test

Author: pashidlos

docker-compose.yml

@@ -16,13 +16,11 @@ services:
       APP_FRONTEND_URL: ${APP_FRONTEND_URL}
       BODY_PARSER_JSON_LIMIT: ${BODY_PARSER_JSON_LIMIT}
       ELASTIC_URL: ${ELASTIC_URL}
-      # VLM: Uncomment to use Ollama running on host machine
       OLLAMA_BASE_URL: http://host.docker.internal:11434
     ports:
       - "${APP_PORT}:3000"
     expose:
       - "${APP_PORT}"
-    # VLM: Uncomment to use Ollama running on host machine
     extra_hosts:
       - host.docker.internal:host-gateway
     depends_on:

src/compare/libs/vlm/README.md

@@ -41,7 +41,7 @@ Set project's image comparison to `vlm` with config:
 }
 ```
 
-Optional custom prompt:
+Optional custom prompt (replaces default system prompt):
 ```json
 {
   "model": "llava:7b",
@@ -50,11 +50,15 @@ Optional custom prompt:
 }
 ```
 
+**Note:** The `prompt` field replaces the entire system prompt. If omitted, a default system prompt is used that focuses on semantic differences while ignoring rendering artifacts.
+
 ## Recommended Models
 
 | Model | Size | Speed | Accuracy | Best For |
 |-------|------|-------|----------|----------|
-| `llava:7b` | 4.7GB | ⚡⚡ | ⭐⭐⭐ | **Recommended** - best balance |
+| `llava:7b` | 4.7GB | ⚡⚡ | ⭐⭐⭐ | **Recommended** - best balance (minimal) |
+| `qwen3-vl:8b` | ~8GB | ⚡⚡ | ⭐⭐⭐ | Minimal model option |
+| `gemma3:latest` | ~ | ⚡⚡ | ⭐⭐⭐ | Minimal model option |
 | `llava:13b` | 8GB | ⚡ | ⭐⭐⭐⭐ | Best accuracy |
 | `moondream` | 1.7GB | ⚡⚡⚡ | ⭐⭐ | Fast, may hallucinate |
 | `minicpm-v` | 5.5GB | ⚡⚡ | ⭐⭐⭐ | Good alternative |
@@ -63,16 +67,23 @@ Optional custom prompt:
 
 | Option | Type | Default | Description |
 |--------|------|---------|-------------|
-| `model` | string | `moondream` | Ollama vision model name |
-| `prompt` | string | `""` | Custom context prepended to system prompt |
-| `temperature` | number | `0.1` | Lower = more consistent results |
+| `model` | string | `llava:7b` | Ollama vision model name |
+| `prompt` | string | System prompt (see below) | Custom prompt for image comparison |
+| `temperature` | number | `0.1` | Lower = more consistent results (0.0-1.0) |
 
 ## How It Works
 
 1. VLM analyzes both images semantically
-2. Returns `YES` (pass) or `NO` (fail) based on meaningful differences
-3. Ignores technical differences (anti-aliasing, sub-pixel, minor spacing)
-4. Provides description of differences found
+2. Returns JSON with `{"identical": true/false, "description": "..."}` 
+3. `identical: true` = images match (pass), `identical: false` = differences found (fail)
+4. Ignores technical differences (anti-aliasing, shadows, 1-2px shifts)
+5. Provides description of differences found
+
+### Default System Prompt
+
+The default prompt instructs the model to:
+- **CHECK** for: data changes, missing/added elements, state changes, structural differences
+- **IGNORE**: rendering artifacts, anti-aliasing, shadows, minor pixel shifts
 
 ## API Endpoints
 
@@ -83,10 +94,3 @@ GET /ollama/models
 # Compare two images (for testing)
 POST /ollama/compare?model=llava:7b&prompt=<prompt>&temperature=0.1
 ```
-
-**Example:**
-```bash
-curl -X POST "http://localhost:3000/ollama/compare?model=llava:7b&prompt=Are%20these%20images%20the%20same&temperature=0.1" \
-  -F "images=@baseline.png" \
-  -F "images=@comparison.png"
-```

src/compare/libs/vlm/ollama.controller.ts

@@ -1,4 +1,13 @@
-import { Controller, Get, Post, Query, HttpException, HttpStatus, UseInterceptors, UploadedFiles } from '@nestjs/common';
+import {
+  Controller,
+  Get,
+  Post,
+  Query,
+  HttpException,
+  HttpStatus,
+  UseInterceptors,
+  UploadedFiles,
+} from '@nestjs/common';
 import { FilesInterceptor } from '@nestjs/platform-express';
 import { ApiTags, ApiConsumes, ApiBody } from '@nestjs/swagger';
 import { OllamaService } from './ollama.service';

src/compare/libs/vlm/ollama.service.spec.ts

@@ -52,9 +52,9 @@ describe('OllamaService', () => {
         text: () => Promise.resolve('Internal Server Error'),
       });
 
-      await expect(
-        service.generate({ model: 'llava', prompt: 'Test' })
-      ).rejects.toThrow('Ollama API returned status 500');
+      await expect(service.generate({ model: 'llava', prompt: 'Test' })).rejects.toThrow(
+        'Ollama API returned status 500'
+      );
     });
 
     it('should throw error when OLLAMA_BASE_URL is not configured', async () => {
@@ -65,9 +65,7 @@ describe('OllamaService', () => {
       } as any;
       const newService = new OllamaService(mockConfigService);
 
-      await expect(
-        newService.generate({ model: 'llava', prompt: 'Test' })
-      ).rejects.toThrow('OLLAMA_BASE_URL');
+      await expect(newService.generate({ model: 'llava', prompt: 'Test' })).rejects.toThrow('OLLAMA_BASE_URL');
     });
   });
 

src/compare/libs/vlm/ollama.service.ts

@@ -1,11 +1,6 @@
 import { Injectable, Logger } from '@nestjs/common';
 import { ConfigService } from '@nestjs/config';
-import {
-  OllamaGenerateRequest,
-  OllamaGenerateResponse,
-  OllamaModel,
-  OllamaModelsResponse,
-} from './ollama.types';
+import { OllamaGenerateRequest, OllamaGenerateResponse, OllamaModel, OllamaModelsResponse } from './ollama.types';
 
 @Injectable()
 export class OllamaService {
@@ -60,4 +55,3 @@ export class OllamaService {
     }
   }
 }
-

src/compare/libs/vlm/ollama.types.ts

@@ -38,4 +38,3 @@ export interface OllamaModel {
 export interface OllamaModelsResponse {
   models: OllamaModel[];
 }
-

src/compare/libs/vlm/vlm.service.spec.ts

@@ -6,11 +6,7 @@ import { NO_BASELINE_RESULT } from '../consts';
 import { DEFAULT_CONFIG, VlmService } from './vlm.service';
 import { OllamaService } from './ollama.service';
 
-const initService = async ({
-  getImageMock = jest.fn(),
-  saveImageMock = jest.fn(),
-  ollamaGenerateMock = jest.fn(),
-}) => {
+const initService = async ({ getImageMock = jest.fn(), saveImageMock = jest.fn(), ollamaGenerateMock = jest.fn() }) => {
   const module: TestingModule = await Test.createTestingModule({
     providers: [
       VlmService,