mirror of
https://github.com/earendil-works/pi.git
synced 2026-06-18 15:54:04 +08:00
fix(ai): Fix a configuration bug with Opus 4.7 adaptive thinking (#3286)
This commit is contained in:
committed by
GitHub
Unverified
parent
72619e9246
commit
d1c6cb1e0f
Generated
+5
-5
@@ -64,9 +64,9 @@
|
||||
}
|
||||
},
|
||||
"node_modules/@anthropic-ai/sdk": {
|
||||
"version": "0.73.0",
|
||||
"resolved": "https://registry.npmjs.org/@anthropic-ai/sdk/-/sdk-0.73.0.tgz",
|
||||
"integrity": "sha512-URURVzhxXGJDGUGFunIOtBlSl7KWvZiAAKY/ttTkZAkXT9bTPqdk2eK0b8qqSxXpikh3QKPnPYpiyX98zf5ebw==",
|
||||
"version": "0.90.0",
|
||||
"resolved": "https://registry.npmjs.org/@anthropic-ai/sdk/-/sdk-0.90.0.tgz",
|
||||
"integrity": "sha512-MzZtPabJF1b0FTDl6Z6H5ljphPwACLGP13lu8MTiB8jXaW/YXlpOp+Po2cVou3MPM5+f5toyLnul9whKCy7fBg==",
|
||||
"license": "MIT",
|
||||
"dependencies": {
|
||||
"json-schema-to-ts": "^3.1.1"
|
||||
@@ -8714,8 +8714,8 @@
|
||||
"version": "0.67.4",
|
||||
"license": "MIT",
|
||||
"dependencies": {
|
||||
"@anthropic-ai/sdk": "^0.73.0",
|
||||
"@aws-sdk/client-bedrock-runtime": "^3.983.0",
|
||||
"@anthropic-ai/sdk": "^0.90.0",
|
||||
"@aws-sdk/client-bedrock-runtime": "^3.1030.0",
|
||||
"@google/genai": "^1.40.0",
|
||||
"@mistralai/mistralai": "1.14.1",
|
||||
"@sinclair/typebox": "^0.34.41",
|
||||
|
||||
@@ -72,8 +72,8 @@
|
||||
"prepublishOnly": "npm run clean && npm run build"
|
||||
},
|
||||
"dependencies": {
|
||||
"@anthropic-ai/sdk": "^0.73.0",
|
||||
"@aws-sdk/client-bedrock-runtime": "^3.983.0",
|
||||
"@anthropic-ai/sdk": "^0.90.0",
|
||||
"@aws-sdk/client-bedrock-runtime": "^3.1030.0",
|
||||
"@google/genai": "^1.40.0",
|
||||
"@mistralai/mistralai": "1.14.1",
|
||||
"@sinclair/typebox": "^0.34.41",
|
||||
|
||||
@@ -50,14 +50,19 @@ export function calculateCost<TApi extends Api>(model: Model<TApi>, usage: Usage
|
||||
*
|
||||
* Supported today:
|
||||
* - GPT-5.2 / GPT-5.3 / GPT-5.4 model families
|
||||
* - Opus 4.6 models (xhigh maps to adaptive effort "max" on Anthropic-compatible providers)
|
||||
* - Opus 4.6+ models (xhigh maps to adaptive effort "max" on Anthropic-compatible providers)
|
||||
*/
|
||||
export function supportsXhigh<TApi extends Api>(model: Model<TApi>): boolean {
|
||||
if (model.id.includes("gpt-5.2") || model.id.includes("gpt-5.3") || model.id.includes("gpt-5.4")) {
|
||||
return true;
|
||||
}
|
||||
|
||||
if (model.id.includes("opus-4-6") || model.id.includes("opus-4.6")) {
|
||||
if (
|
||||
model.id.includes("opus-4-6") ||
|
||||
model.id.includes("opus-4.6") ||
|
||||
model.id.includes("opus-4-7") ||
|
||||
model.id.includes("opus-4.7")
|
||||
) {
|
||||
return true;
|
||||
}
|
||||
|
||||
|
||||
@@ -423,12 +423,14 @@ function handleContentBlockStop(
|
||||
}
|
||||
|
||||
/**
|
||||
* Check if the model supports adaptive thinking (Opus 4.6 and Sonnet 4.6).
|
||||
* Check if the model supports adaptive thinking (Opus 4.6+, Sonnet 4.6).
|
||||
*/
|
||||
function supportsAdaptiveThinking(modelId: string): boolean {
|
||||
return (
|
||||
modelId.includes("opus-4-6") ||
|
||||
modelId.includes("opus-4.6") ||
|
||||
modelId.includes("opus-4-7") ||
|
||||
modelId.includes("opus-4.7") ||
|
||||
modelId.includes("sonnet-4-6") ||
|
||||
modelId.includes("sonnet-4.6")
|
||||
);
|
||||
@@ -437,7 +439,7 @@ function supportsAdaptiveThinking(modelId: string): boolean {
|
||||
function mapThinkingLevelToEffort(
|
||||
level: SimpleStreamOptions["reasoning"],
|
||||
modelId: string,
|
||||
): "low" | "medium" | "high" | "max" {
|
||||
): "low" | "medium" | "high" | "xhigh" | "max" {
|
||||
switch (level) {
|
||||
case "minimal":
|
||||
case "low":
|
||||
@@ -447,7 +449,13 @@ function mapThinkingLevelToEffort(
|
||||
case "high":
|
||||
return "high";
|
||||
case "xhigh":
|
||||
return modelId.includes("opus-4-6") || modelId.includes("opus-4.6") ? "max" : "high";
|
||||
if (modelId.includes("opus-4-6") || modelId.includes("opus-4.6")) {
|
||||
return "max";
|
||||
}
|
||||
if (modelId.includes("opus-4-7") || modelId.includes("opus-4.7")) {
|
||||
return "xhigh";
|
||||
}
|
||||
return "high";
|
||||
default:
|
||||
return "high";
|
||||
}
|
||||
|
||||
@@ -153,7 +153,7 @@ function convertContentBlocks(content: (TextContent | ImageContent)[]):
|
||||
return blocks;
|
||||
}
|
||||
|
||||
export type AnthropicEffort = "low" | "medium" | "high" | "max";
|
||||
export type AnthropicEffort = "low" | "medium" | "high" | "xhigh" | "max";
|
||||
|
||||
export interface AnthropicOptions extends StreamOptions {
|
||||
/**
|
||||
@@ -168,9 +168,10 @@ export interface AnthropicOptions extends StreamOptions {
|
||||
*/
|
||||
thinkingBudgetTokens?: number;
|
||||
/**
|
||||
* Effort level for adaptive thinking (Opus 4.6 and Sonnet 4.6).
|
||||
* Effort level for adaptive thinking (Opus 4.6+ and Sonnet 4.6).
|
||||
* Controls how much thinking Claude allocates:
|
||||
* - "max": Always thinks with no constraints (Opus 4.6 only)
|
||||
* - "xhigh": Highest reasoning level (Opus 4.7)
|
||||
* - "high": Always thinks, deep reasoning (default)
|
||||
* - "medium": Moderate thinking, may skip for simple queries
|
||||
* - "low": Minimal thinking, skips for simple tasks
|
||||
@@ -448,13 +449,15 @@ export const streamAnthropic: StreamFunction<"anthropic-messages", AnthropicOpti
|
||||
};
|
||||
|
||||
/**
|
||||
* Check if a model supports adaptive thinking (Opus 4.6 and Sonnet 4.6)
|
||||
* Check if a model supports adaptive thinking (Opus 4.6+, Sonnet 4.6)
|
||||
*/
|
||||
function supportsAdaptiveThinking(modelId: string): boolean {
|
||||
// Opus 4.6 and Sonnet 4.6 model IDs (with or without date suffix)
|
||||
// Adaptive-thinking model IDs (with or without date suffix)
|
||||
return (
|
||||
modelId.includes("opus-4-6") ||
|
||||
modelId.includes("opus-4.6") ||
|
||||
modelId.includes("opus-4-7") ||
|
||||
modelId.includes("opus-4.7") ||
|
||||
modelId.includes("sonnet-4-6") ||
|
||||
modelId.includes("sonnet-4.6")
|
||||
);
|
||||
@@ -462,7 +465,7 @@ function supportsAdaptiveThinking(modelId: string): boolean {
|
||||
|
||||
/**
|
||||
* Map ThinkingLevel to Anthropic effort levels for adaptive thinking.
|
||||
* Note: effort "max" is only valid on Opus 4.6.
|
||||
* Note: effort "max" is only valid on Opus 4.6, while Opus 4.7 supports "xhigh".
|
||||
*/
|
||||
function mapThinkingLevelToEffort(level: SimpleStreamOptions["reasoning"], modelId: string): AnthropicEffort {
|
||||
switch (level) {
|
||||
@@ -475,7 +478,13 @@ function mapThinkingLevelToEffort(level: SimpleStreamOptions["reasoning"], model
|
||||
case "high":
|
||||
return "high";
|
||||
case "xhigh":
|
||||
return modelId.includes("opus-4-6") || modelId.includes("opus-4.6") ? "max" : "high";
|
||||
if (modelId.includes("opus-4-6") || modelId.includes("opus-4.6")) {
|
||||
return "max";
|
||||
}
|
||||
if (modelId.includes("opus-4-7") || modelId.includes("opus-4.7")) {
|
||||
return "xhigh";
|
||||
}
|
||||
return "high";
|
||||
default:
|
||||
return "high";
|
||||
}
|
||||
|
||||
@@ -0,0 +1,72 @@
|
||||
import { describe, expect, it } from "vitest";
|
||||
import { getModel } from "../src/models.js";
|
||||
import { streamSimple } from "../src/stream.js";
|
||||
import type { Context } from "../src/types.js";
|
||||
|
||||
interface AnthropicThinkingPayload {
|
||||
thinking?: { type: string };
|
||||
output_config?: { effort?: string };
|
||||
}
|
||||
|
||||
function makeContext(): Context {
|
||||
return {
|
||||
systemPrompt: "You are a precise assistant. Follow the user's instructions exactly.",
|
||||
messages: [
|
||||
{
|
||||
role: "user",
|
||||
content:
|
||||
"Compute 48291 * 7317 and 90844 - 17729, add the results, and determine whether the sum is divisible by 11. Reply with exactly this format and nothing else: sum=<sum>; divisibleBy11=<yes|no>",
|
||||
timestamp: Date.now(),
|
||||
},
|
||||
],
|
||||
};
|
||||
}
|
||||
|
||||
describe.skipIf(!process.env.ANTHROPIC_API_KEY)("Anthropic Opus 4.7 smoke", () => {
|
||||
it("streams Claude Opus 4.7 with reasoning enabled", { retry: 2, timeout: 30000 }, async () => {
|
||||
const model = getModel("anthropic", "claude-opus-4-7");
|
||||
let capturedPayload: AnthropicThinkingPayload | undefined;
|
||||
const s = streamSimple(model, makeContext(), {
|
||||
reasoning: "high",
|
||||
maxTokens: 1024,
|
||||
onPayload: (payload) => {
|
||||
capturedPayload = payload as AnthropicThinkingPayload;
|
||||
return payload;
|
||||
},
|
||||
});
|
||||
|
||||
let sawThinking = false;
|
||||
|
||||
for await (const event of s) {
|
||||
if (event.type === "thinking_start" || event.type === "thinking_delta" || event.type === "thinking_end") {
|
||||
sawThinking = true;
|
||||
}
|
||||
}
|
||||
|
||||
const response = await s.result();
|
||||
expect(response.stopReason, response.errorMessage).toBe("stop");
|
||||
expect(response.errorMessage).toBeFalsy();
|
||||
expect(capturedPayload?.thinking).toEqual({ type: "adaptive" });
|
||||
expect(capturedPayload?.output_config).toEqual({ effort: "high" });
|
||||
expect(sawThinking).toBe(true);
|
||||
|
||||
const thinkingBlock = response.content.find((block) => block.type === "thinking");
|
||||
expect(thinkingBlock?.type).toBe("thinking");
|
||||
if (!thinkingBlock || thinkingBlock.type !== "thinking") {
|
||||
throw new Error("Expected thinking block from Claude Opus 4.7");
|
||||
}
|
||||
expect(typeof thinkingBlock.thinkingSignature).toBe("string");
|
||||
const thinkingSignature = thinkingBlock.thinkingSignature;
|
||||
if (!thinkingSignature) {
|
||||
throw new Error("Expected thinking signature from Claude Opus 4.7");
|
||||
}
|
||||
expect(thinkingSignature.length).toBeGreaterThan(0);
|
||||
|
||||
const text = response.content
|
||||
.filter((block) => block.type === "text")
|
||||
.map((block) => block.text)
|
||||
.join("")
|
||||
.trim();
|
||||
expect(text).toBe("sum=353418362; divisibleBy11=yes");
|
||||
});
|
||||
});
|
||||
@@ -1,7 +1,7 @@
|
||||
import { describe, expect, it } from "vitest";
|
||||
import { getModel } from "../src/models.js";
|
||||
import { streamSimple } from "../src/stream.js";
|
||||
import type { Context, Model } from "../src/types.js";
|
||||
import type { Context, Model, SimpleStreamOptions } from "../src/types.js";
|
||||
|
||||
interface AnthropicThinkingPayload {
|
||||
thinking?: { type: string; budget_tokens?: number };
|
||||
@@ -14,7 +14,10 @@ function makePayloadCaptureContext(): Context {
|
||||
};
|
||||
}
|
||||
|
||||
async function capturePayload(model: Model<"anthropic-messages">): Promise<AnthropicThinkingPayload> {
|
||||
async function capturePayload(
|
||||
model: Model<"anthropic-messages">,
|
||||
options?: SimpleStreamOptions,
|
||||
): Promise<AnthropicThinkingPayload> {
|
||||
let capturedPayload: AnthropicThinkingPayload | undefined;
|
||||
const payloadCaptureModel: Model<"anthropic-messages"> = {
|
||||
...model,
|
||||
@@ -22,6 +25,7 @@ async function capturePayload(model: Model<"anthropic-messages">): Promise<Anthr
|
||||
};
|
||||
|
||||
const s = streamSimple(payloadCaptureModel, makePayloadCaptureContext(), {
|
||||
...options,
|
||||
apiKey: "fake-key",
|
||||
onPayload: (payload) => {
|
||||
capturedPayload = payload as AnthropicThinkingPayload;
|
||||
@@ -113,6 +117,27 @@ describe("Anthropic thinking disable payload", () => {
|
||||
expect(payload.thinking).toEqual({ type: "disabled" });
|
||||
expect(payload.output_config).toBeUndefined();
|
||||
});
|
||||
|
||||
it("sends thinking.type=disabled for Claude Opus 4.7 when thinking is off", async () => {
|
||||
const payload = await capturePayload(getModel("anthropic", "claude-opus-4-7"));
|
||||
|
||||
expect(payload.thinking).toEqual({ type: "disabled" });
|
||||
expect(payload.output_config).toBeUndefined();
|
||||
});
|
||||
|
||||
it("uses adaptive thinking for Claude Opus 4.7 when reasoning is enabled", async () => {
|
||||
const payload = await capturePayload(getModel("anthropic", "claude-opus-4-7"), { reasoning: "high" });
|
||||
|
||||
expect(payload.thinking).toEqual({ type: "adaptive" });
|
||||
expect(payload.output_config).toEqual({ effort: "high" });
|
||||
});
|
||||
|
||||
it("maps xhigh reasoning to effort=xhigh for Claude Opus 4.7", async () => {
|
||||
const payload = await capturePayload(getModel("anthropic", "claude-opus-4-7"), { reasoning: "xhigh" });
|
||||
|
||||
expect(payload.thinking).toEqual({ type: "adaptive" });
|
||||
expect(payload.output_config).toEqual({ effort: "xhigh" });
|
||||
});
|
||||
});
|
||||
|
||||
describe.skipIf(!process.env.ANTHROPIC_API_KEY)("Anthropic thinking disable E2E", () => {
|
||||
|
||||
@@ -0,0 +1,78 @@
|
||||
import { describe, expect, it } from "vitest";
|
||||
import { getModel } from "../src/models.js";
|
||||
import { streamBedrock } from "../src/providers/amazon-bedrock.js";
|
||||
import type { Context, Model, SimpleStreamOptions } from "../src/types.js";
|
||||
|
||||
interface BedrockThinkingPayload {
|
||||
additionalModelRequestFields?: {
|
||||
thinking?: { type: string; budget_tokens?: number };
|
||||
output_config?: { effort?: string };
|
||||
anthropic_beta?: string[];
|
||||
};
|
||||
}
|
||||
|
||||
function makeContext(): Context {
|
||||
return {
|
||||
messages: [{ role: "user", content: "Hello", timestamp: Date.now() }],
|
||||
};
|
||||
}
|
||||
|
||||
async function capturePayload(
|
||||
model: Model<"bedrock-converse-stream">,
|
||||
options?: SimpleStreamOptions,
|
||||
): Promise<BedrockThinkingPayload> {
|
||||
let capturedPayload: BedrockThinkingPayload | undefined;
|
||||
const s = streamBedrock(model, makeContext(), {
|
||||
...options,
|
||||
reasoning: options?.reasoning ?? "high",
|
||||
signal: AbortSignal.abort(),
|
||||
onPayload: (payload) => {
|
||||
capturedPayload = payload as BedrockThinkingPayload;
|
||||
return payload;
|
||||
},
|
||||
});
|
||||
|
||||
for await (const event of s) {
|
||||
if (event.type === "error") {
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
if (!capturedPayload) {
|
||||
throw new Error("Expected Bedrock payload to be captured before request abort");
|
||||
}
|
||||
|
||||
return capturedPayload;
|
||||
}
|
||||
|
||||
describe("Bedrock thinking payload", () => {
|
||||
it("uses adaptive thinking for Claude Opus 4.7 when reasoning is enabled", async () => {
|
||||
const baseModel = getModel("amazon-bedrock", "global.anthropic.claude-opus-4-6-v1");
|
||||
const model: Model<"bedrock-converse-stream"> = {
|
||||
...baseModel,
|
||||
id: "global.anthropic.claude-opus-4-7-v1",
|
||||
name: "Claude Opus 4.7 (Global)",
|
||||
};
|
||||
|
||||
const payload = await capturePayload(model);
|
||||
|
||||
expect(payload.additionalModelRequestFields?.thinking).toEqual({ type: "adaptive" });
|
||||
expect(payload.additionalModelRequestFields?.output_config).toEqual({ effort: "high" });
|
||||
expect(payload.additionalModelRequestFields?.anthropic_beta).toBeUndefined();
|
||||
});
|
||||
|
||||
it("maps xhigh reasoning to effort=xhigh for Claude Opus 4.7", async () => {
|
||||
const baseModel = getModel("amazon-bedrock", "global.anthropic.claude-opus-4-6-v1");
|
||||
const model: Model<"bedrock-converse-stream"> = {
|
||||
...baseModel,
|
||||
id: "global.anthropic.claude-opus-4-7-v1",
|
||||
name: "Claude Opus 4.7 (Global)",
|
||||
};
|
||||
|
||||
const payload = await capturePayload(model, { reasoning: "xhigh" });
|
||||
|
||||
expect(payload.additionalModelRequestFields?.thinking).toEqual({ type: "adaptive" });
|
||||
expect(payload.additionalModelRequestFields?.output_config).toEqual({ effort: "xhigh" });
|
||||
expect(payload.additionalModelRequestFields?.anthropic_beta).toBeUndefined();
|
||||
});
|
||||
});
|
||||
@@ -8,6 +8,12 @@ describe("supportsXhigh", () => {
|
||||
expect(supportsXhigh(model!)).toBe(true);
|
||||
});
|
||||
|
||||
it("returns true for Anthropic Opus 4.7 on anthropic-messages API", () => {
|
||||
const model = getModel("anthropic", "claude-opus-4-7");
|
||||
expect(model).toBeDefined();
|
||||
expect(supportsXhigh(model!)).toBe(true);
|
||||
});
|
||||
|
||||
it("returns false for non-Opus Anthropic models", () => {
|
||||
const model = getModel("anthropic", "claude-sonnet-4-5");
|
||||
expect(model).toBeDefined();
|
||||
|
||||
Reference in New Issue
Block a user