V4.14.10 dev (#6674)

* feat: model config with brand-new price calculate machanism (#6616) * fix: image read and json error (Agent) (#6502) * fix: 1.image read 2.JSON parsing error * dataset cite and pause * perf: plancall second parse * add test --------- Co-authored-by: archer <545436317@qq.com> * master message * remove invalid code * wip: model config * feat: model config with brand-new price calculate machanism * merge main branch * ajust calculate way * ajust priceTiers resolve procession * perf: price config code * fix: default price * fix: test * fix: comment * fix test --------- Co-authored-by: YeYuheng <57035043+YYH211@users.noreply.github.com> Co-authored-by: archer <545436317@qq.com> * wip: fix modal UI (#6634) * wip: fix modal UI * fix: maxInputToken set * chore: add price unit for non llm models * chore: replace question mark icon with beta tag (#6672) * feat:rerank too long; fix:rerank ui(agent),embedding returns 0 (#6663) * feat:rerank too long; fix:rerank ui(agent),embedding returns 0 * rerank * fix:rerank function * perf: rerank code * fix rerank * perf: model price ui --------- Co-authored-by: archer <545436317@qq.com> * remove llmtype field * revert model init * fix: filed * fix: model select filter * perf: multiple selector render * remove invalid checker * remove invalid i18n * perf: model selector tip * perf: model selector tip * fix cr * limit pnpm version * fix: i18n * fix action * set default mintoken * update i18n * perf: usage push * fix:rerank model ui (#6677) * fix: tier match error * fix: testr --------- Co-authored-by: Ryo <whoeverimf5@gmail.com> Co-authored-by: YeYuheng <57035043+YYH211@users.noreply.github.com>
2026-05-07 01:02:55 +08:00 · 2026-03-30 10:05:42 +08:00
parent 03dd9c00a8
commit 3f4400a500
108 changed files with 3774 additions and 1469 deletions
@@ -0,0 +1,345 @@
+import { describe, it, expect, vi, beforeEach } from 'vitest';
+import { ModelTypeEnum } from '@fastgpt/global/core/ai/constants';
+import type { RerankModelItemType } from '@fastgpt/global/core/ai/model.schema';
+
+// hoisted：让 mock 实例可在 beforeEach 中重设
+const { mockCountPromptTokens, mockPOST } = vi.hoisted(() => ({
+  mockCountPromptTokens: vi.fn(),
+  mockPOST: vi.fn()
+}));
+
+vi.mock('@fastgpt/service/common/string/tiktoken', () => ({
+  countPromptTokens: mockCountPromptTokens
+}));
+
+vi.mock('@fastgpt/service/common/api/serverRequest', () => ({
+  POST: (...args: any[]) => mockPOST(...args)
+}));
+
+// Mock text2Chunks：按 chunkSize 字符切分，保证测试确定性
+vi.mock('@fastgpt/service/worker/function', () => ({
+  text2Chunks: vi.fn(async ({ text, chunkSize }: { text: string; chunkSize: number }) => {
+    const chunks: string[] = [];
+    for (let i = 0; i < text.length; i += chunkSize) {
+      chunks.push(text.slice(i, i + chunkSize));
+    }
+    return { chunks };
+  })
+}));
+
+// import 放在 mock 之后
+const { reRankRecall } = await import('@fastgpt/service/core/ai/rerank/index');
+
+const mockModel: RerankModelItemType = {
+  provider: 'test',
+  model: 'rerank-test',
+  name: 'Test Rerank',
+  type: ModelTypeEnum.rerank,
+  maxToken: 8000
+};
+
+describe('reRankRecall', () => {
+  beforeEach(() => {
+    mockPOST.mockReset();
+    mockCountPromptTokens.mockReset();
+    mockCountPromptTokens.mockImplementation(async (text: string) => text.length);
+  });
+
+  // ── 基础场景 ──────────────────────────────────────────────────────────────
+
+  it('正常场景：多文档返回正确 id 和 score', async () => {
+    mockPOST.mockResolvedValueOnce({
+      id: 'r1',
+      results: [
+        { index: 1, relevance_score: 0.9 },
+        { index: 0, relevance_score: 0.5 }
+      ],
+      meta: { tokens: { input_tokens: 20, output_tokens: 0 } }
+    });
+
+    const result = await reRankRecall({
+      model: mockModel,
+      query: 'query',
+      documents: [
+        { id: 'doc1', text: 'hello' },
+        { id: 'doc2', text: 'world' }
+      ]
+    });
+
+    expect(result.inputTokens).toBe(20);
+    expect(result.results).toHaveLength(2);
+    expect(result.results.find((r) => r.id === 'doc2')?.score).toBe(0.9);
+    expect(result.results.find((r) => r.id === 'doc1')?.score).toBe(0.5);
+  });
+
+  it('单文档正常召回', async () => {
+    mockPOST.mockResolvedValueOnce({
+      id: 'r1',
+      results: [{ index: 0, relevance_score: 0.75 }],
+      meta: { tokens: { input_tokens: 10, output_tokens: 0 } }
+    });
+
+    const result = await reRankRecall({
+      model: mockModel,
+      query: 'q',
+      documents: [{ id: 'doc1', text: 'hello' }]
+    });
+
+    expect(result.results).toEqual([{ id: 'doc1', score: 0.75 }]);
+    expect(result.inputTokens).toBe(10);
+  });
+
+  // ── 边界值 ────────────────────────────────────────────────────────────────
+
+  it('documents 为空时直接返回空，不发请求', async () => {
+    const result = await reRankRecall({
+      model: mockModel,
+      query: 'q',
+      documents: []
+    });
+
+    expect(result).toEqual({ results: [], inputTokens: 0 });
+    expect(mockPOST).not.toHaveBeenCalled();
+  });
+
+  it('所有文档 text 为空或空白时，返回空结果，不发请求', async () => {
+    const result = await reRankRecall({
+      model: mockModel,
+      query: 'q',
+      documents: [
+        { id: 'doc1', text: '' },
+        { id: 'doc2', text: '   ' }
+      ]
+    });
+
+    expect(result).toEqual({ results: [], inputTokens: 0 });
+    expect(mockPOST).not.toHaveBeenCalled();
+  });
+
+  // ── 复杂场景：文档切分 ────────────────────────────────────────────────────
+
+  it('文档超过 token 预算时切分 chunks，取最高分聚合并返回原始 doc id', async () => {
+    // maxToken=600, query='q'(length=1), docBudget=599
+    // longText length=1100 > 599 → 被切分
+    // chunkSize = floor((1100/1100)*599*0.9) = 539 → 3 chunks (indices 0,1,2)
+    // doc2 'short' length=5 <= 599 → 不切分 (index 3)
+    const longText = 'a'.repeat(1100);
+
+    mockPOST.mockResolvedValueOnce({
+      id: 'r1',
+      // API 按 score 降序返回
+      results: [
+        { index: 0, relevance_score: 0.8 }, // doc1__chunk_0 → doc1
+        { index: 3, relevance_score: 0.6 }, // doc2
+        { index: 1, relevance_score: 0.3 }, // doc1__chunk_1 → doc1（已存在，跳过）
+        { index: 2, relevance_score: 0.1 } // doc1__chunk_2 → doc1（已存在，跳过）
+      ],
+      meta: { tokens: { input_tokens: 30, output_tokens: 0 } }
+    });
+
+    const result = await reRankRecall({
+      model: { ...mockModel, maxToken: 600 },
+      query: 'q',
+      documents: [
+        { id: 'doc1', text: longText },
+        { id: 'doc2', text: 'short' }
+      ]
+    });
+
+    // 返回的 id 应为原始 doc id，不含 __chunk_ 后缀
+    expect(result.results.every((r) => !r.id.includes('__chunk_'))).toBe(true);
+    expect(result.results).toHaveLength(2);
+    expect(result.results.find((r) => r.id === 'doc1')?.score).toBe(0.8);
+    expect(result.results.find((r) => r.id === 'doc2')?.score).toBe(0.6);
+  });
+
+  it('同一文档多个 chunk，非最高分 chunk 排在前面时，仍取第一个（最高分）', async () => {
+    // doc1 3个 chunks (indices 0,1,2)；API 返回 chunk_1 分最高
+    // maxToken=600, query='q'(1), docBudget=599, chunkSize=539
+    const longText = 'b'.repeat(1100);
+
+    mockPOST.mockResolvedValueOnce({
+      id: 'r1',
+      results: [
+        { index: 1, relevance_score: 0.95 }, // chunk_1 最高
+        { index: 0, relevance_score: 0.4 }, // chunk_0 跳过
+        { index: 2, relevance_score: 0.2 } // chunk_2 跳过
+      ],
+      meta: { tokens: { input_tokens: 20, output_tokens: 0 } }
+    });
+
+    const result = await reRankRecall({
+      model: { ...mockModel, maxToken: 600 },
+      query: 'q',
+      documents: [{ id: 'doc1', text: longText }]
+    });
+
+    expect(result.results).toHaveLength(1);
+    expect(result.results[0]).toEqual({ id: 'doc1', score: 0.95 });
+  });
+
+  // ── inputTokens 计算 ──────────────────────────────────────────────────────
+
+  it('API 未返回 meta tokens 时，通过 countPromptTokens 估算', async () => {
+    mockPOST.mockResolvedValueOnce({
+      id: 'r1',
+      results: [{ index: 0, relevance_score: 0.5 }]
+      // 无 meta
+    });
+
+    const result = await reRankRecall({
+      model: mockModel,
+      query: 'test', // length=4
+      documents: [{ id: 'doc1', text: 'hello' }] // text length=5
+    });
+
+    // documentsTextArray.join('\n') = 'hello'（单元素无分隔符）+ 'test' = 'hellotest'(9)
+    expect(result.inputTokens).toBe(9);
+  });
+
+  it('API 返回 meta tokens 时直接使用', async () => {
+    mockPOST.mockResolvedValueOnce({
+      id: 'r1',
+      results: [{ index: 0, relevance_score: 0.5 }],
+      meta: { tokens: { input_tokens: 42, output_tokens: 0 } }
+    });
+
+    const result = await reRankRecall({
+      model: mockModel,
+      query: 'q',
+      documents: [{ id: 'doc1', text: 'hello' }]
+    });
+
+    expect(result.inputTokens).toBe(42);
+  });
+
+  // ── requestUrl / requestAuth ──────────────────────────────────────────────
+
+  it('有 requestUrl 和 requestAuth 时，使用自定义地址和认证头', async () => {
+    mockPOST.mockResolvedValueOnce({
+      id: 'r1',
+      results: [{ index: 0, relevance_score: 0.5 }],
+      meta: { tokens: { input_tokens: 5, output_tokens: 0 } }
+    });
+
+    await reRankRecall({
+      model: {
+        ...mockModel,
+        requestUrl: 'https://custom.rerank.io/rerank',
+        requestAuth: 'secret-key'
+      },
+      query: 'q',
+      documents: [{ id: 'doc1', text: 'hello' }]
+    });
+
+    expect(mockPOST).toHaveBeenCalledWith(
+      'https://custom.rerank.io/rerank',
+      expect.any(Object),
+      expect.objectContaining({
+        headers: expect.objectContaining({
+          Authorization: 'Bearer secret-key'
+        })
+      })
+    );
+  });
+
+  it('未设置 requestUrl 时，使用 baseUrl/rerank', async () => {
+    mockPOST.mockResolvedValueOnce({
+      id: 'r1',
+      results: [{ index: 0, relevance_score: 0.5 }],
+      meta: { tokens: { input_tokens: 5, output_tokens: 0 } }
+    });
+
+    await reRankRecall({
+      model: mockModel,
+      query: 'q',
+      documents: [{ id: 'doc1', text: 'hello' }]
+    });
+
+    const url: string = mockPOST.mock.calls[0][0];
+    expect(url.endsWith('/rerank')).toBe(true);
+  });
+
+  // ── 异常场景 ──────────────────────────────────────────────────────────────
+
+  it('model 为 undefined 时 reject', async () => {
+    await expect(
+      reRankRecall({
+        model: undefined,
+        query: 'q',
+        documents: [{ id: 'doc1', text: 'hello' }]
+      })
+    ).rejects.toThrow('No rerank model');
+  });
+
+  it('query 超过 maxToken 时 reject', async () => {
+    // maxToken=5, query length=26 → docBudget = 5-26 = -21 ≤ 500 → reject
+    await expect(
+      reRankRecall({
+        model: { ...mockModel, maxToken: 5 },
+        query: 'this query is way too long',
+        documents: [{ id: 'doc1', text: 'hello' }]
+      })
+    ).rejects.toThrow('Rerank query too long');
+  });
+
+  it('docBudget === 500 时 reject（边界值）', async () => {
+    // mockCountPromptTokens 按 text.length 计算
+    // maxToken=501, query='q'(length=1) → docBudget = 501-1 = 500 ≤ 500 → reject
+    await expect(
+      reRankRecall({
+        model: { ...mockModel, maxToken: 501 },
+        query: 'q',
+        documents: [{ id: 'doc1', text: 'hello' }]
+      })
+    ).rejects.toThrow('Rerank query too long');
+  });
+
+  it('docBudget === 501 时不因 query 过长 reject', async () => {
+    // maxToken=502, query='q'(length=1) → docBudget = 502-1 = 501 > 500 → 正常发请求
+    mockPOST.mockResolvedValueOnce({
+      id: 'r1',
+      results: [{ index: 0, relevance_score: 0.5 }],
+      meta: { tokens: { input_tokens: 5, output_tokens: 0 } }
+    });
+
+    const result = await reRankRecall({
+      model: { ...mockModel, maxToken: 502 },
+      query: 'q',
+      documents: [{ id: 'doc1', text: 'hello' }]
+    });
+
+    expect(result.results).toHaveLength(1);
+    expect(mockPOST).toHaveBeenCalledOnce();
+  });
+
+  it('API 请求失败时，reject 并传递原始错误', async () => {
+    mockPOST.mockRejectedValueOnce(new Error('Network error'));
+
+    await expect(
+      reRankRecall({
+        model: mockModel,
+        query: 'q',
+        documents: [{ id: 'doc1', text: 'hello' }]
+      })
+    ).rejects.toThrow('Network error');
+  });
+
+  it('API 返回空 results 时，返回空 results', async () => {
+    mockPOST.mockResolvedValueOnce({
+      id: 'r1',
+      results: []
+    });
+
+    const result = await reRankRecall({
+      model: mockModel,
+      query: 'q',
+      documents: [{ id: 'doc1', text: 'hello' }]
+    });
+
+    expect(result.results).toHaveLength(0);
+    expect(mockPOST).toHaveBeenCalledOnce();
+    // 空 results 时提前返回，inputTokens 固定为 0
+    expect(result.inputTokens).toBe(0);
+  });
+});
@@ -811,15 +811,12 @@ describe('getNodeErrResponse', () => {
  });

  it('should pass through optional fields', () => {
-    const usages = [{ totalPoints: 1, tokens: 100, moduleName: 'test' }] as any;
    const result = getNodeErrResponse({
      error: 'fail',
-      nodeDispatchUsages: usages,
      runTimes: 3,
      newVariables: { a: 1 },
      system_memories: { mem: 'val' }
    });
-    expect(result[DispatchNodeResponseKeyEnum.nodeDispatchUsages]).toBe(usages);
    expect(result[DispatchNodeResponseKeyEnum.runTimes]).toBe(3);
    expect(result[DispatchNodeResponseKeyEnum.newVariables]).toEqual({ a: 1 });
    expect(result[DispatchNodeResponseKeyEnum.memories]).toEqual({ mem: 'val' });
@@ -0,0 +1,95 @@
+import { describe, expect, it, vi } from 'vitest';
+import { formatModelChars2Points } from '@fastgpt/service/support/wallet/usage/utils';
+
+// mock findAIModel，避免依赖全局 model map
+const mockModels: Record<string, any> = {
+  'gpt-4': {
+    name: 'GPT-4',
+    model: 'gpt-4',
+    charsPointsPrice: 0,
+    inputPrice: 3,
+    outputPrice: 6
+  },
+  'gpt-3.5': {
+    name: 'GPT-3.5',
+    model: 'gpt-3.5',
+    charsPointsPrice: 2
+  },
+  'tiered-model': {
+    name: 'Tiered',
+    model: 'tiered-model',
+    priceTiers: [
+      { maxInputTokens: 1, inputPrice: 1, outputPrice: 2 },
+      { inputPrice: 5, outputPrice: 10 }
+    ]
+  }
+};
+
+vi.mock('@fastgpt/service/core/ai/model', () => ({
+  findAIModel: (model: string) => mockModels[model]
+}));
+
+describe('formatModelChars2Points', () => {
+  it('should return 0 points and empty name when model not found', () => {
+    const result = formatModelChars2Points({ model: 'non-existent' });
+    expect(result).toEqual({ totalPoints: 0, modelName: '' });
+  });
+
+  it('should return 0 points and empty name when model is empty string', () => {
+    const result = formatModelChars2Points({ model: '' });
+    expect(result).toEqual({ totalPoints: 0, modelName: '' });
+  });
+
+  it('should calculate points with legacy input/output pricing', () => {
+    const result = formatModelChars2Points({
+      model: 'gpt-4',
+      inputTokens: 1000,
+      outputTokens: 500
+    });
+    expect(result.modelName).toBe('GPT-4');
+    // inputPrice:3 * (1000/1000) + outputPrice:6 * (500/1000) = 3 + 3 = 6
+    expect(result.totalPoints).toBe(6);
+  });
+
+  it('should calculate points with comprehensive price', () => {
+    const result = formatModelChars2Points({
+      model: 'gpt-3.5',
+      inputTokens: 2000,
+      outputTokens: 1000
+    });
+    expect(result.modelName).toBe('GPT-3.5');
+    // charsPointsPrice:2 → inputPrice=outputPrice=2
+    // 2 * (2000/1000) + 2 * (1000/1000) = 4 + 2 = 6
+    expect(result.totalPoints).toBe(6);
+  });
+
+  it('should use default 0 tokens when not provided', () => {
+    const result = formatModelChars2Points({ model: 'gpt-4' });
+    expect(result.modelName).toBe('GPT-4');
+    expect(result.totalPoints).toBe(0);
+  });
+
+  it('should support custom multiple parameter', () => {
+    const result = formatModelChars2Points({
+      model: 'gpt-4',
+      inputTokens: 500,
+      outputTokens: 500,
+      multiple: 500
+    });
+    expect(result.modelName).toBe('GPT-4');
+    // inputPrice:3 * (500/500) + outputPrice:6 * (500/500) = 3 + 6 = 9
+    expect(result.totalPoints).toBe(9);
+  });
+
+  it('should calculate points with price tiers', () => {
+    const result = formatModelChars2Points({
+      model: 'tiered-model',
+      inputTokens: 2000,
+      outputTokens: 100
+    });
+    expect(result.modelName).toBe('Tiered');
+    // inputTokens:200 匹配第二梯度 (inputPrice:5, outputPrice:10)
+    // 5 * (2000/1000) + 10 * (100/1000) = 10 + 1 = 11
+    expect(result.totalPoints).toBe(11);
+  });
+});