|
| 1 | +import { vi, describe, it, expect, beforeEach } from "vitest" |
| 2 | + |
| 3 | +// Mock AWS SDK modules before importing the handler |
| 4 | +vi.mock("@aws-sdk/credential-providers", () => ({ |
| 5 | + fromIni: vi.fn(), |
| 6 | +})) |
| 7 | + |
| 8 | +// Define a shared mock for the send function that will be used by all instances |
| 9 | +const sharedMockSend = vi.fn() |
| 10 | + |
| 11 | +vi.mock("@aws-sdk/client-bedrock-runtime", () => ({ |
| 12 | + BedrockRuntimeClient: vi.fn().mockImplementation(() => ({ |
| 13 | + // Ensure all instances of BedrockRuntimeClient use the sharedMockSend |
| 14 | + send: sharedMockSend, |
| 15 | + config: { region: "us-east-1" }, |
| 16 | + })), |
| 17 | + ConverseStreamCommand: vi.fn(), // This will be the mock constructor for ConverseStreamCommand |
| 18 | + ConverseCommand: vi.fn(), |
| 19 | +})) |
| 20 | + |
| 21 | +// Import after mocks are set up |
| 22 | +import { AwsBedrockHandler } from "../bedrock" |
| 23 | +// Import ConverseStreamCommand to check its mock constructor (which is vi.fn() from the mock factory) |
| 24 | +import { ConverseStreamCommand } from "@aws-sdk/client-bedrock-runtime" |
| 25 | + |
| 26 | +describe("AwsBedrockHandler - Extended Thinking", () => { |
| 27 | + let handler: AwsBedrockHandler |
| 28 | + // This will hold the reference to sharedMockSend for use in tests |
| 29 | + let mockSend: typeof sharedMockSend |
| 30 | + |
| 31 | + const mockOptions = { |
| 32 | + awsRegion: "us-east-1", |
| 33 | + apiModelId: "anthropic.claude-3-7-sonnet-20241029-v1:0", |
| 34 | + enableReasoningEffort: false, // Default to false |
| 35 | + modelTemperature: 0.7, |
| 36 | + } |
| 37 | + |
| 38 | + beforeEach(() => { |
| 39 | + // Clear all mocks. This will clear sharedMockSend and the ConverseStreamCommand mock constructor. |
| 40 | + vi.clearAllMocks() |
| 41 | + // Assign the shared mock to mockSend so tests can configure it. |
| 42 | + mockSend = sharedMockSend |
| 43 | + |
| 44 | + // AwsBedrockHandler will instantiate BedrockRuntimeClient, which will get the sharedMockSend. |
| 45 | + handler = new AwsBedrockHandler(mockOptions) |
| 46 | + }) |
| 47 | + |
| 48 | + describe("Extended Thinking Configuration", () => { |
| 49 | + it("should NOT enable extended thinking by default", async () => { |
| 50 | + // Setup mock response |
| 51 | + mockSend.mockResolvedValue({ |
| 52 | + stream: (async function* () { |
| 53 | + yield { messageStart: { role: "assistant" } } |
| 54 | + yield { |
| 55 | + contentBlockStart: { |
| 56 | + start: { text: "Hello" }, |
| 57 | + contentBlockIndex: 0, |
| 58 | + }, |
| 59 | + } |
| 60 | + yield { messageStop: { stopReason: "end_turn" } } |
| 61 | + })(), |
| 62 | + }) |
| 63 | + |
| 64 | + // Create message |
| 65 | + const messages = [{ role: "user" as const, content: "Test message" }] |
| 66 | + const stream = handler.createMessage("", messages) |
| 67 | + |
| 68 | + // Consume stream |
| 69 | + const chunks = [] |
| 70 | + for await (const chunk of stream) { |
| 71 | + chunks.push(chunk) |
| 72 | + } |
| 73 | + |
| 74 | + // Verify the command was called |
| 75 | + expect(ConverseStreamCommand).toHaveBeenCalled() |
| 76 | + const payload = (ConverseStreamCommand as any).mock.calls[0][0] |
| 77 | + |
| 78 | + // Extended thinking should NOT be enabled by default |
| 79 | + expect(payload.anthropic_version).toBeUndefined() |
| 80 | + expect(payload.additionalModelRequestFields).toBeUndefined() |
| 81 | + expect(payload.inferenceConfig.temperature).toBeDefined() |
| 82 | + expect(payload.inferenceConfig.topP).toBeDefined() |
| 83 | + }) |
| 84 | + |
| 85 | + it("should enable extended thinking when explicitly enabled with reasoning budget", async () => { |
| 86 | + // Enable reasoning mode with thinking tokens |
| 87 | + handler = new AwsBedrockHandler({ |
| 88 | + ...mockOptions, |
| 89 | + enableReasoningEffort: true, |
| 90 | + modelMaxThinkingTokens: 5000, |
| 91 | + }) |
| 92 | + |
| 93 | + // Setup mock response with thinking blocks |
| 94 | + mockSend.mockResolvedValue({ |
| 95 | + stream: (async function* () { |
| 96 | + yield { messageStart: { role: "assistant" } } |
| 97 | + yield { |
| 98 | + contentBlockStart: { |
| 99 | + contentBlock: { |
| 100 | + type: "thinking", |
| 101 | + thinking: "Let me think about this...", |
| 102 | + }, |
| 103 | + contentBlockIndex: 0, |
| 104 | + }, |
| 105 | + } |
| 106 | + yield { |
| 107 | + contentBlockStart: { |
| 108 | + start: { text: "Here is my response" }, |
| 109 | + contentBlockIndex: 1, |
| 110 | + }, |
| 111 | + } |
| 112 | + yield { messageStop: { stopReason: "end_turn" } } |
| 113 | + })(), |
| 114 | + }) |
| 115 | + |
| 116 | + // Create message |
| 117 | + const messages = [{ role: "user" as const, content: "Test message" }] |
| 118 | + const stream = handler.createMessage("", messages) |
| 119 | + |
| 120 | + // Consume stream |
| 121 | + const chunks = [] |
| 122 | + for await (const chunk of stream) { |
| 123 | + chunks.push(chunk) |
| 124 | + } |
| 125 | + |
| 126 | + // Verify the command was called |
| 127 | + expect(ConverseStreamCommand).toHaveBeenCalled() |
| 128 | + const payload = (ConverseStreamCommand as any).mock.calls[0][0] |
| 129 | + |
| 130 | + // Extended thinking should be enabled |
| 131 | + expect(payload.anthropic_version).toBe("bedrock-20250514") |
| 132 | + expect(payload.additionalModelRequestFields).toEqual({ |
| 133 | + thinking: { |
| 134 | + type: "enabled", |
| 135 | + budget_tokens: 5000, |
| 136 | + }, |
| 137 | + }) |
| 138 | + // Temperature and topP should be removed |
| 139 | + expect(payload.inferenceConfig.temperature).toBeUndefined() |
| 140 | + expect(payload.inferenceConfig.topP).toBeUndefined() |
| 141 | + |
| 142 | + // Verify thinking content was processed |
| 143 | + const reasoningChunk = chunks.find((c) => c.type === "reasoning") |
| 144 | + expect(reasoningChunk).toBeDefined() |
| 145 | + expect(reasoningChunk?.text).toBe("Let me think about this...") |
| 146 | + }) |
| 147 | + |
| 148 | + it("should NOT enable extended thinking for unsupported models", async () => { |
| 149 | + // Use a model that doesn't support reasoning |
| 150 | + handler = new AwsBedrockHandler({ |
| 151 | + ...mockOptions, |
| 152 | + apiModelId: "anthropic.claude-3-haiku-20240307-v1:0", |
| 153 | + enableReasoningEffort: true, |
| 154 | + modelMaxThinkingTokens: 5000, |
| 155 | + }) |
| 156 | + |
| 157 | + // Setup mock response |
| 158 | + mockSend.mockResolvedValue({ |
| 159 | + stream: (async function* () { |
| 160 | + yield { messageStart: { role: "assistant" } } |
| 161 | + yield { |
| 162 | + contentBlockStart: { |
| 163 | + start: { text: "Hello" }, |
| 164 | + contentBlockIndex: 0, |
| 165 | + }, |
| 166 | + } |
| 167 | + yield { messageStop: { stopReason: "end_turn" } } |
| 168 | + })(), |
| 169 | + }) |
| 170 | + |
| 171 | + // Create message |
| 172 | + const messages = [{ role: "user" as const, content: "Test message" }] |
| 173 | + const stream = handler.createMessage("", messages) |
| 174 | + |
| 175 | + // Consume stream |
| 176 | + const chunks = [] |
| 177 | + for await (const chunk of stream) { |
| 178 | + chunks.push(chunk) |
| 179 | + } |
| 180 | + |
| 181 | + // Verify the command was called |
| 182 | + expect(ConverseStreamCommand).toHaveBeenCalled() |
| 183 | + const payload = (ConverseStreamCommand as any).mock.calls[0][0] |
| 184 | + |
| 185 | + // Extended thinking should NOT be enabled for unsupported models |
| 186 | + expect(payload.anthropic_version).toBeUndefined() |
| 187 | + expect(payload.additionalModelRequestFields).toBeUndefined() |
| 188 | + expect(payload.inferenceConfig.temperature).toBeDefined() |
| 189 | + expect(payload.inferenceConfig.topP).toBeDefined() |
| 190 | + }) |
| 191 | + }) |
| 192 | + |
| 193 | + describe("Stream Processing", () => { |
| 194 | + it("should handle thinking delta events", async () => { |
| 195 | + // Enable reasoning mode |
| 196 | + handler = new AwsBedrockHandler({ |
| 197 | + ...mockOptions, |
| 198 | + enableReasoningEffort: true, |
| 199 | + modelMaxThinkingTokens: 5000, |
| 200 | + }) |
| 201 | + |
| 202 | + // Setup mock response with thinking deltas |
| 203 | + mockSend.mockResolvedValue({ |
| 204 | + stream: (async function* () { |
| 205 | + yield { messageStart: { role: "assistant" } } |
| 206 | + yield { |
| 207 | + contentBlockDelta: { |
| 208 | + delta: { |
| 209 | + type: "thinking_delta", |
| 210 | + thinking: "First part of thinking...", |
| 211 | + }, |
| 212 | + contentBlockIndex: 0, |
| 213 | + }, |
| 214 | + } |
| 215 | + yield { |
| 216 | + contentBlockDelta: { |
| 217 | + delta: { |
| 218 | + type: "thinking_delta", |
| 219 | + thinking: " Second part of thinking.", |
| 220 | + }, |
| 221 | + contentBlockIndex: 0, |
| 222 | + }, |
| 223 | + } |
| 224 | + yield { messageStop: { stopReason: "end_turn" } } |
| 225 | + })(), |
| 226 | + }) |
| 227 | + |
| 228 | + // Create message |
| 229 | + const messages = [{ role: "user" as const, content: "Test message" }] |
| 230 | + const stream = handler.createMessage("", messages) |
| 231 | + |
| 232 | + // Consume stream |
| 233 | + const chunks = [] |
| 234 | + for await (const chunk of stream) { |
| 235 | + chunks.push(chunk) |
| 236 | + } |
| 237 | + |
| 238 | + // Verify thinking deltas were processed |
| 239 | + const reasoningChunks = chunks.filter((c) => c.type === "reasoning") |
| 240 | + expect(reasoningChunks).toHaveLength(2) |
| 241 | + expect(reasoningChunks[0].text).toBe("First part of thinking...") |
| 242 | + expect(reasoningChunks[1].text).toBe(" Second part of thinking.") |
| 243 | + }) |
| 244 | + |
| 245 | + it("should handle signature delta events as reasoning", async () => { |
| 246 | + // Enable reasoning mode |
| 247 | + handler = new AwsBedrockHandler({ |
| 248 | + ...mockOptions, |
| 249 | + enableReasoningEffort: true, |
| 250 | + modelMaxThinkingTokens: 5000, |
| 251 | + }) |
| 252 | + |
| 253 | + // Setup mock response with signature deltas |
| 254 | + mockSend.mockResolvedValue({ |
| 255 | + stream: (async function* () { |
| 256 | + yield { messageStart: { role: "assistant" } } |
| 257 | + yield { |
| 258 | + contentBlockDelta: { |
| 259 | + delta: { |
| 260 | + type: "signature_delta", |
| 261 | + signature: "[Signature content]", |
| 262 | + }, |
| 263 | + contentBlockIndex: 0, |
| 264 | + }, |
| 265 | + } |
| 266 | + yield { messageStop: { stopReason: "end_turn" } } |
| 267 | + })(), |
| 268 | + }) |
| 269 | + |
| 270 | + // Create message |
| 271 | + const messages = [{ role: "user" as const, content: "Test message" }] |
| 272 | + const stream = handler.createMessage("", messages) |
| 273 | + |
| 274 | + // Consume stream |
| 275 | + const chunks = [] |
| 276 | + for await (const chunk of stream) { |
| 277 | + chunks.push(chunk) |
| 278 | + } |
| 279 | + |
| 280 | + // Verify signature delta was processed as reasoning |
| 281 | + const reasoningChunk = chunks.find((c) => c.type === "reasoning") |
| 282 | + expect(reasoningChunk).toBeDefined() |
| 283 | + expect(reasoningChunk?.text).toBe("[Signature content]") |
| 284 | + }) |
| 285 | + }) |
| 286 | +}) |
0 commit comments