alikia2x/cvsa
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix: missing import in @cvsa/core

Browse Source
This commit is contained in:
alikia2x (寒寒) 2025-06-01 21:31:37 +08:00
parent c6b7736dac
commit a1a4abff46
Signed by: alikia2x
GPG Key ID: 56209E0CCD8420C6
11 changed files with 21 additions and 368 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
packages/backend/lib/auth/getJWTsecret.ts
View File

@ -5,9 +5,10 @@ export const getJWTsecret = () => {
if (!secret) {
const response: ErrorResponse = {
message: "JWT_SECRET is not set",
code: "SERVER_ERROR"
code: "SERVER_ERROR",
errors: []
};
return [response, true];
}
return [secret, null];
}
};

3
packages/backend/middleware/rateLimiters.ts
View File

@ -43,7 +43,8 @@ export const registerRateLimiter = async (c: Context<BlankEnv, "/user", {}>, nex
if (!allowed) {
const response: ErrorResponse = {
message: `Too many requests, please retry after ${Math.round(retryAfter)} seconds.`,
code: "RATE_LIMIT_EXCEEDED"
code: "RATE_LIMIT_EXCEEDED",
errors: []
};
return c.json<ErrorResponse>(response, 429);
}

5
packages/backend/routes/captcha/difficulty/GET.ts
View File

@ -16,12 +16,13 @@ export const getCaptchaDifficultyHandler = createHandlers(async (c) => {
if (!difficulty) {
const response: ErrorResponse<unknown> = {
code: "ENTITY_NOT_FOUND",
message: "No difficulty configs found for this route."
message: "No difficulty configs found for this route.",
errors: []
};
return c.json<ErrorResponse<unknown>>(response, 404);
}
return c.json({
"difficulty": difficulty
difficulty: difficulty
});
} catch (e: unknown) {
if (e instanceof ValidationError) {

2
packages/core/package.json
View File

@ -1,7 +1,7 @@
{
"name": "@cvsa/core",
"private": false,
"version": "0.0.1",
"version": "0.0.2",
"scripts": {
"test": "bun --env-file=.env.test run vitest",
"build": "bun build ./index.ts --target node --outdir ./dist"

1
packages/core/types.d.ts vendored
View File

@ -1 +1,2 @@
export * from "./db/schema";
export * from "./index";

179
packages/crawler/ml/benchmark.ts
View File

@ -1,179 +0,0 @@
import { AutoTokenizer, PreTrainedTokenizer } from "@huggingface/transformers";
import * as ort from "onnxruntime";
function softmax(logits: Float32Array): number[] {
const maxLogit = Math.max(...logits);
const exponents = logits.map((logit) => Math.exp(logit - maxLogit));
const sumOfExponents = exponents.reduce((sum, exp) => sum + exp, 0);
return Array.from(exponents.map((exp) => exp / sumOfExponents));
}
// 配置参数
const sentenceTransformerModelName = "alikia2x/jina-embedding-v3-m2v-1024";
const onnxClassifierPath = "./model/video_classifier_v3_17.onnx";
const onnxEmbeddingPath = "./model/embedding_original.onnx";
const testDataPath = "./data/filter/test1.jsonl";
// 初始化会话
const [sessionClassifier, sessionEmbedding] = await Promise.all([
ort.InferenceSession.create(onnxClassifierPath),
ort.InferenceSession.create(onnxEmbeddingPath),
]);
let tokenizer: PreTrainedTokenizer;
// 初始化分词器
async function loadTokenizer() {
const tokenizerConfig = { local_files_only: true };
tokenizer = await AutoTokenizer.from_pretrained(sentenceTransformerModelName, tokenizerConfig);
}
// 新的嵌入生成函数（使用ONNX）
async function getONNXEmbeddings(texts: string[], session: ort.InferenceSession): Promise<number[]> {
const { input_ids } = await tokenizer(texts, {
add_special_tokens: false,
return_tensor: false,
});
// 构造输入参数
const cumsum = (arr: number[]): number[] =>
arr.reduce((acc: number[], num: number, i: number) => [...acc, num + (acc[i - 1] || 0)], []);
const offsets: number[] = [0, ...cumsum(input_ids.slice(0, -1).map((x: string) => x.length))];
const flattened_input_ids = input_ids.flat();
// 准备ONNX输入
const inputs = {
input_ids: new ort.Tensor("int64", new BigInt64Array(flattened_input_ids.map(BigInt)), [
flattened_input_ids.length,
]),
offsets: new ort.Tensor("int64", new BigInt64Array(offsets.map(BigInt)), [offsets.length]),
};
// 执行推理
const { embeddings } = await session.run(inputs);
return Array.from(embeddings.data as Float32Array);
}
// 分类推理函数
async function runClassification(embeddings: number[]): Promise<number[]> {
const inputTensor = new ort.Tensor(
Float32Array.from(embeddings),
[1, 3, 1024],
);
const { logits } = await sessionClassifier.run({ channel_features: inputTensor });
return softmax(logits.data as Float32Array);
}
// 指标计算函数
function calculateMetrics(labels: number[], predictions: number[], elapsedTime: number): {
accuracy: number;
precision: number;
recall: number;
f1: number;
"Class 0 Prec": number;
speed: string;
} {
// 输出label和prediction不一样的index列表
const arr = [];
for (let i = 0; i < labels.length; i++) {
if (labels[i] !== predictions[i] && predictions[i] == 0) {
arr.push([i + 1, labels[i], predictions[i]]);
}
}
console.log(arr);
// 初始化混淆矩阵
const classCount = Math.max(...labels, ...predictions) + 1;
const matrix = Array.from({ length: classCount }, () => Array.from({ length: classCount }, () => 0));
// 填充矩阵
labels.forEach((trueLabel, i) => {
matrix[trueLabel][predictions[i]]++;
});
// 计算各指标
let totalTP = 0, totalFP = 0, totalFN = 0;
for (let c = 0; c < classCount; c++) {
const TP = matrix[c][c];
const FP = matrix.flatMap((row, i) => i === c ? [] : [row[c]]).reduce((a, b) => a + b, 0);
const FN = matrix[c].filter((_, i) => i !== c).reduce((a, b) => a + b, 0);
totalTP += TP;
totalFP += FP;
totalFN += FN;
}
const precision = totalTP / (totalTP + totalFP);
const recall = totalTP / (totalTP + totalFN);
const f1 = 2 * (precision * recall) / (precision + recall) || 0;
// 计算Class 0 Precision
const class0TP = matrix[0][0];
const class0FP = matrix.flatMap((row, i) => i === 0 ? [] : [row[0]]).reduce((a, b) => a + b, 0);
const class0Precision = class0TP / (class0TP + class0FP) || 0;
return {
accuracy: labels.filter((l, i) => l === predictions[i]).length / labels.length,
precision,
recall,
f1,
speed: `${(labels.length / (elapsedTime / 1000)).toFixed(1)} samples/sec`,
"Class 0 Prec": class0Precision,
};
}
// 改造后的评估函数
async function evaluateModel(session: ort.InferenceSession): Promise<{
accuracy: number;
precision: number;
recall: number;
f1: number;
"Class 0 Prec": number;
}> {
const data = await Deno.readTextFile(testDataPath);
const samples = data.split("\n")
.map((line) => {
try {
return JSON.parse(line);
} catch {
return null;
}
})
.filter(Boolean);
const allPredictions: number[] = [];
const allLabels: number[] = [];
const t = new Date().getTime();
for (const sample of samples) {
try {
const embeddings = await getONNXEmbeddings([
sample.title,
sample.description,
sample.tags.join(","),
], session);
const probabilities = await runClassification(embeddings);
allPredictions.push(probabilities.indexOf(Math.max(...probabilities)));
allLabels.push(sample.label);
} catch (error) {
console.error("Processing error:", error);
}
}
const elapsed = new Date().getTime() - t;
return calculateMetrics(allLabels, allPredictions, elapsed);
}
// 主函数
async function main() {
await loadTokenizer();
const metrics = await evaluateModel(sessionEmbedding);
console.log("Model Metrics:");
console.table(metrics);
}
await main();

171
packages/crawler/ml/quant_benchmark.ts
View File

@ -1,171 +0,0 @@
import { AutoTokenizer, PreTrainedTokenizer } from "@huggingface/transformers";
import * as ort from "onnxruntime";
function softmax(logits: Float32Array): number[] {
const maxLogit = Math.max(...logits);
const exponents = logits.map((logit) => Math.exp(logit - maxLogit));
const sumOfExponents = exponents.reduce((sum, exp) => sum + exp, 0);
return Array.from(exponents.map((exp) => exp / sumOfExponents));
}
// 配置参数
const sentenceTransformerModelName = "alikia2x/jina-embedding-v3-m2v-1024";
const onnxClassifierPath = "./model/video_classifier_v3_11.onnx";
const onnxEmbeddingOriginalPath = "./model/embedding_original.onnx";
const onnxEmbeddingQuantizedPath = "./model/embedding_original.onnx";
// 初始化会话
const [sessionClassifier, sessionEmbeddingOriginal, sessionEmbeddingQuantized] = await Promise.all([
ort.InferenceSession.create(onnxClassifierPath),
ort.InferenceSession.create(onnxEmbeddingOriginalPath),
ort.InferenceSession.create(onnxEmbeddingQuantizedPath),
]);
let tokenizer: PreTrainedTokenizer;
// 初始化分词器
async function loadTokenizer() {
const tokenizerConfig = { local_files_only: true };
tokenizer = await AutoTokenizer.from_pretrained(sentenceTransformerModelName, tokenizerConfig);
}
// 新的嵌入生成函数（使用ONNX）
async function getONNXEmbeddings(texts: string[], session: ort.InferenceSession): Promise<number[]> {
const { input_ids } = await tokenizer(texts, {
add_special_tokens: false,
return_tensor: false,
});
// 构造输入参数
const cumsum = (arr: number[]): number[] =>
arr.reduce((acc: number[], num: number, i: number) => [...acc, num + (acc[i - 1] || 0)], []);
const offsets: number[] = [0, ...cumsum(input_ids.slice(0, -1).map((x: string) => x.length))];
const flattened_input_ids = input_ids.flat();
// 准备ONNX输入
const inputs = {
input_ids: new ort.Tensor("int64", new BigInt64Array(flattened_input_ids.map(BigInt)), [
flattened_input_ids.length,
]),
offsets: new ort.Tensor("int64", new BigInt64Array(offsets.map(BigInt)), [offsets.length]),
};
// 执行推理
const { embeddings } = await session.run(inputs);
return Array.from(embeddings.data as Float32Array);
}
// 分类推理函数
async function runClassification(embeddings: number[]): Promise<number[]> {
const inputTensor = new ort.Tensor(
Float32Array.from(embeddings),
[1, 4, 1024],
);
const { logits } = await sessionClassifier.run({ channel_features: inputTensor });
return softmax(logits.data as Float32Array);
}
// 指标计算函数
function calculateMetrics(labels: number[], predictions: number[], elapsedTime: number): {
accuracy: number;
precision: number;
recall: number;
f1: number;
speed: string;
} {
// 初始化混淆矩阵
const classCount = Math.max(...labels, ...predictions) + 1;
const matrix = Array.from({ length: classCount }, () => Array.from({ length: classCount }, () => 0));
// 填充矩阵
labels.forEach((trueLabel, i) => {
matrix[trueLabel][predictions[i]]++;
});
// 计算各指标
let totalTP = 0, totalFP = 0, totalFN = 0;
for (let c = 0; c < classCount; c++) {
const TP = matrix[c][c];
const FP = matrix.flatMap((row, i) => i === c ? [] : [row[c]]).reduce((a, b) => a + b, 0);
const FN = matrix[c].filter((_, i) => i !== c).reduce((a, b) => a + b, 0);
totalTP += TP;
totalFP += FP;
totalFN += FN;
}
const precision = totalTP / (totalTP + totalFP);
const recall = totalTP / (totalTP + totalFN);
const f1 = 2 * (precision * recall) / (precision + recall) || 0;
return {
accuracy: labels.filter((l, i) => l === predictions[i]).length / labels.length,
precision,
recall,
f1,
speed: `${(labels.length / (elapsedTime / 1000)).toFixed(1)} samples/sec`,
};
}
// 改造后的评估函数
async function evaluateModel(session: ort.InferenceSession): Promise<{
accuracy: number;
precision: number;
recall: number;
f1: number;
}> {
const data = await Deno.readTextFile("./data/filter/test1.jsonl");
const samples = data.split("\n")
.map((line) => {
try {
return JSON.parse(line);
} catch {
return null;
}
})
.filter(Boolean);
const allPredictions: number[] = [];
const allLabels: number[] = [];
const t = new Date().getTime();
for (const sample of samples) {
try {
const embeddings = await getONNXEmbeddings([
sample.title,
sample.description,
sample.tags.join(","),
sample.author_info,
], session);
const probabilities = await runClassification(embeddings);
allPredictions.push(probabilities.indexOf(Math.max(...probabilities)));
allLabels.push(sample.label);
} catch (error) {
console.error("Processing error:", error);
}
}
const elapsed = new Date().getTime() - t;
return calculateMetrics(allLabels, allPredictions, elapsed);
}
// 主函数
async function main() {
await loadTokenizer();
// 评估原始模型
const originalMetrics = await evaluateModel(sessionEmbeddingOriginal);
console.log("Original Model Metrics:");
console.table(originalMetrics);
// 评估量化模型
const quantizedMetrics = await evaluateModel(sessionEmbeddingQuantized);
console.log("Quantized Model Metrics:");
console.table(quantizedMetrics);
}
await main();

13
packages/crawler/mq/scheduling.ts
View File

@ -2,7 +2,7 @@ import { findClosestSnapshot, getLatestSnapshot, hasAtLeast2Snapshots } from "db
import { truncate } from "utils/truncate.ts";
import { closetMilestone } from "./exec/snapshotTick.ts";
import { HOUR, MINUTE } from "@core/const/time.ts";
import type { Psql } from "@core/db/global.d.ts";
import type { Psql } from "@core/db/psql.d.ts";
const log = (value: number, base: number = 10) => Math.log(value) / Math.log(base);
@ -12,13 +12,12 @@ const getFactor = (x: number) => {
const c = 100;
const u = 0.601;
const g = 455;
if (x>g) {
return log(b/log(x+1),a);
if (x > g) {
return log(b / log(x + 1), a);
} else {
return log(b / log(x + c), a) + u;
}
else {
return log(b/log(x+c),a)+u;
}
}
};
/*
* Returns the minimum ETA in hours for the next snapshot

4
packages/crawler/mq/task/regularSnapshotInterval.ts
View File

@ -1,6 +1,6 @@
import { findClosestSnapshot, findSnapshotBefore, getLatestSnapshot } from "db/snapshotSchedule.ts";
import { HOUR } from "@core/const/time.ts";
import type { Psql } from "@core/db/global.d.ts";
import type { Psql } from "@core/db/psql";
export const getRegularSnapshotInterval = async (sql: Psql, aid: number) => {
const now = Date.now();
@ -14,7 +14,7 @@ export const getRegularSnapshotInterval = async (sql: Psql, aid: number) => {
if (hoursDiff < 8) return 24;
const viewsDiff = latestSnapshot.views - oldSnapshot.views;
if (viewsDiff === 0) return 72;
const speedPerDay = viewsDiff / (hoursDiff + 0.001) * 24;
const speedPerDay = (viewsDiff / (hoursDiff + 0.001)) * 24;
if (speedPerDay < 6) return 36;
if (speedPerDay < 120) return 24;
if (speedPerDay < 320) return 12;

4
packages/next/bun.lock
View File

@ -4,7 +4,7 @@
"": {
"name": "next",
"dependencies": {
"@cvsa/core": "^0.0.1",
"@cvsa/core": "^0.0.2",
"axios": "^1.9.0",
"framer-motion": "^12.15.0",
"i18next": "^25.2.1",
@ -34,7 +34,7 @@
"@colors/colors": ["@colors/colors@1.6.0", "", {}, "sha512-Ir+AOibqzrIsL6ajt3Rz3LskB7OiMVHqltZmspbW/TJuTVuyOMirVqAkjfY6JISiLHgyNqicAC8AyHHGzNd/dA=="],
"@cvsa/core": ["@cvsa/core@0.0.1", "", { "dependencies": { "@koshnic/ratelimit": "^1.0.3", "chalk": "^5.4.1", "ioredis": "^5.6.1", "logform": "^2.7.0", "postgres": "^3.4.5", "winston": "^3.17.0" } }, "sha512-h7p2AHcvdIA7GCJq4k1sOSGGbs/qjdHa4WlcCh6p1rVgpkpXp6v1Q9lvXca3uqAkInwzXctDSGwKiQp65K5XOg=="],
"@cvsa/core": ["@cvsa/core@0.0.2", "", { "dependencies": { "@koshnic/ratelimit": "^1.0.3", "chalk": "^5.4.1", "ioredis": "^5.6.1", "logform": "^2.7.0", "postgres": "^3.4.5", "winston": "^3.17.0" } }, "sha512-SKiFZYk3+DUCx31R+yFlcMb9S6tbdQdSSV2H+cPNgmCoOcEbBcZvB99iG4vy7wpKOcPy1bDVvUEmIMo3nIxCbQ=="],
"@dabh/diagnostics": ["@dabh/diagnostics@2.0.3", "", { "dependencies": { "colorspace": "1.1.x", "enabled": "2.0.x", "kuler": "^2.0.0" } }, "sha512-hrlQOIi7hAfzsMqlGSFyVucrx38O+j6wiGOf//H2ecvIEqYN4ADBSS2iLMh5UFyDunCNniUIPk/q3riFv45xRA=="],

2
packages/next/package.json
View File

@ -10,7 +10,7 @@
"format": "prettier --write ."
},
"dependencies": {
"@cvsa/core": "^0.0.1",
"@cvsa/core": "^0.0.2",
"axios": "^1.9.0",
"framer-motion": "^12.15.0",
"i18next": "^25.2.1",