test: provider — sampling parameter functions (temperature, topP, topK)

Add 28 tests for ProviderTransform.temperature(), topP(), and topK() which
had zero direct test coverage. These pure functions control LLM sampling
behavior per model family and wrong values cause degraded output quality.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

https://claude.ai/code/session_011NoVCnMW9Kw6eh92ayU7GB

Author: claude

packages/opencode/test/provider/transform.test.ts

@@ -2653,3 +2653,130 @@ describe("ProviderTransform.variants", () => {
     })
   })
 })
+
+// ---------------------------------------------------------------------------
+// Sampling parameter tests: temperature, topP, topK
+// ---------------------------------------------------------------------------
+
+// Minimal stub — temperature/topP/topK only inspect model.id
+function samplingModel(id: string): any {
+  return {
+    id,
+    providerID: "test",
+    api: { id, url: "https://test", npm: "@ai-sdk/openai-compatible" },
+    capabilities: { reasoning: false },
+    limit: { output: 8192 },
+    release_date: "2025-01-01",
+  }
+}
+
+describe("ProviderTransform.temperature", () => {
+  test("returns 0.55 for qwen models", () => {
+    expect(ProviderTransform.temperature(samplingModel("qwen-2.5-72b"))).toBe(0.55)
+    expect(ProviderTransform.temperature(samplingModel("qwen-plus"))).toBe(0.55)
+  })
+
+  test("returns undefined for claude models", () => {
+    expect(ProviderTransform.temperature(samplingModel("claude-3-7-sonnet"))).toBeUndefined()
+    expect(ProviderTransform.temperature(samplingModel("claude-opus-4-5"))).toBeUndefined()
+  })
+
+  test("returns 1.0 for gemini models", () => {
+    expect(ProviderTransform.temperature(samplingModel("gemini-2.5-pro"))).toBe(1.0)
+    expect(ProviderTransform.temperature(samplingModel("gemini-3.0-flash"))).toBe(1.0)
+  })
+
+  test("returns 1.0 for glm-4.6 and glm-4.7 models", () => {
+    expect(ProviderTransform.temperature(samplingModel("glm-4.6-plus"))).toBe(1.0)
+    expect(ProviderTransform.temperature(samplingModel("glm-4.7"))).toBe(1.0)
+  })
+
+  test("returns 1.0 for minimax-m2 models", () => {
+    expect(ProviderTransform.temperature(samplingModel("minimax-m2-01"))).toBe(1.0)
+  })
+
+  test("returns 1.0 for kimi-k2-thinking", () => {
+    expect(ProviderTransform.temperature(samplingModel("kimi-k2-thinking"))).toBe(1.0)
+  })
+
+  test("returns 1.0 for kimi-k2.5 (matches 'k2.' pattern)", () => {
+    expect(ProviderTransform.temperature(samplingModel("kimi-k2.5"))).toBe(1.0)
+  })
+
+  test("returns 1.0 for kimi-k2p5 (matches 'k2p' pattern)", () => {
+    expect(ProviderTransform.temperature(samplingModel("kimi-k2p5"))).toBe(1.0)
+  })
+
+  test("returns 1.0 for kimi-k2-5 (matches 'k2-5' pattern)", () => {
+    expect(ProviderTransform.temperature(samplingModel("kimi-k2-5-latest"))).toBe(1.0)
+  })
+
+  test("returns 0.6 for base kimi-k2", () => {
+    expect(ProviderTransform.temperature(samplingModel("kimi-k2"))).toBe(0.6)
+  })
+
+  test("returns undefined for generic/unknown models", () => {
+    expect(ProviderTransform.temperature(samplingModel("gpt-5.2"))).toBeUndefined()
+    expect(ProviderTransform.temperature(samplingModel("llama-3.3-70b"))).toBeUndefined()
+  })
+})
+
+describe("ProviderTransform.topP", () => {
+  test("returns 1 for qwen models", () => {
+    expect(ProviderTransform.topP(samplingModel("qwen-2.5-72b"))).toBe(1)
+  })
+
+  test("returns 0.95 for minimax-m2", () => {
+    expect(ProviderTransform.topP(samplingModel("minimax-m2-01"))).toBe(0.95)
+  })
+
+  test("returns 0.95 for gemini", () => {
+    expect(ProviderTransform.topP(samplingModel("gemini-2.5-pro"))).toBe(0.95)
+  })
+
+  test("returns 0.95 for kimi-k2.5", () => {
+    expect(ProviderTransform.topP(samplingModel("kimi-k2.5"))).toBe(0.95)
+  })
+
+  test("returns 0.95 for kimi-k2p5", () => {
+    expect(ProviderTransform.topP(samplingModel("kimi-k2p5"))).toBe(0.95)
+  })
+
+  test("returns 0.95 for kimi-k2-5", () => {
+    expect(ProviderTransform.topP(samplingModel("kimi-k2-5"))).toBe(0.95)
+  })
+
+  test("returns undefined for other models", () => {
+    expect(ProviderTransform.topP(samplingModel("claude-opus-4-5"))).toBeUndefined()
+    expect(ProviderTransform.topP(samplingModel("gpt-5.2"))).toBeUndefined()
+  })
+})
+
+describe("ProviderTransform.topK", () => {
+  test("returns 40 for minimax-m2. (dot pattern)", () => {
+    expect(ProviderTransform.topK(samplingModel("minimax-m2.5"))).toBe(40)
+  })
+
+  test("returns 40 for minimax-m25 (m25 pattern)", () => {
+    expect(ProviderTransform.topK(samplingModel("minimax-m25"))).toBe(40)
+  })
+
+  test("returns 40 for minimax-m21 (m21 pattern)", () => {
+    expect(ProviderTransform.topK(samplingModel("minimax-m21"))).toBe(40)
+  })
+
+  test("returns 20 for base minimax-m2", () => {
+    expect(ProviderTransform.topK(samplingModel("minimax-m2"))).toBe(20)
+    expect(ProviderTransform.topK(samplingModel("minimax-m2-01"))).toBe(20)
+  })
+
+  test("returns 64 for gemini", () => {
+    expect(ProviderTransform.topK(samplingModel("gemini-2.5-pro"))).toBe(64)
+  })
+
+  test("returns undefined for other models", () => {
+    expect(ProviderTransform.topK(samplingModel("claude-opus-4-5"))).toBeUndefined()
+    expect(ProviderTransform.topK(samplingModel("gpt-5.2"))).toBeUndefined()
+    expect(ProviderTransform.topK(samplingModel("qwen-2.5-72b"))).toBeUndefined()
+  })
+})