratio matching uses the base model name.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Calcium-Ion]

这个修改的意义是？

[sxjeru]

贵项目当前已支持 gemini-3-flash-preview-minimal 通过添加后缀控制 思考等级 参数。

本 pr 应当能让 gemma 4 也支持控制思考参数。

[Calcium-Ion]

你这个pr的修改并不能支持后缀控制。

[sxjeru]

应该可以用了。

[coderabbitai]

🧹 Nitpick comments (1)

setting/ratio_setting/model_ratio.go (1)

726-739: ⚡ Quick win

Add table-driven tests for adapter on/off and suffix trimming cases.

This path is config-gated and easy to regress; a small test matrix would lock behavior down.

✅ Proposed test diff

+// setting/ratio_setting/model_ratio_test.go
+package ratio_setting
+
+import (
+	"testing"
+
+	"github.com/QuantumNous/new-api/setting/model_setting"
+)
+
+func TestNormalizeGemma4EffortModelName(t *testing.T) {
+	gs := model_setting.GetGeminiSettings()
+	orig := gs.ThinkingAdapterEnabled
+	defer func() { gs.ThinkingAdapterEnabled = orig }()
+
+	tests := []struct {
+		name      string
+		enabled   bool
+		input     string
+		expected  string
+	}{
+		{"adapter off keeps suffix", false, "gemma-4-26b-it-minimal", "gemma-4-26b-it-minimal"},
+		{"adapter on trims suffix", true, "gemma-4-26b-it-minimal", "gemma-4-26b-it"},
+		{"adapter on non-effort unchanged", true, "gemma-4-26b-it", "gemma-4-26b-it"},
+		{"non-gemma unchanged", true, "gemini-2.5-flash", "gemini-2.5-flash"},
+	}
+
+	for _, tt := range tests {
+		gs.ThinkingAdapterEnabled = tt.enabled
+		if got := NormalizeGemma4EffortModelName(tt.input); got != tt.expected {
+			t.Fatalf("%s: got %q, want %q", tt.name, got, tt.expected)
+		}
+	}
+}

Also applies to: 752-752

🤖 Prompt for AI Agents

Verify each finding against the current code and only fix it if needed.

In `@setting/ratio_setting/model_ratio.go` around lines 726 - 739, Add
table-driven unit tests for NormalizeGemma4EffortModelName covering combinations
of ThinkingAdapterEnabled true/false and model name variants (no gemma prefix,
gemma-4 with no suffix, gemma-4 with effort suffixes like -minimal/-high) to
ensure suffix trimming only occurs when adapter is enabled; use subtests or a
loop with cases that assert expected output from NormalizeGemma4EffortModelName
given each input and toggle
model_setting.GetGeminiSettings().ThinkingAdapterEnabled accordingly. Also add
the same matrix tests for the related function referenced near line 752 (the
other Normalize* function that uses reasoning.TrimEffortSuffix) to prevent
regressions. Ensure tests restore any global settings after each case to avoid
cross-test pollution.

🤖 Prompt for all review comments with AI agents

Verify each finding against the current code and only fix it if needed.

Nitpick comments:
In `@setting/ratio_setting/model_ratio.go`:
- Around line 726-739: Add table-driven unit tests for
NormalizeGemma4EffortModelName covering combinations of ThinkingAdapterEnabled
true/false and model name variants (no gemma prefix, gemma-4 with no suffix,
gemma-4 with effort suffixes like -minimal/-high) to ensure suffix trimming only
occurs when adapter is enabled; use subtests or a loop with cases that assert
expected output from NormalizeGemma4EffortModelName given each input and toggle
model_setting.GetGeminiSettings().ThinkingAdapterEnabled accordingly. Also add
the same matrix tests for the related function referenced near line 752 (the
other Normalize* function that uses reasoning.TrimEffortSuffix) to prevent
regressions. Ensure tests restore any global settings after each case to avoid
cross-test pollution.

---

ℹ️ Review info

⚙️ Run configuration

Configuration used: Organization UI

Review profile: CHILL

Plan: Pro

Run ID: 13eb6fae-a17b-4fd6-b6df-a3f1a381f3fa

📥 Commits

Reviewing files that changed from the base of the PR and between f56397b and 168eaf5.

📒 Files selected for processing (1)

• setting/ratio_setting/model_ratio.go