- Update test.invoices.nanonets.ts to use DualAgentOrchestrator for JSON extraction - Enable streaming token callback for real-time progress visibility - Add markdown caching to avoid re-running Nanonets OCR for cached files - Update test.bankstatements.minicpm.ts and test.invoices.minicpm.ts with streaming - Update dependencies to @push.rocks/smartai@0.11.1 and @push.rocks/smartagent@1.2.8
561 lines
18 KiB
TypeScript
561 lines
18 KiB
TypeScript
/**
|
|
* Invoice extraction test using MiniCPM-V (visual extraction)
|
|
*
|
|
* Consensus approach:
|
|
* 1. Pass 1: Fast JSON extraction
|
|
* 2. Pass 2: Confirm with thinking enabled
|
|
* 3. If mismatch: repeat until consensus or max attempts
|
|
*/
|
|
import { tap, expect } from '@git.zone/tstest/tapbundle';
|
|
import * as fs from 'fs';
|
|
import * as path from 'path';
|
|
import { execSync } from 'child_process';
|
|
import * as os from 'os';
|
|
import { ensureMiniCpm } from './helpers/docker.js';
|
|
|
|
const OLLAMA_URL = 'http://localhost:11434';
|
|
const MODEL = 'openbmb/minicpm-v4.5:q8_0';
|
|
|
|
interface IInvoice {
|
|
invoice_number: string;
|
|
invoice_date: string;
|
|
vendor_name: string;
|
|
currency: string;
|
|
net_amount: number;
|
|
vat_amount: number;
|
|
total_amount: number;
|
|
}
|
|
|
|
/**
|
|
* Convert PDF to PNG images using ImageMagick
|
|
*/
|
|
function convertPdfToImages(pdfPath: string): string[] {
|
|
const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'pdf-convert-'));
|
|
const outputPattern = path.join(tempDir, 'page-%d.png');
|
|
|
|
try {
|
|
execSync(
|
|
`convert -density 300 -quality 95 "${pdfPath}" -background white -alpha remove "${outputPattern}"`,
|
|
{ stdio: 'pipe' }
|
|
);
|
|
|
|
const files = fs.readdirSync(tempDir).filter((f) => f.endsWith('.png')).sort();
|
|
const images: string[] = [];
|
|
|
|
for (const file of files) {
|
|
const imagePath = path.join(tempDir, file);
|
|
const imageData = fs.readFileSync(imagePath);
|
|
images.push(imageData.toString('base64'));
|
|
}
|
|
|
|
return images;
|
|
} finally {
|
|
fs.rmSync(tempDir, { recursive: true, force: true });
|
|
}
|
|
}
|
|
|
|
const JSON_PROMPT = `Extract invoice data from this image. Return ONLY a JSON object with these exact fields:
|
|
{
|
|
"invoice_number": "the invoice number (not VAT ID, not customer ID)",
|
|
"invoice_date": "YYYY-MM-DD format",
|
|
"vendor_name": "company that issued the invoice",
|
|
"currency": "EUR, USD, or GBP",
|
|
"net_amount": 0.00,
|
|
"vat_amount": 0.00,
|
|
"total_amount": 0.00
|
|
}
|
|
Return only the JSON, no explanation.`;
|
|
|
|
/**
|
|
* Query MiniCPM-V for JSON output (fast, no thinking) with streaming
|
|
*/
|
|
async function queryJsonFast(images: string[]): Promise<string> {
|
|
const startTime = Date.now();
|
|
process.stdout.write(` [Fast] `);
|
|
|
|
const response = await fetch(`${OLLAMA_URL}/api/chat`, {
|
|
method: 'POST',
|
|
headers: { 'Content-Type': 'application/json' },
|
|
body: JSON.stringify({
|
|
model: MODEL,
|
|
messages: [{
|
|
role: 'user',
|
|
content: JSON_PROMPT,
|
|
images: images,
|
|
}],
|
|
stream: true,
|
|
options: {
|
|
num_ctx: 32768,
|
|
num_predict: 1000,
|
|
temperature: 0.1,
|
|
},
|
|
}),
|
|
});
|
|
|
|
if (!response.ok) {
|
|
throw new Error(`Ollama API error: ${response.status}`);
|
|
}
|
|
|
|
let content = '';
|
|
const reader = response.body!.getReader();
|
|
const decoder = new TextDecoder();
|
|
|
|
try {
|
|
while (true) {
|
|
const { done, value } = await reader.read();
|
|
if (done) break;
|
|
|
|
const chunk = decoder.decode(value, { stream: true });
|
|
for (const line of chunk.split('\n').filter(l => l.trim())) {
|
|
try {
|
|
const json = JSON.parse(line);
|
|
const token = json.message?.content || '';
|
|
if (token) {
|
|
process.stdout.write(token);
|
|
content += token;
|
|
}
|
|
} catch {
|
|
// Ignore parse errors for partial chunks
|
|
}
|
|
}
|
|
}
|
|
} finally {
|
|
const elapsed = ((Date.now() - startTime) / 1000).toFixed(1);
|
|
process.stdout.write(` (${elapsed}s)\n`);
|
|
}
|
|
|
|
return content.trim();
|
|
}
|
|
|
|
/**
|
|
* Query MiniCPM-V for JSON output with thinking enabled (slower, more accurate) with streaming
|
|
*/
|
|
async function queryJsonWithThinking(images: string[]): Promise<string> {
|
|
const startTime = Date.now();
|
|
process.stdout.write(` [Think] `);
|
|
|
|
const response = await fetch(`${OLLAMA_URL}/api/chat`, {
|
|
method: 'POST',
|
|
headers: { 'Content-Type': 'application/json' },
|
|
body: JSON.stringify({
|
|
model: MODEL,
|
|
messages: [{
|
|
role: 'user',
|
|
content: `Think carefully about this invoice image, then ${JSON_PROMPT}`,
|
|
images: images,
|
|
}],
|
|
stream: true,
|
|
options: {
|
|
num_ctx: 32768,
|
|
num_predict: 2000,
|
|
temperature: 0.1,
|
|
},
|
|
}),
|
|
});
|
|
|
|
if (!response.ok) {
|
|
throw new Error(`Ollama API error: ${response.status}`);
|
|
}
|
|
|
|
let content = '';
|
|
let thinkingContent = '';
|
|
let thinkingStarted = false;
|
|
let outputStarted = false;
|
|
const reader = response.body!.getReader();
|
|
const decoder = new TextDecoder();
|
|
|
|
try {
|
|
while (true) {
|
|
const { done, value } = await reader.read();
|
|
if (done) break;
|
|
|
|
const chunk = decoder.decode(value, { stream: true });
|
|
for (const line of chunk.split('\n').filter(l => l.trim())) {
|
|
try {
|
|
const json = JSON.parse(line);
|
|
|
|
// Stream thinking tokens
|
|
const thinking = json.message?.thinking || '';
|
|
if (thinking) {
|
|
if (!thinkingStarted) {
|
|
process.stdout.write(`THINKING: `);
|
|
thinkingStarted = true;
|
|
}
|
|
process.stdout.write(thinking);
|
|
thinkingContent += thinking;
|
|
}
|
|
|
|
// Stream content tokens
|
|
const token = json.message?.content || '';
|
|
if (token) {
|
|
if (!outputStarted) {
|
|
if (thinkingStarted) process.stdout.write('\n [Think] ');
|
|
process.stdout.write(`OUTPUT: `);
|
|
outputStarted = true;
|
|
}
|
|
process.stdout.write(token);
|
|
content += token;
|
|
}
|
|
} catch {
|
|
// Ignore parse errors for partial chunks
|
|
}
|
|
}
|
|
}
|
|
} finally {
|
|
const elapsed = ((Date.now() - startTime) / 1000).toFixed(1);
|
|
process.stdout.write(` (${elapsed}s)\n`);
|
|
}
|
|
|
|
return content.trim();
|
|
}
|
|
|
|
/**
|
|
* Parse amount from string (handles European format)
|
|
*/
|
|
function parseAmount(s: string | number | undefined): number {
|
|
if (s === undefined || s === null) return 0;
|
|
if (typeof s === 'number') return s;
|
|
const match = s.match(/([\d.,]+)/);
|
|
if (!match) return 0;
|
|
const numStr = match[1];
|
|
// Handle European format: 1.234,56 → 1234.56
|
|
const normalized = numStr.includes(',') && numStr.indexOf(',') > numStr.lastIndexOf('.')
|
|
? numStr.replace(/\./g, '').replace(',', '.')
|
|
: numStr.replace(/,/g, '');
|
|
return parseFloat(normalized) || 0;
|
|
}
|
|
|
|
/**
|
|
* Extract invoice number from potentially verbose response
|
|
*/
|
|
function extractInvoiceNumber(s: string | undefined): string {
|
|
if (!s) return '';
|
|
let clean = s.replace(/\*\*/g, '').replace(/`/g, '').trim();
|
|
const patterns = [
|
|
/\b([A-Z]{2,3}\d{10,})\b/i, // IEE2022006460244
|
|
/\b([A-Z]\d{8,})\b/i, // R0014359508
|
|
/\b(INV[-\s]?\d{4}[-\s]?\d+)\b/i, // INV-2024-001
|
|
/\b(\d{7,})\b/, // 1579087430
|
|
];
|
|
for (const pattern of patterns) {
|
|
const match = clean.match(pattern);
|
|
if (match) return match[1];
|
|
}
|
|
return clean.replace(/[^A-Z0-9-]/gi, '').trim() || clean;
|
|
}
|
|
|
|
/**
|
|
* Extract date (YYYY-MM-DD) from response
|
|
*/
|
|
function extractDate(s: string | undefined): string {
|
|
if (!s) return '';
|
|
let clean = s.replace(/\*\*/g, '').replace(/`/g, '').trim();
|
|
const isoMatch = clean.match(/(\d{4}-\d{2}-\d{2})/);
|
|
if (isoMatch) return isoMatch[1];
|
|
// Try DD/MM/YYYY or DD.MM.YYYY
|
|
const dmyMatch = clean.match(/(\d{1,2})[\/.](\d{1,2})[\/.](\d{4})/);
|
|
if (dmyMatch) {
|
|
return `${dmyMatch[3]}-${dmyMatch[2].padStart(2, '0')}-${dmyMatch[1].padStart(2, '0')}`;
|
|
}
|
|
return clean.replace(/[^\d-]/g, '').trim();
|
|
}
|
|
|
|
/**
|
|
* Extract currency
|
|
*/
|
|
function extractCurrency(s: string | undefined): string {
|
|
if (!s) return 'EUR';
|
|
const upper = s.toUpperCase();
|
|
if (upper.includes('EUR') || upper.includes('€')) return 'EUR';
|
|
if (upper.includes('USD') || upper.includes('$')) return 'USD';
|
|
if (upper.includes('GBP') || upper.includes('£')) return 'GBP';
|
|
return 'EUR';
|
|
}
|
|
|
|
/**
|
|
* Extract JSON from response (handles markdown code blocks)
|
|
*/
|
|
function extractJsonFromResponse(response: string): Record<string, unknown> | null {
|
|
// Try to find JSON in markdown code block
|
|
const codeBlockMatch = response.match(/```(?:json)?\s*([\s\S]*?)```/);
|
|
const jsonStr = codeBlockMatch ? codeBlockMatch[1].trim() : response.trim();
|
|
|
|
try {
|
|
return JSON.parse(jsonStr);
|
|
} catch {
|
|
// Try to find JSON object pattern
|
|
const jsonMatch = jsonStr.match(/\{[\s\S]*\}/);
|
|
if (jsonMatch) {
|
|
try {
|
|
return JSON.parse(jsonMatch[0]);
|
|
} catch {
|
|
return null;
|
|
}
|
|
}
|
|
return null;
|
|
}
|
|
}
|
|
|
|
/**
|
|
* Parse JSON response into IInvoice
|
|
*/
|
|
function parseJsonToInvoice(response: string): IInvoice | null {
|
|
const parsed = extractJsonFromResponse(response);
|
|
if (!parsed) return null;
|
|
|
|
return {
|
|
invoice_number: extractInvoiceNumber(String(parsed.invoice_number || '')),
|
|
invoice_date: extractDate(String(parsed.invoice_date || '')),
|
|
vendor_name: String(parsed.vendor_name || '').replace(/\*\*/g, '').replace(/`/g, '').trim(),
|
|
currency: extractCurrency(String(parsed.currency || '')),
|
|
net_amount: parseAmount(parsed.net_amount as string | number),
|
|
vat_amount: parseAmount(parsed.vat_amount as string | number),
|
|
total_amount: parseAmount(parsed.total_amount as string | number),
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Compare two invoices for consensus (key fields must match)
|
|
*/
|
|
function invoicesMatch(a: IInvoice, b: IInvoice): boolean {
|
|
const numMatch = a.invoice_number.toLowerCase() === b.invoice_number.toLowerCase();
|
|
const dateMatch = a.invoice_date === b.invoice_date;
|
|
const totalMatch = Math.abs(a.total_amount - b.total_amount) < 0.02;
|
|
return numMatch && dateMatch && totalMatch;
|
|
}
|
|
|
|
/**
|
|
* Extract invoice data using consensus approach:
|
|
* 1. Pass 1: Fast JSON extraction
|
|
* 2. Pass 2: Confirm with thinking enabled
|
|
* 3. If mismatch: repeat until consensus or max 5 attempts
|
|
*/
|
|
async function extractInvoiceFromImages(images: string[]): Promise<IInvoice> {
|
|
console.log(` [Vision] Processing ${images.length} page(s) with ${MODEL} (consensus)`);
|
|
|
|
const MAX_ATTEMPTS = 5;
|
|
let attempt = 0;
|
|
|
|
while (attempt < MAX_ATTEMPTS) {
|
|
attempt++;
|
|
console.log(` [Attempt ${attempt}/${MAX_ATTEMPTS}]`);
|
|
|
|
// PASS 1: Fast JSON extraction
|
|
console.log(` [Pass 1] Fast extraction...`);
|
|
const fastResponse = await queryJsonFast(images);
|
|
const fastInvoice = parseJsonToInvoice(fastResponse);
|
|
|
|
if (!fastInvoice) {
|
|
console.log(` [Pass 1] JSON parsing failed, retrying...`);
|
|
continue;
|
|
}
|
|
console.log(` [Pass 1] Result: ${fastInvoice.invoice_number} | ${fastInvoice.invoice_date} | ${fastInvoice.total_amount} ${fastInvoice.currency}`);
|
|
|
|
// PASS 2: Confirm with thinking
|
|
console.log(` [Pass 2] Thinking confirmation...`);
|
|
const thinkResponse = await queryJsonWithThinking(images);
|
|
const thinkInvoice = parseJsonToInvoice(thinkResponse);
|
|
|
|
if (!thinkInvoice) {
|
|
console.log(` [Pass 2] JSON parsing failed, retrying...`);
|
|
continue;
|
|
}
|
|
console.log(` [Pass 2] Result: ${thinkInvoice.invoice_number} | ${thinkInvoice.invoice_date} | ${thinkInvoice.total_amount} ${thinkInvoice.currency}`);
|
|
|
|
// Check consensus
|
|
if (invoicesMatch(fastInvoice, thinkInvoice)) {
|
|
console.log(` [Consensus] MATCH - using result`);
|
|
return thinkInvoice; // Prefer thinking result
|
|
}
|
|
|
|
console.log(` [Consensus] MISMATCH - repeating...`);
|
|
console.log(` Fast: ${fastInvoice.invoice_number} | ${fastInvoice.invoice_date} | ${fastInvoice.total_amount}`);
|
|
console.log(` Think: ${thinkInvoice.invoice_number} | ${thinkInvoice.invoice_date} | ${thinkInvoice.total_amount}`);
|
|
}
|
|
|
|
// Max attempts reached - do one final thinking pass and use that
|
|
console.log(` [Final] Max attempts reached, using final thinking pass`);
|
|
const finalResponse = await queryJsonWithThinking(images);
|
|
const finalInvoice = parseJsonToInvoice(finalResponse);
|
|
|
|
if (finalInvoice) {
|
|
console.log(` [Final] Result: ${finalInvoice.invoice_number} | ${finalInvoice.invoice_date} | ${finalInvoice.total_amount} ${finalInvoice.currency}`);
|
|
return finalInvoice;
|
|
}
|
|
|
|
// Return empty invoice if all else fails
|
|
console.log(` [Final] All parsing failed, returning empty`);
|
|
return {
|
|
invoice_number: '',
|
|
invoice_date: '',
|
|
vendor_name: '',
|
|
currency: 'EUR',
|
|
net_amount: 0,
|
|
vat_amount: 0,
|
|
total_amount: 0,
|
|
};
|
|
}
|
|
|
|
/**
|
|
* Normalize date to YYYY-MM-DD
|
|
*/
|
|
function normalizeDate(dateStr: string | null): string {
|
|
if (!dateStr) return '';
|
|
if (/^\d{4}-\d{2}-\d{2}$/.test(dateStr)) return dateStr;
|
|
|
|
const monthMap: Record<string, string> = {
|
|
JAN: '01', FEB: '02', MAR: '03', APR: '04', MAY: '05', JUN: '06',
|
|
JUL: '07', AUG: '08', SEP: '09', OCT: '10', NOV: '11', DEC: '12',
|
|
};
|
|
|
|
let match = dateStr.match(/^(\d{1,2})-([A-Z]{3})-(\d{4})$/i);
|
|
if (match) {
|
|
return `${match[3]}-${monthMap[match[2].toUpperCase()] || '01'}-${match[1].padStart(2, '0')}`;
|
|
}
|
|
|
|
match = dateStr.match(/^(\d{1,2})[\/.](\d{1,2})[\/.](\d{4})$/);
|
|
if (match) {
|
|
return `${match[3]}-${match[2].padStart(2, '0')}-${match[1].padStart(2, '0')}`;
|
|
}
|
|
|
|
return dateStr;
|
|
}
|
|
|
|
/**
|
|
* Compare extracted invoice against expected
|
|
*/
|
|
function compareInvoice(
|
|
extracted: IInvoice,
|
|
expected: IInvoice
|
|
): { match: boolean; errors: string[] } {
|
|
const errors: string[] = [];
|
|
|
|
// Compare invoice number (normalize by removing spaces and case)
|
|
const extNum = extracted.invoice_number?.replace(/\s/g, '').toLowerCase() || '';
|
|
const expNum = expected.invoice_number?.replace(/\s/g, '').toLowerCase() || '';
|
|
if (extNum !== expNum) {
|
|
errors.push(`invoice_number: expected "${expected.invoice_number}", got "${extracted.invoice_number}"`);
|
|
}
|
|
|
|
// Compare date
|
|
if (normalizeDate(extracted.invoice_date) !== normalizeDate(expected.invoice_date)) {
|
|
errors.push(`invoice_date: expected "${expected.invoice_date}", got "${extracted.invoice_date}"`);
|
|
}
|
|
|
|
// Compare total amount (with tolerance)
|
|
if (Math.abs(extracted.total_amount - expected.total_amount) > 0.02) {
|
|
errors.push(`total_amount: expected ${expected.total_amount}, got ${extracted.total_amount}`);
|
|
}
|
|
|
|
// Compare currency
|
|
if (extracted.currency?.toUpperCase() !== expected.currency?.toUpperCase()) {
|
|
errors.push(`currency: expected "${expected.currency}", got "${extracted.currency}"`);
|
|
}
|
|
|
|
return { match: errors.length === 0, errors };
|
|
}
|
|
|
|
/**
|
|
* Find all test cases (PDF + JSON pairs) in .nogit/invoices/
|
|
*/
|
|
function findTestCases(): Array<{ name: string; pdfPath: string; jsonPath: string }> {
|
|
const testDir = path.join(process.cwd(), '.nogit/invoices');
|
|
if (!fs.existsSync(testDir)) {
|
|
return [];
|
|
}
|
|
|
|
const files = fs.readdirSync(testDir);
|
|
const pdfFiles = files.filter((f) => f.endsWith('.pdf'));
|
|
const testCases: Array<{ name: string; pdfPath: string; jsonPath: string }> = [];
|
|
|
|
for (const pdf of pdfFiles) {
|
|
const baseName = pdf.replace('.pdf', '');
|
|
const jsonFile = `${baseName}.json`;
|
|
if (files.includes(jsonFile)) {
|
|
testCases.push({
|
|
name: baseName,
|
|
pdfPath: path.join(testDir, pdf),
|
|
jsonPath: path.join(testDir, jsonFile),
|
|
});
|
|
}
|
|
}
|
|
|
|
testCases.sort((a, b) => a.name.localeCompare(b.name));
|
|
return testCases;
|
|
}
|
|
|
|
// Tests
|
|
|
|
tap.test('setup: ensure Docker containers are running', async () => {
|
|
console.log('\n[Setup] Checking Docker containers...\n');
|
|
const minicpmOk = await ensureMiniCpm();
|
|
expect(minicpmOk).toBeTrue();
|
|
console.log('\n[Setup] All containers ready!\n');
|
|
});
|
|
|
|
tap.test('should have MiniCPM-V model loaded', async () => {
|
|
const response = await fetch(`${OLLAMA_URL}/api/tags`);
|
|
const data = await response.json();
|
|
const modelNames = data.models.map((m: { name: string }) => m.name);
|
|
expect(modelNames.some((name: string) => name.includes('minicpm'))).toBeTrue();
|
|
});
|
|
|
|
const testCases = findTestCases();
|
|
console.log(`\nFound ${testCases.length} invoice test cases (MiniCPM-V)\n`);
|
|
|
|
let passedCount = 0;
|
|
let failedCount = 0;
|
|
const processingTimes: number[] = [];
|
|
|
|
for (const testCase of testCases) {
|
|
tap.test(`should extract invoice: ${testCase.name}`, async () => {
|
|
const expected: IInvoice = JSON.parse(fs.readFileSync(testCase.jsonPath, 'utf-8'));
|
|
console.log(`\n=== ${testCase.name} ===`);
|
|
console.log(`Expected: ${expected.invoice_number} | ${expected.invoice_date} | ${expected.total_amount} ${expected.currency}`);
|
|
|
|
const startTime = Date.now();
|
|
const images = convertPdfToImages(testCase.pdfPath);
|
|
console.log(` Pages: ${images.length}`);
|
|
|
|
const extracted = await extractInvoiceFromImages(images);
|
|
console.log(` Extracted: ${extracted.invoice_number} | ${extracted.invoice_date} | ${extracted.total_amount} ${extracted.currency}`);
|
|
|
|
const elapsedMs = Date.now() - startTime;
|
|
processingTimes.push(elapsedMs);
|
|
|
|
const result = compareInvoice(extracted, expected);
|
|
|
|
if (result.match) {
|
|
passedCount++;
|
|
console.log(` Result: MATCH (${(elapsedMs / 1000).toFixed(1)}s)`);
|
|
} else {
|
|
failedCount++;
|
|
console.log(` Result: MISMATCH (${(elapsedMs / 1000).toFixed(1)}s)`);
|
|
result.errors.forEach((e) => console.log(` - ${e}`));
|
|
}
|
|
|
|
expect(result.match).toBeTrue();
|
|
});
|
|
}
|
|
|
|
tap.test('summary', async () => {
|
|
const totalInvoices = testCases.length;
|
|
const accuracy = totalInvoices > 0 ? (passedCount / totalInvoices) * 100 : 0;
|
|
const totalTimeMs = processingTimes.reduce((a, b) => a + b, 0);
|
|
const avgTimeSec = processingTimes.length > 0 ? totalTimeMs / processingTimes.length / 1000 : 0;
|
|
|
|
console.log(`\n========================================`);
|
|
console.log(` Invoice Extraction Summary (${MODEL})`);
|
|
console.log(`========================================`);
|
|
console.log(` Method: Consensus (fast + thinking)`);
|
|
console.log(` Passed: ${passedCount}/${totalInvoices}`);
|
|
console.log(` Failed: ${failedCount}/${totalInvoices}`);
|
|
console.log(` Accuracy: ${accuracy.toFixed(1)}%`);
|
|
console.log(`----------------------------------------`);
|
|
console.log(` Total time: ${(totalTimeMs / 1000).toFixed(1)}s`);
|
|
console.log(` Avg per inv: ${avgTimeSec.toFixed(1)}s`);
|
|
console.log(`========================================\n`);
|
|
});
|
|
|
|
export default tap.start();
|