chore: reduce layout threshold to 0.2, tune vLLM memory, and add test scripts and reports

backend/config/pipeline_config_vllm.yaml

@@ -25,7 +25,7 @@ SubModules:
     model_name: PP-DocLayoutV3
     model_dir: null
     batch_size: 8
-    threshold: 0.3
+    threshold: 0.2
     layout_nms: True
     layout_unclip_ratio: [1.0, 1.0]
     layout_merge_bboxes_mode:

backend/config/vllm_config.yaml

@@ -1,6 +1,6 @@
 # vLLM backend tuning for paddleocr genai_server
 # Docs: https://www.paddleocr.ai/latest/en/version3.x/pipeline_usage/PaddleOCR-VL.html#331-server-side-parameter-adjustment
-gpu-memory-utilization: 0.6
+gpu-memory-utilization: 0.35
 max-num-seqs: 4
 enforce-eager: true
 max-model-len: 2048

backend/pfm-web-app/reparse_unparsed.js

@@ -0,0 +1,44 @@
+const { Client } = require('pg');
+
+async function main() {
+  const client = new Client({
+    host: process.env.PGHOST || "paddleocr-db",
+    port: parseInt(process.env.PGPORT || "5432"),
+    user: process.env.PGUSER || "postgres",
+    password: process.env.PGPASSWORD || "postgres",
+    database: process.env.PGDATABASE || "dopfm",
+  });
+
+  await client.connect();
+  console.log('Connected to PG database.');
+
+  const res = await client.query("SELECT id, filename FROM documents WHERE parsed = false;");
+  console.log(`Found ${res.rows.length} documents to parse.`);
+
+  for (let i = 0; i < res.rows.length; i++) {
+    const row = res.rows[i];
+    console.log(`[${i+1}/${res.rows.length}] Reparsing ${row.filename} (ID: ${row.id})...`);
+    try {
+      const response = await fetch('http://localhost:3000/api/parse', {
+        method: 'POST',
+        headers: { 'Content-Type': 'application/json' },
+        body: JSON.stringify({ filename: row.filename })
+      });
+      if (response.ok) {
+        console.log(`Successfully triggered parse for ${row.filename}. Status: ${response.status}`);
+      } else {
+        console.error(`Failed to parse ${row.filename}. Status: ${response.status}, Error: ${await response.text()}`);
+      }
+    } catch (err) {
+      console.error(`Fetch error for ${row.filename}:`, err.message);
+    }
+  }
+
+  await client.end();
+  console.log('Done reparsing.');
+}
+
+main().catch(err => {
+  console.error('Fatal error:', err);
+  process.exit(1);
+});

backend/pfm-web-app/test_guided.js

@@ -0,0 +1,113 @@
+const fs = require('fs');
+const path = require('path');
+
+const PROXY_URL = 'http://localhost:8000/api/vllm-proxy/v1/chat/completions';
+const IMAGE_PATH = path.join(__dirname, '..', 'sources', 'test-images', '1782870899198-sample_do.jpeg');
+
+async function testGuidedDecoding() {
+  console.log('Reading test image from:', IMAGE_PATH);
+  if (!fs.existsSync(IMAGE_PATH)) {
+    console.error('Test image not found!');
+    process.exit(1);
+  }
+
+  const imageBuffer = fs.readFileSync(IMAGE_PATH);
+  const base64Image = imageBuffer.toString('base64');
+  const imageUrl = `data:image/jpeg;base64,${base64Image}`;
+
+  // JSON Schema for DO metadata
+  const jsonSchema = {
+    type: "object",
+    properties: {
+      tanggal: { type: "string" },
+      noPo: { type: "string" },
+      noSo: { type: "string" },
+      noDo: { type: "string" },
+      kepadaYth: { type: "string" },
+      orderUntuk: { type: "string" },
+      alamat: { type: "string" },
+      platTruk: { type: "string" },
+      namaDriver: { type: "string" },
+      namaPenerima: { type: "string" },
+      items: {
+        type: "array",
+        items: {
+          type: "object",
+          properties: {
+            nomor_sku: { type: "string" },
+            nama_barang: { type: "string" },
+            banyak: { type: "string" },
+            jumlah: { type: "string" }
+          },
+          required: ["nomor_sku", "nama_barang", "banyak", "jumlah"]
+        }
+      }
+    },
+    required: ["tanggal", "noPo", "noSo", "noDo", "kepadaYth", "items"]
+  };
+
+  const payload = {
+    model: "PaddleOCR-VL-1.6-0.9B",
+    messages: [
+      {
+        role: "user",
+        content: [
+          {
+            type: "image_url",
+            image_url: {
+              url: imageUrl
+            }
+          },
+          {
+            type: "text",
+            text: "Extract all structural details from this Delivery Order. Match the requested JSON Schema exactly."
+          }
+        ]
+      }
+    ],
+    temperature: 0.1,
+    max_tokens: 1024,
+    guided_json: JSON.stringify(jsonSchema) // standard vLLM guided JSON schema format
+  };
+
+  console.log('Sending request to vLLM proxy with JSON Schema...');
+  try {
+    const startTime = Date.now();
+    const response = await fetch(PROXY_URL, {
+      method: 'POST',
+      headers: {
+        'Content-Type': 'application/json'
+      },
+      body: JSON.stringify(payload)
+    });
+
+    console.log(`Response status: ${response.status} (${response.statusText})`);
+    const duration = ((Date.now() - startTime) / 1000).toFixed(2);
+    console.log(`Request completed in ${duration}s`);
+
+    const result = await response.json();
+    if (response.ok) {
+      console.log('=== SUCCESS RESPONSE ===');
+      console.log(JSON.stringify(result, null, 2));
+      
+      const content = result.choices?.[0]?.message?.content;
+      console.log('\n=== EXTRACTED CONTENT ===');
+      console.log(content);
+      
+      try {
+        const parsed = JSON.parse(content);
+        console.log('\nValid JSON parsed successfully! ✅');
+        console.log(parsed);
+      } catch (err) {
+        console.error('\nFailed to parse content as JSON! ❌', err.message);
+      }
+    } else {
+      console.error('=== ERROR RESPONSE ===');
+      console.error(result);
+    }
+  } catch (error) {
+    console.error('Request failed:', error);
+  }
+}
+
+testGuidedDecoding();

backend/pfm-web-app/test_notfound_images.js

@@ -0,0 +1,109 @@
+const fs = require('fs');
+const path = require('path');
+
+const PIPELINE_URL = 'http://localhost:8000/layout-parsing';
+const PROXY_URL = 'http://localhost:8000/api/vllm-proxy/v1/chat/completions';
+const UPLOADS_DIR = path.join(__dirname, '..', 'uploads');
+
+const notFoundImages = [
+  '1782888211457-rotated_1782888204591_rotated_1782888199962_rotated_1782888195402_CAP7202641176939787142.jpg',
+  '1782875064223-CAP2007290974474760139.jpg',
+  '1782884586047-CAP9169719214882332189.jpg',
+  '1782893409879-CAP4330863738757813156.jpg'
+];
+
+async function testImage(filename) {
+  console.log(`\n========================================`);
+  console.log(`TESTING FILE: ${filename}`);
+  console.log(`========================================`);
+
+  const filePath = path.join(UPLOADS_DIR, filename);
+  if (!fs.existsSync(filePath)) {
+    console.error(`File does not exist on disk: ${filePath}`);
+    return;
+  }
+
+  const fileBuffer = fs.readFileSync(filePath);
+  const base64Image = fileBuffer.toString('base64');
+
+  // Test 1: Hit the Layout Parsing Pipeline API
+  console.log('\n--- Test 1: Layout Parsing Pipeline (Standard) ---');
+  try {
+    const res = await fetch(PIPELINE_URL, {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({
+        file: base64Image,
+        matchHistoryJob: false,
+        useLayoutDetection: true,
+        fileType: 1,
+        useDocUnwarping: false,
+        useDocOrientationClassify: true
+      })
+    });
+
+    if (res.ok) {
+      const data = await res.json();
+      const markdown = data.result?.layoutParsingResults?.[0]?.markdown?.text || data.layoutParsingResults?.[0]?.markdown?.text || '';
+      console.log('Resulting Markdown snippet (first 300 chars):');
+      console.log(markdown.substring(0, 300));
+      console.log(`\nDoes it contain PO, SO, DO or Tanggal?`);
+      console.log(`- "Tanggal": ${/Tanggal/i.test(markdown)}`);
+      console.log(`- "SO": ${/SO/i.test(markdown)}`);
+      console.log(`- "DO": ${/DO/i.test(markdown)}`);
+      console.log(`- "PO": ${/PO/i.test(markdown)}`);
+    } else {
+      console.error(`Pipeline returned status ${res.status}: ${await res.text()}`);
+    }
+  } catch (err) {
+    console.error('Pipeline test failed:', err.message);
+  }
+
+  // Test 2: Direct vLLM completions with simple extraction prompt
+  console.log('\n--- Test 2: Direct vLLM Simple Extraction ---');
+  try {
+    const res = await fetch(PROXY_URL, {
+      method: 'POST',
+      headers: { 'Content-Type': 'application/json' },
+      body: JSON.stringify({
+        model: "PaddleOCR-VL-1.6-0.9B",
+        messages: [
+          {
+            role: "user",
+            content: [
+              {
+                type: "image_url",
+                image_url: { url: `data:image/jpeg;base64,${base64Image}` }
+              },
+              {
+                type: "text",
+                text: "Read the top right section of the document. Extract Tanggal, No. SO, No. DO, and No. PO."
+              }
+            ]
+          }
+        ],
+        temperature: 0.1,
+        max_tokens: 300
+      })
+    });
+
+    if (res.ok) {
+      const data = await res.json();
+      const content = data.choices?.[0]?.message?.content;
+      console.log('vLLM Response:');
+      console.log(content);
+    } else {
+      console.error(`vLLM proxy returned status ${res.status}: ${await res.text()}`);
+    }
+  } catch (err) {
+    console.error('vLLM test failed:', err.message);
+  }
+}
+
+async function runAll() {
+  for (const filename of notFoundImages) {
+    await testImage(filename);
+  }
+}
+
+runAll();