host.today/ht-docker-ai
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(tests): stabilize OCR extraction tests and manage GPU containers

Browse Source
This commit is contained in:
Jürgen Kunz
2026-01-18 23:00:24 +00:00
parent 09ea7440e8
commit ae28a64902
4 changed files with 451 additions and 425 deletions
Download Patch File Download Diff File Expand all files Collapse all files

10
changelog.md
View File

@@ -1,5 +1,15 @@
# Changelog # Changelog
## 2026-01-18 - 1.13.2 - fix(tests)
stabilize OCR extraction tests and manage GPU containers
- Add stopAllGpuContainers() and call it before starting GPU images to free GPU memory.
- Remove PaddleOCR-VL image configs and associated ensure helpers from docker test helper to simplify images list.
- Split invoice/bankstatement tests into two sequential stages: Stage 1 runs Nanonets OCR to produce markdown files, Stage 2 stops Nanonets and runs model extraction from saved markdown (avoids GPU contention).
- Introduce temporary markdown directory handling and cleanup; add stopNanonets() and container running checks in tests.
- Switch bank statement extraction model from qwen3:8b to gpt-oss:20b; add request timeout and improved logging/console output across tests.
- Refactor extractWithConsensus and extraction functions to accept document identifiers, improve error messages and JSON extraction robustness.
## 2026-01-18 - 1.13.1 - fix(image_support_files) ## 2026-01-18 - 1.13.1 - fix(image_support_files)
remove PaddleOCR-VL server scripts from image_support_files remove PaddleOCR-VL server scripts from image_support_files

109
test/helpers/docker.ts
View File

@@ -2,10 +2,6 @@ import { execSync } from 'child_process';
// Project container names (only manage these) // Project container names (only manage these)
const PROJECT_CONTAINERS = [ const PROJECT_CONTAINERS = [
'paddleocr-vl-test',
'paddleocr-vl-gpu-test',
'paddleocr-vl-cpu-test',
'paddleocr-vl-full-test',
'minicpm-test', 'minicpm-test',
'nanonets-test', 'nanonets-test',
]; ];
@@ -24,30 +20,6 @@ export interface IImageConfig {
} }
export const IMAGES = { export const IMAGES = {
paddleocrVlGpu: {
name: 'paddleocr-vl-gpu',
dockerfile: 'Dockerfile_paddleocr_vl_gpu',
buildContext: '.',
containerName: 'paddleocr-vl-test',
ports: ['8000:8000'],
volumes: ['ht-huggingface-cache:/root/.cache/huggingface'],
gpus: true,
healthEndpoint: 'http://localhost:8000/health',
healthTimeout: 300000, // 5 minutes for model loading
} as IImageConfig,
paddleocrVlCpu: {
name: 'paddleocr-vl-cpu',
dockerfile: 'Dockerfile_paddleocr_vl_cpu',
buildContext: '.',
containerName: 'paddleocr-vl-test',
ports: ['8000:8000'],
volumes: ['ht-huggingface-cache:/root/.cache/huggingface'],
gpus: false,
healthEndpoint: 'http://localhost:8000/health',
healthTimeout: 300000,
} as IImageConfig,
minicpm: { minicpm: {
name: 'minicpm45v', name: 'minicpm45v',
dockerfile: 'Dockerfile_minicpm45v_gpu', dockerfile: 'Dockerfile_minicpm45v_gpu',
@@ -60,22 +32,6 @@ export const IMAGES = {
healthTimeout: 120000, healthTimeout: 120000,
} as IImageConfig, } as IImageConfig,
// Full PaddleOCR-VL pipeline with PP-DocLayoutV2 + structured JSON output
paddleocrVlFull: {
name: 'paddleocr-vl-full',
dockerfile: 'Dockerfile_paddleocr_vl_full',
buildContext: '.',
containerName: 'paddleocr-vl-full-test',
ports: ['8000:8000'],
volumes: [
'ht-huggingface-cache:/root/.cache/huggingface',
'ht-paddleocr-cache:/root/.paddleocr',
],
gpus: true,
healthEndpoint: 'http://localhost:8000/health',
healthTimeout: 600000, // 10 minutes for model loading (vLLM + PP-DocLayoutV2)
} as IImageConfig,
// Nanonets-OCR-s - Document OCR optimized VLM (Qwen2.5-VL-3B fine-tuned) // Nanonets-OCR-s - Document OCR optimized VLM (Qwen2.5-VL-3B fine-tuned)
nanonetsOcr: { nanonetsOcr: {
name: 'nanonets-ocr', name: 'nanonets-ocr',
@@ -140,7 +96,7 @@ export function removeContainer(containerName: string): void {
} }
/** /**
* Stop all project containers that conflict with the required one * Stop all project containers that conflict with the required one (port-based)
*/ */
export function stopConflictingContainers(requiredContainer: string, requiredPort: string): void { export function stopConflictingContainers(requiredContainer: string, requiredPort: string): void {
// Stop project containers using the same port // Stop project containers using the same port
@@ -158,6 +114,24 @@ export function stopConflictingContainers(requiredContainer: string, requiredPor
} }
} }
/**
* Stop all GPU-consuming project containers (for GPU memory management)
* This ensures GPU memory is freed before starting a new GPU service
*/
export function stopAllGpuContainers(exceptContainer?: string): void {
for (const container of PROJECT_CONTAINERS) {
if (container === exceptContainer) continue;
if (isContainerRunning(container)) {
console.log(`[Docker] Stopping GPU container: ${container}`);
exec(`docker stop ${container}`, true);
// Give the GPU a moment to free memory
}
}
// Brief pause to allow GPU memory to be released
execSync('sleep 2');
}
/** /**
* Build a Docker image * Build a Docker image
*/ */
@@ -234,6 +208,11 @@ export async function ensureService(config: IImageConfig): Promise<boolean> {
buildImage(config); buildImage(config);
} }
// For GPU services, stop ALL other GPU containers to free GPU memory
if (config.gpus) {
stopAllGpuContainers(config.containerName);
}
// Stop conflicting containers on the same port // Stop conflicting containers on the same port
const mainPort = config.ports[0]; const mainPort = config.ports[0];
stopConflictingContainers(config.containerName, mainPort); stopConflictingContainers(config.containerName, mainPort);
@@ -254,21 +233,7 @@ export async function ensureService(config: IImageConfig): Promise<boolean> {
} }
/** /**
* Ensure PaddleOCR-VL GPU service is running * Ensure MiniCPM service is running (Ollama with GPU)
*/
export async function ensurePaddleOcrVlGpu(): Promise<boolean> {
return ensureService(IMAGES.paddleocrVlGpu);
}
/**
* Ensure PaddleOCR-VL CPU service is running
*/
export async function ensurePaddleOcrVlCpu(): Promise<boolean> {
return ensureService(IMAGES.paddleocrVlCpu);
}
/**
* Ensure MiniCPM service is running
*/ */
export async function ensureMiniCpm(): Promise<boolean> { export async function ensureMiniCpm(): Promise<boolean> {
return ensureService(IMAGES.minicpm); return ensureService(IMAGES.minicpm);
@@ -286,30 +251,6 @@ export function isGpuAvailable(): boolean {
} }
} }
/**
* Ensure PaddleOCR-VL service (auto-detect GPU/CPU)
*/
export async function ensurePaddleOcrVl(): Promise<boolean> {
if (isGpuAvailable()) {
console.log('[Docker] GPU detected, using GPU image');
return ensurePaddleOcrVlGpu();
} else {
console.log('[Docker] No GPU detected, using CPU image');
return ensurePaddleOcrVlCpu();
}
}
/**
* Ensure PaddleOCR-VL Full Pipeline service (PP-DocLayoutV2 + structured output)
* This is the recommended service for production use - outputs structured JSON/Markdown
*/
export async function ensurePaddleOcrVlFull(): Promise<boolean> {
if (!isGpuAvailable()) {
console.log('[Docker] WARNING: Full pipeline requires GPU, but none detected');
}
return ensureService(IMAGES.paddleocrVlFull);
}
/** /**
* Ensure an Ollama model is pulled and available * Ensure an Ollama model is pulled and available
* Uses the MiniCPM container (which runs Ollama) to pull the model * Uses the MiniCPM container (which runs Ollama) to pull the model

489
test/test.bankstatements.nanonets.ts
View File

@@ -1,25 +1,26 @@
/** /**
* Bank statement extraction using Nanonets-OCR-s + Qwen3 (two-stage pipeline) * Bank statement extraction using Nanonets-OCR-s + GPT-OSS 20B (sequential two-stage pipeline)
* *
* Stage 1: Nanonets-OCR-s converts document pages to markdown (its strength) * Stage 1: Nanonets-OCR-s converts ALL document pages to markdown (stop after completion)
* Stage 2: Qwen3 extracts structured JSON from the combined markdown * Stage 2: GPT-OSS 20B extracts structured JSON from saved markdown (after Nanonets stops)
* *
* This leverages each model's strengths: * This approach avoids GPU contention by running services sequentially.
* - Nanonets: Document OCR with semantic tags
* - Qwen3: Text understanding and JSON extraction
*/ */
import { tap, expect } from '@git.zone/tstest/tapbundle'; import { tap, expect } from '@git.zone/tstest/tapbundle';
import * as fs from 'fs'; import * as fs from 'fs';
import * as path from 'path'; import * as path from 'path';
import { execSync } from 'child_process'; import { execSync } from 'child_process';
import * as os from 'os'; import * as os from 'os';
import { ensureNanonetsOcr, ensureMiniCpm } from './helpers/docker.js'; import { ensureNanonetsOcr, ensureMiniCpm, removeContainer, isContainerRunning } from './helpers/docker.js';
const NANONETS_URL = 'http://localhost:8000/v1'; const NANONETS_URL = 'http://localhost:8000/v1';
const NANONETS_MODEL = 'nanonets/Nanonets-OCR-s'; const NANONETS_MODEL = 'nanonets/Nanonets-OCR-s';
const OLLAMA_URL = 'http://localhost:11434'; const OLLAMA_URL = 'http://localhost:11434';
const QWEN_MODEL = 'qwen3:8b'; const EXTRACTION_MODEL = 'gpt-oss:20b';
// Temp directory for storing markdown between stages
const TEMP_MD_DIR = path.join(os.tmpdir(), 'nanonets-markdown');
interface ITransaction { interface ITransaction {
date: string; date: string;
@@ -27,6 +28,14 @@ interface ITransaction {
amount: number; amount: number;
} }
interface ITestCase {
name: string;
pdfPath: string;
jsonPath: string;
markdownPath?: string;
images?: string[];
}
// Nanonets-specific prompt for document OCR to markdown // Nanonets-specific prompt for document OCR to markdown
const NANONETS_OCR_PROMPT = `Extract the text from the above document as if you were reading it naturally. const NANONETS_OCR_PROMPT = `Extract the text from the above document as if you were reading it naturally.
Return the tables in html format. Return the tables in html format.
@@ -35,24 +44,10 @@ If there is an image in the document and image caption is not present, add a sma
Watermarks should be wrapped in brackets. Ex: <watermark>OFFICIAL COPY</watermark>. Watermarks should be wrapped in brackets. Ex: <watermark>OFFICIAL COPY</watermark>.
Page numbers should be wrapped in brackets. Ex: <page_number>14</page_number>.`; Page numbers should be wrapped in brackets. Ex: <page_number>14</page_number>.`;
// JSON extraction prompt for Qwen3 // JSON extraction prompt for GPT-OSS 20B
const JSON_EXTRACTION_PROMPT = `You are a financial data extractor. Below is a bank statement converted to text/markdown. Extract ALL transactions from it as a JSON array. const JSON_EXTRACTION_PROMPT = `Extract ALL transactions from this bank statement as JSON array. Each transaction: {"date": "YYYY-MM-DD", "counterparty": "NAME", "amount": -25.99}. Amount negative for debits, positive for credits. Only include actual transactions, not balances. Return ONLY JSON array, no explanation.
IMPORTANT RULES: STATEMENT:
1. Each transaction has: date, description/counterparty, and an amount
2. Amount is NEGATIVE for money going OUT (debits, payments, withdrawals)
3. Amount is POSITIVE for money coming IN (credits, deposits, refunds)
4. Date format: YYYY-MM-DD
5. Do NOT include: opening balance, closing balance, subtotals, headers, or summary rows
6. Only include actual transactions with a specific date and amount
Return ONLY this JSON format, no explanation:
[
{"date": "2021-06-01", "counterparty": "COMPANY NAME", "amount": -25.99},
{"date": "2021-06-02", "counterparty": "DEPOSIT FROM", "amount": 100.00}
]
BANK STATEMENT TEXT:
`; `;
/** /**
@@ -63,7 +58,6 @@ function convertPdfToImages(pdfPath: string): string[] {
const outputPattern = path.join(tempDir, 'page-%d.png'); const outputPattern = path.join(tempDir, 'page-%d.png');
try { try {
// Use 150 DPI to keep images within model's context length
execSync( execSync(
`convert -density 150 -quality 90 "${pdfPath}" -background white -alpha remove "${outputPattern}"`, `convert -density 150 -quality 90 "${pdfPath}" -background white -alpha remove "${outputPattern}"`,
{ stdio: 'pipe' } { stdio: 'pipe' }
@@ -85,10 +79,9 @@ function convertPdfToImages(pdfPath: string): string[] {
} }
/** /**
* Stage 1: Convert a single page to markdown using Nanonets-OCR-s * Convert a single page to markdown using Nanonets-OCR-s
*/ */
async function convertPageToMarkdown(image: string, pageNum: number): Promise<string> { async function convertPageToMarkdown(image: string, pageNum: number): Promise<string> {
console.log(` [Nanonets] Converting page ${pageNum} to markdown...`);
const startTime = Date.now(); const startTime = Date.now();
const response = await fetch(`${NANONETS_URL}/chat/completions`, { const response = await fetch(`${NANONETS_URL}/chat/completions`, {
@@ -115,21 +108,20 @@ async function convertPageToMarkdown(image: string, pageNum: number): Promise<st
if (!response.ok) { if (!response.ok) {
const errorText = await response.text(); const errorText = await response.text();
console.log(` [Nanonets] ERROR page ${pageNum}: ${response.status} - ${errorText}`); throw new Error(`Nanonets API error: ${response.status} - ${errorText}`);
throw new Error(`Nanonets API error: ${response.status}`);
} }
const data = await response.json(); const data = await response.json();
const content = (data.choices?.[0]?.message?.content || '').trim(); const content = (data.choices?.[0]?.message?.content || '').trim();
console.log(` [Nanonets] Page ${pageNum} converted (${elapsed}s, ${content.length} chars)`); console.log(` Page ${pageNum}: ${content.length} chars (${elapsed}s)`);
return content; return content;
} }
/** /**
* Stage 1: Convert all pages to markdown using Nanonets-OCR-s * Convert all pages of a document to markdown
*/ */
async function convertDocumentToMarkdown(images: string[]): Promise<string> { async function convertDocumentToMarkdown(images: string[], docName: string): Promise<string> {
console.log(` [Stage 1] Converting ${images.length} page(s) to markdown with Nanonets-OCR-s...`); console.log(` [${docName}] Converting ${images.length} page(s)...`);
const markdownPages: string[] = []; const markdownPages: string[] = [];
@@ -139,21 +131,55 @@ async function convertDocumentToMarkdown(images: string[]): Promise<string> {
} }
const fullMarkdown = markdownPages.join('\n\n'); const fullMarkdown = markdownPages.join('\n\n');
console.log(` [Stage 1] Complete: ${fullMarkdown.length} chars total`); console.log(` [${docName}] Complete: ${fullMarkdown.length} chars total`);
return fullMarkdown; return fullMarkdown;
} }
/** /**
* Ensure Qwen3 model is available * Stop Nanonets container
*/ */
async function ensureQwen3(): Promise<boolean> { function stopNanonets(): void {
console.log(' [Docker] Stopping Nanonets container...');
try {
execSync('docker stop nanonets-test 2>/dev/null || true', { stdio: 'pipe' });
// Wait for GPU memory to be released
execSync('sleep 5', { stdio: 'pipe' });
console.log(' [Docker] Nanonets stopped');
} catch {
console.log(' [Docker] Nanonets was not running');
}
}
/**
* Ensure GPT-OSS 20B model is available and warmed up
*/
async function ensureExtractionModel(): Promise<boolean> {
try { try {
const response = await fetch(`${OLLAMA_URL}/api/tags`); const response = await fetch(`${OLLAMA_URL}/api/tags`);
if (response.ok) { if (response.ok) {
const data = await response.json(); const data = await response.json();
const models = data.models || []; const models = data.models || [];
if (models.some((m: { name: string }) => m.name === QWEN_MODEL)) { if (models.some((m: { name: string }) => m.name === EXTRACTION_MODEL)) {
console.log(` [Ollama] Model available: ${QWEN_MODEL}`); console.log(` [Ollama] Model available: ${EXTRACTION_MODEL}`);
// Warmup: send a simple request to ensure model is loaded
console.log(` [Ollama] Warming up model...`);
const warmupResponse = await fetch(`${OLLAMA_URL}/api/chat`, {
method: 'POST',
headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({
model: EXTRACTION_MODEL,
messages: [{ role: 'user', content: 'Return: [{"test": 1}]' }],
stream: false,
}),
signal: AbortSignal.timeout(120000),
});
if (warmupResponse.ok) {
const warmupData = await warmupResponse.json();
console.log(` [Ollama] Warmup complete (${warmupData.message?.content?.length || 0} chars)`);
}
return true; return true;
} }
} }
@@ -161,77 +187,92 @@ async function ensureQwen3(): Promise<boolean> {
return false; return false;
} }
console.log(` [Ollama] Pulling ${QWEN_MODEL}...`); console.log(` [Ollama] Pulling ${EXTRACTION_MODEL}...`);
const pullResponse = await fetch(`${OLLAMA_URL}/api/pull`, { const pullResponse = await fetch(`${OLLAMA_URL}/api/pull`, {
method: 'POST', method: 'POST',
headers: { 'Content-Type': 'application/json' }, headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ name: QWEN_MODEL, stream: false }), body: JSON.stringify({ name: EXTRACTION_MODEL, stream: false }),
}); });
return pullResponse.ok; return pullResponse.ok;
} }
/** /**
* Stage 2: Extract transactions from markdown using Qwen3 * Extract transactions from markdown using GPT-OSS 20B (streaming)
*/ */
async function extractTransactionsFromMarkdown(markdown: string, queryId: string): Promise<ITransaction[]> { async function extractTransactionsFromMarkdown(markdown: string, queryId: string): Promise<ITransaction[]> {
console.log(` [${queryId}] Sending markdown to ${QWEN_MODEL}...`); console.log(` [${queryId}] Sending to ${EXTRACTION_MODEL}...`);
console.log(` [${queryId}] Markdown length: ${markdown.length}`);
const startTime = Date.now(); const startTime = Date.now();
const fullPrompt = JSON_EXTRACTION_PROMPT + markdown;
console.log(` [${queryId}] Prompt preview: ${fullPrompt.substring(0, 200)}...`);
const response = await fetch(`${OLLAMA_URL}/api/chat`, { const response = await fetch(`${OLLAMA_URL}/api/chat`, {
method: 'POST', method: 'POST',
headers: { 'Content-Type': 'application/json' }, headers: { 'Content-Type': 'application/json' },
body: JSON.stringify({ body: JSON.stringify({
model: QWEN_MODEL, model: EXTRACTION_MODEL,
messages: [{ messages: [{
role: 'user', role: 'user',
content: JSON_EXTRACTION_PROMPT + markdown, content: fullPrompt,
}], }],
stream: false, stream: true,
options: {
num_predict: 8000,
temperature: 0.1,
},
}), }),
signal: AbortSignal.timeout(600000), // 10 minute timeout
}); });
const elapsed = ((Date.now() - startTime) / 1000).toFixed(1);
if (!response.ok) { if (!response.ok) {
const elapsed = ((Date.now() - startTime) / 1000).toFixed(1);
console.log(` [${queryId}] ERROR: ${response.status} (${elapsed}s)`); console.log(` [${queryId}] ERROR: ${response.status} (${elapsed}s)`);
throw new Error(`Ollama API error: ${response.status}`); throw new Error(`Ollama API error: ${response.status}`);
} }
const data = await response.json(); // Stream the response and log to console
const content = (data.message?.content || '').trim(); let content = '';
console.log(` [${queryId}] Response received (${elapsed}s, ${content.length} chars)`); const reader = response.body!.getReader();
const decoder = new TextDecoder();
process.stdout.write(` [${queryId}] `);
while (true) {
const { done, value } = await reader.read();
if (done) break;
const chunk = decoder.decode(value, { stream: true });
// Each line is a JSON object
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
}
}
}
const elapsed = ((Date.now() - startTime) / 1000).toFixed(1);
console.log(`\n [${queryId}] Done: ${content.length} chars (${elapsed}s)`);
return parseJsonResponse(content, queryId); return parseJsonResponse(content, queryId);
} }
/** /**
* Sanitize JSON string - fix common issues from LLM output * Sanitize JSON string
*/ */
function sanitizeJson(jsonStr: string): string { function sanitizeJson(jsonStr: string): string {
let s = jsonStr; let s = jsonStr;
// Fix +number (e.g., +93.80 -> 93.80) - JSON doesn't allow + prefix
s = s.replace(/"amount"\s*:\s*\+/g, '"amount": '); s = s.replace(/"amount"\s*:\s*\+/g, '"amount": ');
s = s.replace(/:\s*\+(\d)/g, ': $1'); s = s.replace(/:\s*\+(\d)/g, ': $1');
// Fix European number format with thousands separator
s = s.replace(/"amount"\s*:\s*(-?)(\d{1,3})\.(\d{3})\.(\d{2})\b/g, '"amount": $1$2$3.$4'); s = s.replace(/"amount"\s*:\s*(-?)(\d{1,3})\.(\d{3})\.(\d{2})\b/g, '"amount": $1$2$3.$4');
// Fix trailing commas before ] or }
s = s.replace(/,\s*([}\]])/g, '$1'); s = s.replace(/,\s*([}\]])/g, '$1');
// Fix unescaped newlines/tabs inside strings
s = s.replace(/"([^"\\]*)\n([^"]*)"/g, '"$1 $2"'); s = s.replace(/"([^"\\]*)\n([^"]*)"/g, '"$1 $2"');
s = s.replace(/"([^"\\]*)\t([^"]*)"/g, '"$1 $2"'); s = s.replace(/"([^"\\]*)\t([^"]*)"/g, '"$1 $2"');
// Remove control characters
s = s.replace(/[\x00-\x08\x0B\x0C\x0E-\x1F]/g, ' '); s = s.replace(/[\x00-\x08\x0B\x0C\x0E-\x1F]/g, ' ');
return s; return s;
} }
@@ -243,7 +284,6 @@ function parseAmount(value: unknown): number {
if (typeof value !== 'string') return 0; if (typeof value !== 'string') return 0;
let s = value.replace(/[€$£\s]/g, '').replace('−', '-').replace('–', '-'); let s = value.replace(/[€$£\s]/g, '').replace('−', '-').replace('–', '-');
// European format: comma is decimal
if (s.includes(',') && s.indexOf(',') > s.lastIndexOf('.')) { if (s.includes(',') && s.indexOf(',') > s.lastIndexOf('.')) {
s = s.replace(/\./g, '').replace(',', '.'); s = s.replace(/\./g, '').replace(',', '.');
} else { } else {
@@ -256,16 +296,14 @@ function parseAmount(value: unknown): number {
* Parse JSON response into transactions * Parse JSON response into transactions
*/ */
function parseJsonResponse(response: string, queryId: string): ITransaction[] { function parseJsonResponse(response: string, queryId: string): ITransaction[] {
console.log(` [${queryId}] Parsing response...`); // Remove thinking tags if present
// Remove thinking tags if present (Qwen3 may include <think>...</think>)
let cleanResponse = response.replace(/<think>[\s\S]*?<\/think>/g, '').trim(); let cleanResponse = response.replace(/<think>[\s\S]*?<\/think>/g, '').trim();
// Try to find JSON in markdown code block // Debug: show what we're working with
console.log(` [${queryId}] Response preview: ${cleanResponse.substring(0, 300)}...`);
const codeBlockMatch = cleanResponse.match(/```(?:json)?\s*([\s\S]*?)```/); const codeBlockMatch = cleanResponse.match(/```(?:json)?\s*([\s\S]*?)```/);
let jsonStr = codeBlockMatch ? codeBlockMatch[1].trim() : cleanResponse; let jsonStr = codeBlockMatch ? codeBlockMatch[1].trim() : cleanResponse;
// Sanitize JSON
jsonStr = sanitizeJson(jsonStr); jsonStr = sanitizeJson(jsonStr);
try { try {
@@ -280,11 +318,10 @@ function parseJsonResponse(response: string, queryId: string): ITransaction[] {
return txs; return txs;
} }
} catch (e) { } catch (e) {
console.log(` [${queryId}] Direct parse failed: ${(e as Error).message}`); // Try to find a JSON array in the text
// Try to find JSON array pattern
const arrayMatch = jsonStr.match(/\[[\s\S]*\]/); const arrayMatch = jsonStr.match(/\[[\s\S]*\]/);
if (arrayMatch) { if (arrayMatch) {
console.log(` [${queryId}] Array match found: ${arrayMatch[0].length} chars`);
try { try {
const parsed = JSON.parse(sanitizeJson(arrayMatch[0])); const parsed = JSON.parse(sanitizeJson(arrayMatch[0]));
if (Array.isArray(parsed)) { if (Array.isArray(parsed)) {
@@ -296,93 +333,36 @@ function parseJsonResponse(response: string, queryId: string): ITransaction[] {
console.log(` [${queryId}] Parsed ${txs.length} transactions (array match)`); console.log(` [${queryId}] Parsed ${txs.length} transactions (array match)`);
return txs; return txs;
} }
} catch (e2) { } catch (innerErr) {
console.log(` [${queryId}] Array parse failed: ${(e2 as Error).message}`); console.log(` [${queryId}] Array parse error: ${(innerErr as Error).message}`);
} }
} else {
console.log(` [${queryId}] No JSON array found in response`);
} }
} }
console.log(` [${queryId}] PARSE FAILED - returning empty array`); console.log(` [${queryId}] PARSE FAILED`);
return []; return [];
} }
/** /**
* Compare two transaction arrays for consensus * Extract transactions (single pass)
*/ */
function transactionArraysMatch(a: ITransaction[], b: ITransaction[]): boolean { async function extractTransactions(markdown: string, docName: string): Promise<ITransaction[]> {
if (a.length !== b.length) return false; console.log(` [${docName}] Extracting...`);
const txs = await extractTransactionsFromMarkdown(markdown, docName);
for (let i = 0; i < a.length; i++) { console.log(` [${docName}] Extracted ${txs.length} transactions`);
const dateMatch = a[i].date === b[i].date; return txs;
const amountMatch = Math.abs(a[i].amount - b[i].amount) < 0.01;
if (!dateMatch || !amountMatch) return false;
}
return true;
} }
/** /**
* Stage 2: Extract transactions using Qwen3 with consensus * Compare transactions
*/
async function extractWithConsensus(markdown: string): Promise<ITransaction[]> {
const MAX_ATTEMPTS = 3;
console.log(` [Stage 2] Extracting transactions with ${QWEN_MODEL} (consensus)...`);
for (let attempt = 1; attempt <= MAX_ATTEMPTS; attempt++) {
console.log(`\n [Stage 2] --- Attempt ${attempt}/${MAX_ATTEMPTS} ---`);
// Extract twice in parallel
const [txs1, txs2] = await Promise.all([
extractTransactionsFromMarkdown(markdown, `A${attempt}Q1`),
extractTransactionsFromMarkdown(markdown, `A${attempt}Q2`),
]);
console.log(` [Stage 2] Results: Q1=${txs1.length} txs, Q2=${txs2.length} txs`);
if (txs1.length > 0 && transactionArraysMatch(txs1, txs2)) {
console.log(` [Stage 2] CONSENSUS REACHED: ${txs1.length} transactions`);
return txs1;
}
console.log(` [Stage 2] NO CONSENSUS`);
}
// Fallback: use last response
console.log(`\n [Stage 2] === FALLBACK ===`);
const fallback = await extractTransactionsFromMarkdown(markdown, 'FALLBACK');
console.log(` [Stage 2] ~ FALLBACK RESULT: ${fallback.length} transactions`);
return fallback;
}
/**
* Full pipeline: PDF -> Images -> Markdown -> JSON
*/
async function extractTransactions(images: string[]): Promise<ITransaction[]> {
// Stage 1: Convert to markdown
const markdown = await convertDocumentToMarkdown(images);
// Stage 2: Extract transactions with consensus
const transactions = await extractWithConsensus(markdown);
// Log all transactions
console.log(`\n [Result] Extracted ${transactions.length} transactions:`);
for (let i = 0; i < transactions.length; i++) {
const tx = transactions[i];
console.log(` ${(i + 1).toString().padStart(2)}. ${tx.date} | ${tx.counterparty.substring(0, 30).padEnd(30)} | ${tx.amount >= 0 ? '+' : ''}${tx.amount.toFixed(2)}`);
}
return transactions;
}
/**
* Compare extracted transactions against expected
*/ */
function compareTransactions( function compareTransactions(
extracted: ITransaction[], extracted: ITransaction[],
expected: ITransaction[] expected: ITransaction[]
): { matches: number; total: number; errors: string[]; variations: string[] } { ): { matches: number; total: number; errors: string[] } {
const errors: string[] = []; const errors: string[] = [];
const variations: string[] = [];
let matches = 0; let matches = 0;
for (let i = 0; i < expected.length; i++) { for (let i = 0; i < expected.length; i++) {
@@ -390,7 +370,7 @@ function compareTransactions(
const ext = extracted[i]; const ext = extracted[i];
if (!ext) { if (!ext) {
errors.push(`Missing transaction ${i}: ${exp.date} ${exp.counterparty}`); errors.push(`Missing tx ${i}: ${exp.date} ${exp.counterparty}`);
continue; continue;
} }
@@ -399,11 +379,8 @@ function compareTransactions(
if (dateMatch && amountMatch) { if (dateMatch && amountMatch) {
matches++; matches++;
if (ext.counterparty !== exp.counterparty) {
variations.push(`[${i}] "${exp.counterparty}" -> "${ext.counterparty}"`);
}
} else { } else {
errors.push(`Mismatch at ${i}: expected ${exp.date}/${exp.amount}, got ${ext.date}/${ext.amount}`); errors.push(`Mismatch ${i}: exp ${exp.date}/${exp.amount}, got ${ext.date}/${ext.amount}`);
} }
} }
@@ -411,23 +388,20 @@ function compareTransactions(
errors.push(`Extra transactions: ${extracted.length - expected.length}`); errors.push(`Extra transactions: ${extracted.length - expected.length}`);
} }
return { matches, total: expected.length, errors, variations }; return { matches, total: expected.length, errors };
} }
/** /**
* Find all test cases (PDF + JSON pairs) in .nogit/ * Find all test cases
*/ */
function findTestCases(): Array<{ name: string; pdfPath: string; jsonPath: string }> { function findTestCases(): ITestCase[] {
const testDir = path.join(process.cwd(), '.nogit'); const testDir = path.join(process.cwd(), '.nogit');
if (!fs.existsSync(testDir)) { if (!fs.existsSync(testDir)) return [];
return [];
}
const files = fs.readdirSync(testDir); const files = fs.readdirSync(testDir);
const pdfFiles = files.filter((f: string) => f.endsWith('.pdf')); const testCases: ITestCase[] = [];
const testCases: Array<{ name: string; pdfPath: string; jsonPath: string }> = [];
for (const pdf of pdfFiles) { for (const pdf of files.filter((f: string) => f.endsWith('.pdf'))) {
const baseName = pdf.replace('.pdf', ''); const baseName = pdf.replace('.pdf', '');
const jsonFile = `${baseName}.json`; const jsonFile = `${baseName}.json`;
if (files.includes(jsonFile)) { if (files.includes(jsonFile)) {
@@ -442,72 +416,142 @@ function findTestCases(): Array<{ name: string; pdfPath: string; jsonPath: strin
return testCases.sort((a, b) => a.name.localeCompare(b.name)); return testCases.sort((a, b) => a.name.localeCompare(b.name));
} }
// Tests // ============ TESTS ============
tap.test('setup: ensure containers are running', async () => { const testCases = findTestCases();
console.log('\n[Setup] Checking Docker containers...\n'); console.log(`\nFound ${testCases.length} bank statement test cases\n`);
// Nanonets for OCR // Ensure temp directory exists
const nanonetsOk = await ensureNanonetsOcr(); if (!fs.existsSync(TEMP_MD_DIR)) {
expect(nanonetsOk).toBeTrue(); fs.mkdirSync(TEMP_MD_DIR, { recursive: true });
}
// -------- STAGE 1: OCR with Nanonets --------
// Check if all markdown files already exist
function allMarkdownFilesExist(): boolean {
for (const tc of testCases) {
const mdPath = path.join(TEMP_MD_DIR, `${tc.name}.md`);
if (!fs.existsSync(mdPath)) {
return false;
}
}
return true;
}
// Track whether we need to run Stage 1
let stage1Needed = !allMarkdownFilesExist();
tap.test('Stage 1: Setup Nanonets', async () => {
console.log('\n========== STAGE 1: Nanonets OCR ==========\n');
if (!stage1Needed) {
console.log(' [SKIP] All markdown files already exist, skipping Nanonets setup');
return;
}
const ok = await ensureNanonetsOcr();
expect(ok).toBeTrue();
});
tap.test('Stage 1: Convert all documents to markdown', async () => {
if (!stage1Needed) {
console.log(' [SKIP] Using existing markdown files from previous run\n');
// Load existing markdown paths
for (const tc of testCases) {
tc.markdownPath = path.join(TEMP_MD_DIR, `${tc.name}.md`);
console.log(` Loaded: ${tc.markdownPath}`);
}
return;
}
console.log('\n Converting all PDFs to markdown with Nanonets-OCR-s...\n');
for (const tc of testCases) {
console.log(`\n === ${tc.name} ===`);
// Convert PDF to images
const images = convertPdfToImages(tc.pdfPath);
console.log(` Pages: ${images.length}`);
// Convert to markdown
const markdown = await convertDocumentToMarkdown(images, tc.name);
// Save markdown to temp file
const mdPath = path.join(TEMP_MD_DIR, `${tc.name}.md`);
fs.writeFileSync(mdPath, markdown);
tc.markdownPath = mdPath;
console.log(` Saved: ${mdPath}`);
}
console.log('\n Stage 1 complete: All documents converted to markdown\n');
});
tap.test('Stage 1: Stop Nanonets', async () => {
if (!stage1Needed) {
console.log(' [SKIP] Nanonets was not started');
return;
}
stopNanonets();
// Verify it's stopped
await new Promise(resolve => setTimeout(resolve, 3000));
expect(isContainerRunning('nanonets-test')).toBeFalse();
});
// -------- STAGE 2: Extraction with GPT-OSS 20B --------
tap.test('Stage 2: Setup Ollama + GPT-OSS 20B', async () => {
console.log('\n========== STAGE 2: GPT-OSS 20B Extraction ==========\n');
// Ollama for Qwen3
const ollamaOk = await ensureMiniCpm(); const ollamaOk = await ensureMiniCpm();
expect(ollamaOk).toBeTrue(); expect(ollamaOk).toBeTrue();
// Qwen3 model const extractionOk = await ensureExtractionModel();
const qwenOk = await ensureQwen3(); expect(extractionOk).toBeTrue();
expect(qwenOk).toBeTrue();
console.log('\n[Setup] All containers ready!\n');
}); });
tap.test('should have models available', async () => {
// Check Nanonets
const nanonetsResp = await fetch(`${NANONETS_URL}/models`);
expect(nanonetsResp.ok).toBeTrue();
// Check Qwen3
const ollamaResp = await fetch(`${OLLAMA_URL}/api/tags`);
expect(ollamaResp.ok).toBeTrue();
const data = await ollamaResp.json();
const modelNames = data.models.map((m: { name: string }) => m.name);
expect(modelNames.some((name: string) => name.includes('qwen3'))).toBeTrue();
});
const testCases = findTestCases();
console.log(`\nFound ${testCases.length} bank statement test cases (Nanonets + Qwen3)\n`);
let passedCount = 0; let passedCount = 0;
let failedCount = 0; let failedCount = 0;
for (const testCase of testCases) { for (const tc of testCases) {
tap.test(`should extract: ${testCase.name}`, async () => { tap.test(`Stage 2: Extract ${tc.name}`, async () => {
const expected: ITransaction[] = JSON.parse(fs.readFileSync(testCase.jsonPath, 'utf-8')); const expected: ITransaction[] = JSON.parse(fs.readFileSync(tc.jsonPath, 'utf-8'));
console.log(`\n=== ${testCase.name} ===`); console.log(`\n === ${tc.name} ===`);
console.log(`Expected: ${expected.length} transactions`); console.log(` Expected: ${expected.length} transactions`);
const images = convertPdfToImages(testCase.pdfPath); // Load saved markdown
console.log(` Pages: ${images.length}`); const mdPath = path.join(TEMP_MD_DIR, `${tc.name}.md`);
if (!fs.existsSync(mdPath)) {
throw new Error(`Markdown not found: ${mdPath}. Run Stage 1 first.`);
}
const markdown = fs.readFileSync(mdPath, 'utf-8');
console.log(` Markdown: ${markdown.length} chars`);
const extracted = await extractTransactions(images); // Extract transactions (single pass)
console.log(` Extracted: ${extracted.length} transactions`); const extracted = await extractTransactions(markdown, tc.name);
const result = compareTransactions(extracted, expected); // Log results
const perfectMatch = result.matches === result.total && extracted.length === expected.length; console.log(` Extracted: ${extracted.length} transactions`);
for (let i = 0; i < Math.min(extracted.length, 5); i++) {
if (perfectMatch) { const tx = extracted[i];
passedCount++; console.log(` ${i + 1}. ${tx.date} | ${tx.counterparty.substring(0, 25).padEnd(25)} | ${tx.amount >= 0 ? '+' : ''}${tx.amount.toFixed(2)}`);
console.log(` Result: PASS (${result.matches}/${result.total})`); }
} else { if (extracted.length > 5) {
failedCount++; console.log(` ... and ${extracted.length - 5} more`);
console.log(` Result: FAIL (${result.matches}/${result.total})`);
result.errors.slice(0, 10).forEach((e) => console.log(` - ${e}`));
} }
if (result.variations.length > 0) { // Compare
console.log(` Counterparty variations (${result.variations.length}):`); const result = compareTransactions(extracted, expected);
result.variations.slice(0, 5).forEach((v) => console.log(` ${v}`)); const pass = result.matches === result.total && extracted.length === expected.length;
if (pass) {
passedCount++;
console.log(` Result: PASS (${result.matches}/${result.total})`);
} else {
failedCount++;
console.log(` Result: FAIL (${result.matches}/${result.total})`);
result.errors.slice(0, 5).forEach(e => console.log(` - ${e}`));
} }
expect(result.matches).toEqual(result.total); expect(result.matches).toEqual(result.total);
@@ -515,16 +559,27 @@ for (const testCase of testCases) {
}); });
} }
tap.test('summary', async () => { tap.test('Summary', async () => {
const total = testCases.length;
console.log(`\n======================================================`); console.log(`\n======================================================`);
console.log(` Bank Statement Summary (Nanonets + Qwen3 Pipeline)`); console.log(` Bank Statement Summary (Nanonets + GPT-OSS 20B Sequential)`);
console.log(`======================================================`); console.log(`======================================================`);
console.log(` Stage 1: Nanonets-OCR-s (document -> markdown)`); console.log(` Stage 1: Nanonets-OCR-s (document -> markdown)`);
console.log(` Stage 2: Qwen3 8B (markdown -> JSON)`); console.log(` Stage 2: GPT-OSS 20B (markdown -> JSON)`);
console.log(` Passed: ${passedCount}/${total}`); console.log(` Passed: ${passedCount}/${testCases.length}`);
console.log(` Failed: ${failedCount}/${total}`); console.log(` Failed: ${failedCount}/${testCases.length}`);
console.log(`======================================================\n`); console.log(`======================================================\n`);
// Only cleanup temp files if ALL tests passed
if (failedCount === 0 && passedCount === testCases.length) {
try {
fs.rmSync(TEMP_MD_DIR, { recursive: true, force: true });
console.log(` Cleaned up temp directory: ${TEMP_MD_DIR}\n`);
} catch {
// Ignore
}
} else {
console.log(` Keeping temp directory for debugging: ${TEMP_MD_DIR}\n`);
}
}); });
export default tap.start(); export default tap.start();

268
test/test.invoices.nanonets.ts
View File

@@ -1,19 +1,17 @@
/** /**
* Invoice extraction using Nanonets-OCR-s + Qwen3 (two-stage pipeline) * Invoice extraction using Nanonets-OCR-s + Qwen3 (sequential two-stage pipeline)
* *
* Stage 1: Nanonets-OCR-s converts document pages to markdown (its strength) * Stage 1: Nanonets-OCR-s converts ALL document pages to markdown (stop after completion)
* Stage 2: Qwen3 extracts structured JSON from the combined markdown * Stage 2: Qwen3 extracts structured JSON from saved markdown (after Nanonets stops)
* *
* This leverages each model's strengths: * This approach avoids GPU contention by running services sequentially.
* - Nanonets: Document OCR with semantic tags
* - Qwen3: Text understanding and JSON extraction
*/ */
import { tap, expect } from '@git.zone/tstest/tapbundle'; import { tap, expect } from '@git.zone/tstest/tapbundle';
import * as fs from 'fs'; import * as fs from 'fs';
import * as path from 'path'; import * as path from 'path';
import { execSync } from 'child_process'; import { execSync } from 'child_process';
import * as os from 'os'; import * as os from 'os';
import { ensureNanonetsOcr, ensureMiniCpm } from './helpers/docker.js'; import { ensureNanonetsOcr, ensureMiniCpm, isContainerRunning } from './helpers/docker.js';
const NANONETS_URL = 'http://localhost:8000/v1'; const NANONETS_URL = 'http://localhost:8000/v1';
const NANONETS_MODEL = 'nanonets/Nanonets-OCR-s'; const NANONETS_MODEL = 'nanonets/Nanonets-OCR-s';
@@ -21,6 +19,9 @@ const NANONETS_MODEL = 'nanonets/Nanonets-OCR-s';
const OLLAMA_URL = 'http://localhost:11434'; const OLLAMA_URL = 'http://localhost:11434';
const QWEN_MODEL = 'qwen3:8b'; const QWEN_MODEL = 'qwen3:8b';
// Temp directory for storing markdown between stages
const TEMP_MD_DIR = path.join(os.tmpdir(), 'nanonets-invoices-markdown');
interface IInvoice { interface IInvoice {
invoice_number: string; invoice_number: string;
invoice_date: string; invoice_date: string;
@@ -31,6 +32,13 @@ interface IInvoice {
total_amount: number; total_amount: number;
} }
interface ITestCase {
name: string;
pdfPath: string;
jsonPath: string;
markdownPath?: string;
}
// Nanonets-specific prompt for document OCR to markdown // Nanonets-specific prompt for document OCR to markdown
const NANONETS_OCR_PROMPT = `Extract the text from the above document as if you were reading it naturally. const NANONETS_OCR_PROMPT = `Extract the text from the above document as if you were reading it naturally.
Return the tables in html format. Return the tables in html format.
@@ -66,14 +74,13 @@ INVOICE TEXT:
`; `;
/** /**
* Convert PDF to PNG images using ImageMagick * Convert PDF to PNG images
*/ */
function convertPdfToImages(pdfPath: string): string[] { function convertPdfToImages(pdfPath: string): string[] {
const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'pdf-convert-')); const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), 'pdf-convert-'));
const outputPattern = path.join(tempDir, 'page-%d.png'); const outputPattern = path.join(tempDir, 'page-%d.png');
try { try {
// Use 150 DPI to keep images within model's context length
execSync( execSync(
`convert -density 150 -quality 90 "${pdfPath}" -background white -alpha remove "${outputPattern}"`, `convert -density 150 -quality 90 "${pdfPath}" -background white -alpha remove "${outputPattern}"`,
{ stdio: 'pipe' } { stdio: 'pipe' }
@@ -95,10 +102,9 @@ function convertPdfToImages(pdfPath: string): string[] {
} }
/** /**
* Stage 1: Convert a single page to markdown using Nanonets-OCR-s * Convert a single page to markdown using Nanonets-OCR-s
*/ */
async function convertPageToMarkdown(image: string, pageNum: number): Promise<string> { async function convertPageToMarkdown(image: string, pageNum: number): Promise<string> {
console.log(` [Nanonets] Converting page ${pageNum} to markdown...`);
const startTime = Date.now(); const startTime = Date.now();
const response = await fetch(`${NANONETS_URL}/chat/completions`, { const response = await fetch(`${NANONETS_URL}/chat/completions`, {
@@ -125,21 +131,20 @@ async function convertPageToMarkdown(image: string, pageNum: number): Promise<st
if (!response.ok) { if (!response.ok) {
const errorText = await response.text(); const errorText = await response.text();
console.log(` [Nanonets] ERROR page ${pageNum}: ${response.status} - ${errorText}`); throw new Error(`Nanonets API error: ${response.status} - ${errorText}`);
throw new Error(`Nanonets API error: ${response.status}`);
} }
const data = await response.json(); const data = await response.json();
const content = (data.choices?.[0]?.message?.content || '').trim(); const content = (data.choices?.[0]?.message?.content || '').trim();
console.log(` [Nanonets] Page ${pageNum} converted (${elapsed}s, ${content.length} chars)`); console.log(` Page ${pageNum}: ${content.length} chars (${elapsed}s)`);
return content; return content;
} }
/** /**
* Stage 1: Convert all pages to markdown using Nanonets-OCR-s * Convert all pages of a document to markdown
*/ */
async function convertDocumentToMarkdown(images: string[]): Promise<string> { async function convertDocumentToMarkdown(images: string[], docName: string): Promise<string> {
console.log(` [Stage 1] Converting ${images.length} page(s) to markdown with Nanonets-OCR-s...`); console.log(` [${docName}] Converting ${images.length} page(s)...`);
const markdownPages: string[] = []; const markdownPages: string[] = [];
@@ -149,10 +154,24 @@ async function convertDocumentToMarkdown(images: string[]): Promise<string> {
} }
const fullMarkdown = markdownPages.join('\n\n'); const fullMarkdown = markdownPages.join('\n\n');
console.log(` [Stage 1] Complete: ${fullMarkdown.length} chars total`); console.log(` [${docName}] Complete: ${fullMarkdown.length} chars total`);
return fullMarkdown; return fullMarkdown;
} }
/**
* Stop Nanonets container
*/
function stopNanonets(): void {
console.log(' [Docker] Stopping Nanonets container...');
try {
execSync('docker stop nanonets-test 2>/dev/null || true', { stdio: 'pipe' });
execSync('sleep 5', { stdio: 'pipe' });
console.log(' [Docker] Nanonets stopped');
} catch {
console.log(' [Docker] Nanonets was not running');
}
}
/** /**
* Ensure Qwen3 model is available * Ensure Qwen3 model is available
*/ */
@@ -190,7 +209,6 @@ function parseAmount(s: string | number | undefined): number {
const match = s.match(/([\d.,]+)/); const match = s.match(/([\d.,]+)/);
if (!match) return 0; if (!match) return 0;
const numStr = match[1]; const numStr = match[1];
// Handle European format: 1.234,56 -> 1234.56
const normalized = numStr.includes(',') && numStr.indexOf(',') > numStr.lastIndexOf('.') const normalized = numStr.includes(',') && numStr.indexOf(',') > numStr.lastIndexOf('.')
? numStr.replace(/\./g, '').replace(',', '.') ? numStr.replace(/\./g, '').replace(',', '.')
: numStr.replace(/,/g, ''); : numStr.replace(/,/g, '');
@@ -204,10 +222,10 @@ function extractInvoiceNumber(s: string | undefined): string {
if (!s) return ''; if (!s) return '';
let clean = s.replace(/\*\*/g, '').replace(/`/g, '').trim(); let clean = s.replace(/\*\*/g, '').replace(/`/g, '').trim();
const patterns = [ const patterns = [
/\b([A-Z]{2,3}\d{10,})\b/i, // IEE2022006460244 /\b([A-Z]{2,3}\d{10,})\b/i,
/\b([A-Z]\d{8,})\b/i, // R0014359508 /\b([A-Z]\d{8,})\b/i,
/\b(INV[-\s]?\d{4}[-\s]?\d+)\b/i, // INV-2024-001 /\b(INV[-\s]?\d{4}[-\s]?\d+)\b/i,
/\b(\d{7,})\b/, // 1579087430 /\b(\d{7,})\b/,
]; ];
for (const pattern of patterns) { for (const pattern of patterns) {
const match = clean.match(pattern); const match = clean.match(pattern);
@@ -224,7 +242,6 @@ function extractDate(s: string | undefined): string {
let clean = s.replace(/\*\*/g, '').replace(/`/g, '').trim(); let clean = s.replace(/\*\*/g, '').replace(/`/g, '').trim();
const isoMatch = clean.match(/(\d{4}-\d{2}-\d{2})/); const isoMatch = clean.match(/(\d{4}-\d{2}-\d{2})/);
if (isoMatch) return isoMatch[1]; 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})/); const dmyMatch = clean.match(/(\d{1,2})[\/.](\d{1,2})[\/.](\d{4})/);
if (dmyMatch) { if (dmyMatch) {
return `${dmyMatch[3]}-${dmyMatch[2].padStart(2, '0')}-${dmyMatch[1].padStart(2, '0')}`; return `${dmyMatch[3]}-${dmyMatch[2].padStart(2, '0')}-${dmyMatch[1].padStart(2, '0')}`;
@@ -245,20 +262,16 @@ function extractCurrency(s: string | undefined): string {
} }
/** /**
* Extract JSON from response (handles markdown code blocks) * Extract JSON from response
*/ */
function extractJsonFromResponse(response: string): Record<string, unknown> | null { function extractJsonFromResponse(response: string): Record<string, unknown> | null {
// Remove thinking tags if present (Qwen3 may include <think>...</think>)
let cleanResponse = response.replace(/<think>[\s\S]*?<\/think>/g, '').trim(); let cleanResponse = response.replace(/<think>[\s\S]*?<\/think>/g, '').trim();
// Try to find JSON in markdown code block
const codeBlockMatch = cleanResponse.match(/```(?:json)?\s*([\s\S]*?)```/); const codeBlockMatch = cleanResponse.match(/```(?:json)?\s*([\s\S]*?)```/);
const jsonStr = codeBlockMatch ? codeBlockMatch[1].trim() : cleanResponse; const jsonStr = codeBlockMatch ? codeBlockMatch[1].trim() : cleanResponse;
try { try {
return JSON.parse(jsonStr); return JSON.parse(jsonStr);
} catch { } catch {
// Try to find JSON object pattern
const jsonMatch = jsonStr.match(/\{[\s\S]*\}/); const jsonMatch = jsonStr.match(/\{[\s\S]*\}/);
if (jsonMatch) { if (jsonMatch) {
try { try {
@@ -290,15 +303,16 @@ function parseJsonToInvoice(response: string): IInvoice | null {
} }
/** /**
* Stage 2: Extract invoice from markdown using Qwen3 * Extract invoice from markdown using Qwen3
*/ */
async function extractInvoiceFromMarkdown(markdown: string, queryId: string): Promise<IInvoice | null> { async function extractInvoiceFromMarkdown(markdown: string, queryId: string): Promise<IInvoice | null> {
console.log(` [${queryId}] Sending markdown to ${QWEN_MODEL}...`); console.log(` [${queryId}] Sending to ${QWEN_MODEL}...`);
const startTime = Date.now(); const startTime = Date.now();
const response = await fetch(`${OLLAMA_URL}/api/chat`, { const response = await fetch(`${OLLAMA_URL}/api/chat`, {
method: 'POST', method: 'POST',
headers: { 'Content-Type': 'application/json' }, headers: { 'Content-Type': 'application/json' },
signal: AbortSignal.timeout(600000), // 10 minute timeout for large documents
body: JSON.stringify({ body: JSON.stringify({
model: QWEN_MODEL, model: QWEN_MODEL,
messages: [{ messages: [{
@@ -322,13 +336,13 @@ async function extractInvoiceFromMarkdown(markdown: string, queryId: string): Pr
const data = await response.json(); const data = await response.json();
const content = (data.message?.content || '').trim(); const content = (data.message?.content || '').trim();
console.log(` [${queryId}] Response received (${elapsed}s, ${content.length} chars)`); console.log(` [${queryId}] Response: ${content.length} chars (${elapsed}s)`);
return parseJsonToInvoice(content); return parseJsonToInvoice(content);
} }
/** /**
* Compare two invoices for consensus (key fields must match) * Compare two invoices for consensus
*/ */
function invoicesMatch(a: IInvoice, b: IInvoice): boolean { function invoicesMatch(a: IInvoice, b: IInvoice): boolean {
const numMatch = a.invoice_number.toLowerCase() === b.invoice_number.toLowerCase(); const numMatch = a.invoice_number.toLowerCase() === b.invoice_number.toLowerCase();
@@ -338,45 +352,39 @@ function invoicesMatch(a: IInvoice, b: IInvoice): boolean {
} }
/** /**
* Stage 2: Extract invoice using Qwen3 with consensus * Extract with consensus
*/ */
async function extractWithConsensus(markdown: string): Promise<IInvoice> { async function extractWithConsensus(markdown: string, docName: string): Promise<IInvoice> {
const MAX_ATTEMPTS = 3; const MAX_ATTEMPTS = 3;
console.log(` [Stage 2] Extracting invoice with ${QWEN_MODEL} (consensus)...`);
for (let attempt = 1; attempt <= MAX_ATTEMPTS; attempt++) { for (let attempt = 1; attempt <= MAX_ATTEMPTS; attempt++) {
console.log(`\n [Stage 2] --- Attempt ${attempt}/${MAX_ATTEMPTS} ---`); console.log(` [${docName}] Attempt ${attempt}/${MAX_ATTEMPTS}`);
// Extract twice const inv1 = await extractInvoiceFromMarkdown(markdown, `${docName}-A${attempt}Q1`);
const inv1 = await extractInvoiceFromMarkdown(markdown, `A${attempt}Q1`); const inv2 = await extractInvoiceFromMarkdown(markdown, `${docName}-A${attempt}Q2`);
const inv2 = await extractInvoiceFromMarkdown(markdown, `A${attempt}Q2`);
if (!inv1 || !inv2) { if (!inv1 || !inv2) {
console.log(` [Stage 2] Parsing failed, retrying...`); console.log(` [${docName}] Parsing failed, retrying...`);
continue; continue;
} }
console.log(` [Stage 2] Q1: ${inv1.invoice_number} | ${inv1.invoice_date} | ${inv1.total_amount} ${inv1.currency}`); console.log(` [${docName}] Q1: ${inv1.invoice_number} | ${inv1.invoice_date} | ${inv1.total_amount}`);
console.log(` [Stage 2] Q2: ${inv2.invoice_number} | ${inv2.invoice_date} | ${inv2.total_amount} ${inv2.currency}`); console.log(` [${docName}] Q2: ${inv2.invoice_number} | ${inv2.invoice_date} | ${inv2.total_amount}`);
if (invoicesMatch(inv1, inv2)) { if (invoicesMatch(inv1, inv2)) {
console.log(` [Stage 2] CONSENSUS REACHED`); console.log(` [${docName}] CONSENSUS`);
return inv2; return inv2;
} }
console.log(` [${docName}] No consensus`);
console.log(` [Stage 2] NO CONSENSUS`);
} }
// Fallback: use last response // Fallback
console.log(`\n [Stage 2] === FALLBACK ===`); const fallback = await extractInvoiceFromMarkdown(markdown, `${docName}-FALLBACK`);
const fallback = await extractInvoiceFromMarkdown(markdown, 'FALLBACK');
if (fallback) { if (fallback) {
console.log(` [Stage 2] ~ FALLBACK: ${fallback.invoice_number} | ${fallback.invoice_date} | ${fallback.total_amount}`); console.log(` [${docName}] FALLBACK: ${fallback.invoice_number} | ${fallback.invoice_date} | ${fallback.total_amount}`);
return fallback; return fallback;
} }
// Return empty invoice if all else fails
return { return {
invoice_number: '', invoice_number: '',
invoice_date: '', invoice_date: '',
@@ -388,19 +396,6 @@ async function extractWithConsensus(markdown: string): Promise<IInvoice> {
}; };
} }
/**
* Full pipeline: PDF -> Images -> Markdown -> JSON
*/
async function extractInvoice(images: string[]): Promise<IInvoice> {
// Stage 1: Convert to markdown
const markdown = await convertDocumentToMarkdown(images);
// Stage 2: Extract invoice with consensus
const invoice = await extractWithConsensus(markdown);
return invoice;
}
/** /**
* Normalize date to YYYY-MM-DD * Normalize date to YYYY-MM-DD
*/ */
@@ -435,45 +430,38 @@ function compareInvoice(
): { match: boolean; errors: string[] } { ): { match: boolean; errors: string[] } {
const errors: string[] = []; const errors: string[] = [];
// Compare invoice number (normalize by removing spaces and case)
const extNum = extracted.invoice_number?.replace(/\s/g, '').toLowerCase() || ''; const extNum = extracted.invoice_number?.replace(/\s/g, '').toLowerCase() || '';
const expNum = expected.invoice_number?.replace(/\s/g, '').toLowerCase() || ''; const expNum = expected.invoice_number?.replace(/\s/g, '').toLowerCase() || '';
if (extNum !== expNum) { if (extNum !== expNum) {
errors.push(`invoice_number: expected "${expected.invoice_number}", got "${extracted.invoice_number}"`); errors.push(`invoice_number: exp "${expected.invoice_number}", got "${extracted.invoice_number}"`);
} }
// Compare date
if (normalizeDate(extracted.invoice_date) !== normalizeDate(expected.invoice_date)) { if (normalizeDate(extracted.invoice_date) !== normalizeDate(expected.invoice_date)) {
errors.push(`invoice_date: expected "${expected.invoice_date}", got "${extracted.invoice_date}"`); errors.push(`invoice_date: exp "${expected.invoice_date}", got "${extracted.invoice_date}"`);
} }
// Compare total amount (with tolerance)
if (Math.abs(extracted.total_amount - expected.total_amount) > 0.02) { if (Math.abs(extracted.total_amount - expected.total_amount) > 0.02) {
errors.push(`total_amount: expected ${expected.total_amount}, got ${extracted.total_amount}`); errors.push(`total_amount: exp ${expected.total_amount}, got ${extracted.total_amount}`);
} }
// Compare currency
if (extracted.currency?.toUpperCase() !== expected.currency?.toUpperCase()) { if (extracted.currency?.toUpperCase() !== expected.currency?.toUpperCase()) {
errors.push(`currency: expected "${expected.currency}", got "${extracted.currency}"`); errors.push(`currency: exp "${expected.currency}", got "${extracted.currency}"`);
} }
return { match: errors.length === 0, errors }; return { match: errors.length === 0, errors };
} }
/** /**
* Find all test cases (PDF + JSON pairs) in .nogit/invoices/ * Find all test cases
*/ */
function findTestCases(): Array<{ name: string; pdfPath: string; jsonPath: string }> { function findTestCases(): ITestCase[] {
const testDir = path.join(process.cwd(), '.nogit/invoices'); const testDir = path.join(process.cwd(), '.nogit/invoices');
if (!fs.existsSync(testDir)) { if (!fs.existsSync(testDir)) return [];
return [];
}
const files = fs.readdirSync(testDir); const files = fs.readdirSync(testDir);
const pdfFiles = files.filter((f) => f.endsWith('.pdf')); const testCases: ITestCase[] = [];
const testCases: Array<{ name: string; pdfPath: string; jsonPath: string }> = [];
for (const pdf of pdfFiles) { for (const pdf of files.filter((f) => f.endsWith('.pdf'))) {
const baseName = pdf.replace('.pdf', ''); const baseName = pdf.replace('.pdf', '');
const jsonFile = `${baseName}.json`; const jsonFile = `${baseName}.json`;
if (files.includes(jsonFile)) { if (files.includes(jsonFile)) {
@@ -485,90 +473,114 @@ function findTestCases(): Array<{ name: string; pdfPath: string; jsonPath: strin
} }
} }
testCases.sort((a, b) => a.name.localeCompare(b.name)); return testCases.sort((a, b) => a.name.localeCompare(b.name));
return testCases;
} }
// Tests // ============ TESTS ============
tap.test('setup: ensure containers are running', async () => { const testCases = findTestCases();
console.log('\n[Setup] Checking Docker containers...\n'); console.log(`\nFound ${testCases.length} invoice test cases\n`);
// Nanonets for OCR // Ensure temp directory exists
const nanonetsOk = await ensureNanonetsOcr(); if (!fs.existsSync(TEMP_MD_DIR)) {
expect(nanonetsOk).toBeTrue(); fs.mkdirSync(TEMP_MD_DIR, { recursive: true });
}
// -------- STAGE 1: OCR with Nanonets --------
tap.test('Stage 1: Setup Nanonets', async () => {
console.log('\n========== STAGE 1: Nanonets OCR ==========\n');
const ok = await ensureNanonetsOcr();
expect(ok).toBeTrue();
});
tap.test('Stage 1: Convert all invoices to markdown', async () => {
console.log('\n Converting all invoice PDFs to markdown with Nanonets-OCR-s...\n');
for (const tc of testCases) {
console.log(`\n === ${tc.name} ===`);
const images = convertPdfToImages(tc.pdfPath);
console.log(` Pages: ${images.length}`);
const markdown = await convertDocumentToMarkdown(images, tc.name);
const mdPath = path.join(TEMP_MD_DIR, `${tc.name}.md`);
fs.writeFileSync(mdPath, markdown);
tc.markdownPath = mdPath;
console.log(` Saved: ${mdPath}`);
}
console.log('\n Stage 1 complete: All invoices converted to markdown\n');
});
tap.test('Stage 1: Stop Nanonets', async () => {
stopNanonets();
await new Promise(resolve => setTimeout(resolve, 3000));
expect(isContainerRunning('nanonets-test')).toBeFalse();
});
// -------- STAGE 2: Extraction with Qwen3 --------
tap.test('Stage 2: Setup Ollama + Qwen3', async () => {
console.log('\n========== STAGE 2: Qwen3 Extraction ==========\n');
// Ollama for Qwen3
const ollamaOk = await ensureMiniCpm(); const ollamaOk = await ensureMiniCpm();
expect(ollamaOk).toBeTrue(); expect(ollamaOk).toBeTrue();
// Qwen3 model
const qwenOk = await ensureQwen3(); const qwenOk = await ensureQwen3();
expect(qwenOk).toBeTrue(); expect(qwenOk).toBeTrue();
console.log('\n[Setup] All containers ready!\n');
}); });
tap.test('should have models available', async () => {
// Check Nanonets
const nanonetsResp = await fetch(`${NANONETS_URL}/models`);
expect(nanonetsResp.ok).toBeTrue();
// Check Qwen3
const ollamaResp = await fetch(`${OLLAMA_URL}/api/tags`);
expect(ollamaResp.ok).toBeTrue();
const data = await ollamaResp.json();
const modelNames = data.models.map((m: { name: string }) => m.name);
expect(modelNames.some((name: string) => name.includes('qwen3'))).toBeTrue();
});
const testCases = findTestCases();
console.log(`\nFound ${testCases.length} invoice test cases (Nanonets + Qwen3)\n`);
let passedCount = 0; let passedCount = 0;
let failedCount = 0; let failedCount = 0;
const processingTimes: number[] = []; const processingTimes: number[] = [];
for (const testCase of testCases) { for (const tc of testCases) {
tap.test(`should extract invoice: ${testCase.name}`, async () => { tap.test(`Stage 2: Extract ${tc.name}`, async () => {
const expected: IInvoice = JSON.parse(fs.readFileSync(testCase.jsonPath, 'utf-8')); const expected: IInvoice = JSON.parse(fs.readFileSync(tc.jsonPath, 'utf-8'));
console.log(`\n=== ${testCase.name} ===`); console.log(`\n === ${tc.name} ===`);
console.log(`Expected: ${expected.invoice_number} | ${expected.invoice_date} | ${expected.total_amount} ${expected.currency}`); console.log(` Expected: ${expected.invoice_number} | ${expected.invoice_date} | ${expected.total_amount} ${expected.currency}`);
const startTime = Date.now(); const startTime = Date.now();
const images = convertPdfToImages(testCase.pdfPath);
console.log(` Pages: ${images.length}`);
const extracted = await extractInvoice(images); const mdPath = path.join(TEMP_MD_DIR, `${tc.name}.md`);
console.log(` Extracted: ${extracted.invoice_number} | ${extracted.invoice_date} | ${extracted.total_amount} ${extracted.currency}`); if (!fs.existsSync(mdPath)) {
throw new Error(`Markdown not found: ${mdPath}. Run Stage 1 first.`);
}
const markdown = fs.readFileSync(mdPath, 'utf-8');
console.log(` Markdown: ${markdown.length} chars`);
const extracted = await extractWithConsensus(markdown, tc.name);
const elapsedMs = Date.now() - startTime; const elapsedMs = Date.now() - startTime;
processingTimes.push(elapsedMs); processingTimes.push(elapsedMs);
console.log(` Extracted: ${extracted.invoice_number} | ${extracted.invoice_date} | ${extracted.total_amount} ${extracted.currency}`);
const result = compareInvoice(extracted, expected); const result = compareInvoice(extracted, expected);
if (result.match) { if (result.match) {
passedCount++; passedCount++;
console.log(` Result: MATCH (${(elapsedMs / 1000).toFixed(1)}s)`); console.log(` Result: MATCH (${(elapsedMs / 1000).toFixed(1)}s)`);
} else { } else {
failedCount++; failedCount++;
console.log(` Result: MISMATCH (${(elapsedMs / 1000).toFixed(1)}s)`); console.log(` Result: MISMATCH (${(elapsedMs / 1000).toFixed(1)}s)`);
result.errors.forEach((e) => console.log(` - ${e}`)); result.errors.forEach(e => console.log(` - ${e}`));
} }
expect(result.match).toBeTrue(); expect(result.match).toBeTrue();
}); });
} }
tap.test('summary', async () => { tap.test('Summary', async () => {
const totalInvoices = testCases.length; const totalInvoices = testCases.length;
const accuracy = totalInvoices > 0 ? (passedCount / totalInvoices) * 100 : 0; const accuracy = totalInvoices > 0 ? (passedCount / totalInvoices) * 100 : 0;
const totalTimeMs = processingTimes.reduce((a, b) => a + b, 0); const totalTimeMs = processingTimes.reduce((a, b) => a + b, 0);
const avgTimeSec = processingTimes.length > 0 ? totalTimeMs / processingTimes.length / 1000 : 0; const avgTimeSec = processingTimes.length > 0 ? totalTimeMs / processingTimes.length / 1000 : 0;
console.log(`\n========================================`); console.log(`\n========================================`);
console.log(` Invoice Extraction Summary`); console.log(` Invoice Summary (Nanonets + Qwen3)`);
console.log(` (Nanonets + Qwen3 Pipeline)`);
console.log(`========================================`); console.log(`========================================`);
console.log(` Stage 1: Nanonets-OCR-s (doc -> md)`); console.log(` Stage 1: Nanonets-OCR-s (doc -> md)`);
console.log(` Stage 2: Qwen3 8B (md -> JSON)`); console.log(` Stage 2: Qwen3 8B (md -> JSON)`);
@@ -579,6 +591,14 @@ tap.test('summary', async () => {
console.log(` Total time: ${(totalTimeMs / 1000).toFixed(1)}s`); console.log(` Total time: ${(totalTimeMs / 1000).toFixed(1)}s`);
console.log(` Avg per inv: ${avgTimeSec.toFixed(1)}s`); console.log(` Avg per inv: ${avgTimeSec.toFixed(1)}s`);
console.log(`========================================\n`); console.log(`========================================\n`);
// Cleanup temp files
try {
fs.rmSync(TEMP_MD_DIR, { recursive: true, force: true });
console.log(` Cleaned up temp directory: ${TEMP_MD_DIR}\n`);
} catch {
// Ignore
}
}); });
export default tap.start(); export default tap.start();