Image-Text-to-Text
Transformers
English
vision-language-model
vlm
surveillance
iot
gemma
vl-jepa
multimodal
object-detection
video-analytics
Instructions to use hardiksa/arcisvlm with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- Transformers
How to use hardiksa/arcisvlm with Transformers:
# Use a pipeline as a high-level helper from transformers import pipeline pipe = pipeline("image-text-to-text", model="hardiksa/arcisvlm")# Load model directly from transformers import AutoModel model = AutoModel.from_pretrained("hardiksa/arcisvlm", dtype="auto") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- vLLM
How to use hardiksa/arcisvlm with vLLM:
Install from pip and serve model
# Install vLLM from pip: pip install vllm # Start the vLLM server: vllm serve "hardiksa/arcisvlm" # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:8000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "hardiksa/arcisvlm", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker
docker model run hf.co/hardiksa/arcisvlm
- SGLang
How to use hardiksa/arcisvlm with SGLang:
Install from pip and serve model
# Install SGLang from pip: pip install sglang # Start the SGLang server: python3 -m sglang.launch_server \ --model-path "hardiksa/arcisvlm" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "hardiksa/arcisvlm", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }'Use Docker images
docker run --gpus all \ --shm-size 32g \ -p 30000:30000 \ -v ~/.cache/huggingface:/root/.cache/huggingface \ --env "HF_TOKEN=<secret>" \ --ipc=host \ lmsysorg/sglang:latest \ python3 -m sglang.launch_server \ --model-path "hardiksa/arcisvlm" \ --host 0.0.0.0 \ --port 30000 # Call the server using curl (OpenAI-compatible API): curl -X POST "http://localhost:30000/v1/completions" \ -H "Content-Type: application/json" \ --data '{ "model": "hardiksa/arcisvlm", "prompt": "Once upon a time,", "max_tokens": 512, "temperature": 0.5 }' - Docker Model Runner
How to use hardiksa/arcisvlm with Docker Model Runner:
docker model run hf.co/hardiksa/arcisvlm
| """ | |
| End-to-end integration tests for ArcisVLM cloud pipeline. | |
| Tests the full API contract, agent routing, alert CRUD, WebSocket connectivity, | |
| and orchestrator task classification — all without a GPU. | |
| Model inference is mocked; the goal is to verify the API layer and agent routing. | |
| """ | |
| import pytest | |
| import sys | |
| import os | |
| import base64 | |
| import io | |
| import json | |
| from unittest.mock import patch, MagicMock | |
| # Ensure project root is on path | |
| sys.path.insert(0, os.path.join(os.path.dirname(__file__), "..")) | |
| # Set env to skip real model loading | |
| os.environ["ARCISVLM_CONFIG"] = "configs/scale_1.3b.yaml" | |
| def _make_test_image_base64() -> str: | |
| """Generate a small random test image as base64-encoded JPEG.""" | |
| # Create a minimal valid JPEG without needing PIL | |
| # Use a 2x2 red pixel BMP converted concept — but simpler: just use raw bytes | |
| # Actually, let's create a tiny valid JPEG | |
| # Minimal JPEG: SOI + APP0 + minimal frame | |
| # Easier approach: create with PIL if available, otherwise use a known minimal JPEG | |
| try: | |
| from PIL import Image | |
| img = Image.new("RGB", (4, 4), color=(128, 64, 32)) | |
| buf = io.BytesIO() | |
| img.save(buf, format="JPEG") | |
| return base64.b64encode(buf.getvalue()).decode("utf-8") | |
| except ImportError: | |
| # Minimal valid JPEG bytes (1x1 pixel white) | |
| minimal_jpeg = bytes([ | |
| 0xFF, 0xD8, 0xFF, 0xE0, 0x00, 0x10, 0x4A, 0x46, 0x49, 0x46, 0x00, | |
| 0x01, 0x01, 0x00, 0x00, 0x01, 0x00, 0x01, 0x00, 0x00, 0xFF, 0xDB, | |
| 0x00, 0x43, 0x00, 0x08, 0x06, 0x06, 0x07, 0x06, 0x05, 0x08, 0x07, | |
| 0x07, 0x07, 0x09, 0x09, 0x08, 0x0A, 0x0C, 0x14, 0x0D, 0x0C, 0x0B, | |
| 0x0B, 0x0C, 0x19, 0x12, 0x13, 0x0F, 0x14, 0x1D, 0x1A, 0x1F, 0x1E, | |
| 0x1D, 0x1A, 0x1C, 0x1C, 0x20, 0x24, 0x2E, 0x27, 0x20, 0x22, 0x2C, | |
| 0x23, 0x1C, 0x1C, 0x28, 0x37, 0x29, 0x2C, 0x30, 0x31, 0x34, 0x34, | |
| 0x34, 0x1F, 0x27, 0x39, 0x3D, 0x38, 0x32, 0x3C, 0x2E, 0x33, 0x34, | |
| 0x32, 0xFF, 0xC0, 0x00, 0x0B, 0x08, 0x00, 0x01, 0x00, 0x01, 0x01, | |
| 0x01, 0x11, 0x00, 0xFF, 0xC4, 0x00, 0x1F, 0x00, 0x00, 0x01, 0x05, | |
| 0x01, 0x01, 0x01, 0x01, 0x01, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00, | |
| 0x00, 0x00, 0x00, 0x01, 0x02, 0x03, 0x04, 0x05, 0x06, 0x07, 0x08, | |
| 0x09, 0x0A, 0x0B, 0xFF, 0xC4, 0x00, 0xB5, 0x10, 0x00, 0x02, 0x01, | |
| 0x03, 0x03, 0x02, 0x04, 0x03, 0x05, 0x05, 0x04, 0x04, 0x00, 0x00, | |
| 0x01, 0x7D, 0x01, 0x02, 0x03, 0x00, 0x04, 0x11, 0x05, 0x12, 0x21, | |
| 0x31, 0x41, 0x06, 0x13, 0x51, 0x61, 0x07, 0x22, 0x71, 0x14, 0x32, | |
| 0x81, 0x91, 0xA1, 0x08, 0x23, 0x42, 0xB1, 0xC1, 0x15, 0x52, 0xD1, | |
| 0xF0, 0x24, 0x33, 0x62, 0x72, 0x82, 0x09, 0x0A, 0x16, 0x17, 0x18, | |
| 0x19, 0x1A, 0x25, 0x26, 0x27, 0x28, 0x29, 0x2A, 0x34, 0x35, 0x36, | |
| 0x37, 0x38, 0x39, 0x3A, 0x43, 0x44, 0x45, 0x46, 0x47, 0x48, 0x49, | |
| 0x4A, 0x53, 0x54, 0x55, 0x56, 0x57, 0x58, 0x59, 0x5A, 0x63, 0x64, | |
| 0x65, 0x66, 0x67, 0x68, 0x69, 0x6A, 0x73, 0x74, 0x75, 0x76, 0x77, | |
| 0x78, 0x79, 0x7A, 0x83, 0x84, 0x85, 0x86, 0x87, 0x88, 0x89, 0x8A, | |
| 0x92, 0x93, 0x94, 0x95, 0x96, 0x97, 0x98, 0x99, 0x9A, 0xA2, 0xA3, | |
| 0xA4, 0xA5, 0xA6, 0xA7, 0xA8, 0xA9, 0xAA, 0xB2, 0xB3, 0xB4, 0xB5, | |
| 0xB6, 0xB7, 0xB8, 0xB9, 0xBA, 0xC2, 0xC3, 0xC4, 0xC5, 0xC6, 0xC7, | |
| 0xC8, 0xC9, 0xCA, 0xD2, 0xD3, 0xD4, 0xD5, 0xD6, 0xD7, 0xD8, 0xD9, | |
| 0xDA, 0xE1, 0xE2, 0xE3, 0xE4, 0xE5, 0xE6, 0xE7, 0xE8, 0xE9, 0xEA, | |
| 0xF1, 0xF2, 0xF3, 0xF4, 0xF5, 0xF6, 0xF7, 0xF8, 0xF9, 0xFA, 0xFF, | |
| 0xDA, 0x00, 0x08, 0x01, 0x01, 0x00, 0x00, 0x3F, 0x00, 0x7B, 0x94, | |
| 0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0xFF, 0xD9, | |
| ]) | |
| return base64.b64encode(minimal_jpeg).decode("utf-8") | |
| def _mock_result(answer="mocked answer", confidence=0.85, expert_used="vqa", **kwargs): | |
| """Create a mock Result object.""" | |
| from agents.base import Result | |
| return Result( | |
| answer=answer, | |
| confidence=confidence, | |
| expert_used=expert_used, | |
| task_id=kwargs.get("task_id", "test-task-001"), | |
| metadata=kwargs.get("metadata", {}), | |
| ) | |
| # --------------------------------------------------------------------------- | |
| # Fixtures | |
| # --------------------------------------------------------------------------- | |
| def client(): | |
| """Create a TestClient with mocked model manager for integration tests.""" | |
| from fastapi.testclient import TestClient | |
| from api.main import app | |
| with TestClient(app) as c: | |
| yield c | |
| def test_image_b64(): | |
| """A small test image as base64.""" | |
| return _make_test_image_base64() | |
| # --------------------------------------------------------------------------- | |
| # 1. Health endpoint returns model info | |
| # --------------------------------------------------------------------------- | |
| class TestHealthIntegration: | |
| def test_health_returns_200_with_model_info(self, client): | |
| resp = client.get("/health") | |
| assert resp.status_code == 200 | |
| data = resp.json() | |
| assert data["model"] == "arcisvlm-1.6b" | |
| assert data["version"] == "1.0.0" | |
| assert "status" in data | |
| assert "model_loaded" in data | |
| assert "agents_ready" in data | |
| def test_health_status_is_valid(self, client): | |
| resp = client.get("/health") | |
| data = resp.json() | |
| assert data["status"] in ("ok", "starting") | |
| # --------------------------------------------------------------------------- | |
| # 2–5. POST /api/v1/query with various task_types (mocked inference) | |
| # --------------------------------------------------------------------------- | |
| class TestInferenceIntegration: | |
| """Tests POST /api/v1/query with different task_types, mocking the submit_task.""" | |
| def test_vqa_query_with_image(self, client, test_image_b64): | |
| """POST /api/v1/query with a test image gets a VQA answer.""" | |
| mock_result = _mock_result( | |
| answer="This is a red building", | |
| expert_used="vqa", | |
| confidence=0.92, | |
| ) | |
| with patch("api.routes.inference.get_model_manager") as mock_mgr: | |
| manager = MagicMock() | |
| manager.camera_manager = None | |
| manager.submit_task.return_value = mock_result | |
| mock_mgr.return_value = manager | |
| resp = client.post("/api/v1/query", json={ | |
| "question": "What is in this image?", | |
| "image_base64": test_image_b64, | |
| "task_type": "vqa", | |
| }) | |
| assert resp.status_code == 200 | |
| data = resp.json() | |
| assert data["answer"] == "This is a red building" | |
| assert data["expert_used"] == "vqa" | |
| assert data["confidence"] == 0.92 | |
| assert "processing_time_ms" in data | |
| def test_detect_task_type(self, client, test_image_b64): | |
| """POST /api/v1/query with task_type=detect gets detection response.""" | |
| mock_result = _mock_result( | |
| answer="Detected 3 objects: person, car, dog", | |
| expert_used="detect", | |
| confidence=0.88, | |
| metadata={"objects": ["person", "car", "dog"], "count": 3}, | |
| ) | |
| with patch("api.routes.inference.get_model_manager") as mock_mgr: | |
| manager = MagicMock() | |
| manager.camera_manager = None | |
| manager.submit_task.return_value = mock_result | |
| mock_mgr.return_value = manager | |
| resp = client.post("/api/v1/query", json={ | |
| "question": "Detect objects in the scene", | |
| "image_base64": test_image_b64, | |
| "task_type": "detect", | |
| }) | |
| assert resp.status_code == 200 | |
| data = resp.json() | |
| assert "detect" in data["answer"].lower() or "object" in data["answer"].lower() | |
| assert data["expert_used"] == "detect" | |
| def test_count_task_type(self, client, test_image_b64): | |
| """POST /api/v1/query with task_type=count gets counting response.""" | |
| mock_result = _mock_result( | |
| answer="There are 5 people in the image", | |
| expert_used="count", | |
| confidence=0.90, | |
| metadata={"count": 5}, | |
| ) | |
| with patch("api.routes.inference.get_model_manager") as mock_mgr: | |
| manager = MagicMock() | |
| manager.camera_manager = None | |
| manager.submit_task.return_value = mock_result | |
| mock_mgr.return_value = manager | |
| resp = client.post("/api/v1/query", json={ | |
| "question": "How many people are there?", | |
| "image_base64": test_image_b64, | |
| "task_type": "count", | |
| }) | |
| assert resp.status_code == 200 | |
| data = resp.json() | |
| assert data["expert_used"] == "count" | |
| assert data["confidence"] > 0 | |
| def test_caption_task_type(self, client, test_image_b64): | |
| """POST /api/v1/query with task_type=caption gets caption response.""" | |
| mock_result = _mock_result( | |
| answer="A busy city street with cars and pedestrians under a cloudy sky", | |
| expert_used="caption", | |
| confidence=0.87, | |
| ) | |
| with patch("api.routes.inference.get_model_manager") as mock_mgr: | |
| manager = MagicMock() | |
| manager.camera_manager = None | |
| manager.submit_task.return_value = mock_result | |
| mock_mgr.return_value = manager | |
| resp = client.post("/api/v1/query", json={ | |
| "question": "Describe this scene", | |
| "image_base64": test_image_b64, | |
| "task_type": "caption", | |
| }) | |
| assert resp.status_code == 200 | |
| data = resp.json() | |
| assert data["expert_used"] == "caption" | |
| assert len(data["answer"]) > 0 | |
| def test_query_without_image_returns_400(self, client): | |
| """POST /api/v1/query with no image source returns 400.""" | |
| resp = client.post("/api/v1/query", json={ | |
| "question": "What is this?", | |
| }) | |
| assert resp.status_code == 400 | |
| # --------------------------------------------------------------------------- | |
| # 6. Alert rule CRUD: create → list → delete | |
| # --------------------------------------------------------------------------- | |
| class TestAlertCRUDIntegration: | |
| """Full CRUD cycle for alert rules.""" | |
| def test_alert_rule_crud_lifecycle(self, client): | |
| """Create an alert rule, verify it in the list, then delete it.""" | |
| # Patch get_model_manager for alerts routes to have an orchestrator | |
| from agents.mother import MotherOrchestrator | |
| orchestrator = MotherOrchestrator(max_agents_per_type=2) | |
| with patch("api.routes.alerts.get_model_manager") as mock_mgr: | |
| manager = MagicMock() | |
| manager.orchestrator = orchestrator | |
| manager.alert_history = [] | |
| mock_mgr.return_value = manager | |
| # CREATE | |
| rule_payload = { | |
| "rule_id": "integ-test-fire-alert", | |
| "condition_type": "presence", | |
| "target_object": "fire", | |
| "action": "webhook", | |
| } | |
| resp = client.post("/api/v1/alerts/rules", json=rule_payload) | |
| assert resp.status_code == 200 | |
| data = resp.json() | |
| assert data["rule_id"] == "integ-test-fire-alert" | |
| assert data["condition_type"] == "presence" | |
| assert data["target_object"] == "fire" | |
| assert data["enabled"] is True | |
| # LIST — verify the rule appears | |
| resp = client.get("/api/v1/alerts/rules") | |
| assert resp.status_code == 200 | |
| rules = resp.json() | |
| assert isinstance(rules, list) | |
| rule_ids = [r["rule_id"] for r in rules] | |
| assert "integ-test-fire-alert" in rule_ids | |
| # DELETE | |
| resp = client.delete("/api/v1/alerts/rules/integ-test-fire-alert") | |
| assert resp.status_code == 200 | |
| assert resp.json()["deleted"] == "integ-test-fire-alert" | |
| # Verify deletion — rule should be gone | |
| resp = client.get("/api/v1/alerts/rules") | |
| assert resp.status_code == 200 | |
| rules_after = resp.json() | |
| rule_ids_after = [r["rule_id"] for r in rules_after] | |
| assert "integ-test-fire-alert" not in rule_ids_after | |
| def test_delete_nonexistent_rule_returns_404(self, client): | |
| """Deleting a rule that doesn't exist returns 404.""" | |
| from agents.mother import MotherOrchestrator | |
| orchestrator = MotherOrchestrator(max_agents_per_type=2) | |
| with patch("api.routes.alerts.get_model_manager") as mock_mgr: | |
| manager = MagicMock() | |
| manager.orchestrator = orchestrator | |
| mock_mgr.return_value = manager | |
| resp = client.delete("/api/v1/alerts/rules/nonexistent-rule") | |
| assert resp.status_code == 404 | |
| def test_alert_history_returns_list(self, client): | |
| """GET /api/v1/alerts/history returns a list.""" | |
| with patch("api.routes.alerts.get_model_manager") as mock_mgr: | |
| manager = MagicMock() | |
| manager.alert_history = [] | |
| mock_mgr.return_value = manager | |
| resp = client.get("/api/v1/alerts/history") | |
| assert resp.status_code == 200 | |
| assert isinstance(resp.json(), list) | |
| # --------------------------------------------------------------------------- | |
| # 7. Agent status endpoint returns all 8 agent types | |
| # --------------------------------------------------------------------------- | |
| class TestAgentStatusIntegration: | |
| def test_agent_types_returns_all_eight(self, client): | |
| """GET /api/v1/agents/types returns all 8 expert types.""" | |
| resp = client.get("/api/v1/agents/types") | |
| assert resp.status_code == 200 | |
| data = resp.json() | |
| expected_types = {"vqa", "detect", "alert", "caption", "track", "count", "ocr", "reason"} | |
| actual_types = set(data["expert_types"]) | |
| assert expected_types == actual_types | |
| def test_agents_status_endpoint(self, client): | |
| """GET /api/v1/agents/status returns proper structure.""" | |
| resp = client.get("/api/v1/agents/status") | |
| assert resp.status_code == 200 | |
| data = resp.json() | |
| assert "agents" in data | |
| assert "total_agents" in data | |
| assert isinstance(data["agents"], dict) | |
| assert isinstance(data["total_agents"], int) | |
| # --------------------------------------------------------------------------- | |
| # 8. Streams endpoint returns camera list | |
| # --------------------------------------------------------------------------- | |
| class TestStreamsIntegration: | |
| def test_streams_status_returns_list(self, client): | |
| """GET /api/v1/streams/status returns a list (empty when no cameras).""" | |
| resp = client.get("/api/v1/streams/status") | |
| assert resp.status_code == 200 | |
| data = resp.json() | |
| assert isinstance(data, list) | |
| def test_start_stream_requires_fields(self, client): | |
| """POST /api/v1/streams/start with missing fields returns 422.""" | |
| resp = client.post("/api/v1/streams/start", json={}) | |
| assert resp.status_code == 422 | |
| # --------------------------------------------------------------------------- | |
| # 9. WebSocket /ws/chat connects and can send/receive | |
| # --------------------------------------------------------------------------- | |
| class TestWebSocketIntegration: | |
| def test_ws_chat_connect_and_send(self, client): | |
| """WebSocket /ws/chat connects and responds to a question.""" | |
| mock_result = _mock_result( | |
| answer="The scene shows a parking lot", | |
| expert_used="vqa", | |
| confidence=0.80, | |
| ) | |
| with patch("api.deps.get_model_manager") as mock_mgr: | |
| manager = MagicMock() | |
| manager.camera_manager = None | |
| manager.submit_task.return_value = mock_result | |
| manager.model = None | |
| mock_mgr.return_value = manager | |
| with client.websocket_connect("/ws/chat") as ws: | |
| ws.send_text(json.dumps({ | |
| "question": "What is happening here?", | |
| "task_type": "vqa", | |
| })) | |
| resp = ws.receive_json() | |
| assert resp["type"] == "answer" | |
| assert resp["answer"] == "The scene shows a parking lot" | |
| assert resp["expert_used"] == "vqa" | |
| def test_ws_chat_missing_question(self, client): | |
| """WebSocket /ws/chat returns error when question is missing.""" | |
| with client.websocket_connect("/ws/chat") as ws: | |
| ws.send_text(json.dumps({"task_type": "vqa"})) | |
| resp = ws.receive_json() | |
| assert "error" in resp | |
| # --------------------------------------------------------------------------- | |
| # 10. Agent routing: MotherOrchestrator classify_task | |
| # --------------------------------------------------------------------------- | |
| class TestAgentRoutingIntegration: | |
| """Verify MotherOrchestrator routes queries to the correct agent type.""" | |
| def test_tracking_query_routes_to_track(self): | |
| """'track the person' routes to TrackingAgent.""" | |
| from agents.mother import classify_task | |
| from agents.base import EXPERT_TRACK | |
| result = classify_task("track the person moving across the room") | |
| assert result == EXPERT_TRACK | |
| def test_caption_query_routes_to_caption(self): | |
| """'describe this scene' routes to CaptionAgent.""" | |
| from agents.mother import classify_task | |
| from agents.base import EXPERT_CAPTION | |
| result = classify_task("describe this scene in detail") | |
| assert result == EXPERT_CAPTION | |
| def test_counting_query_routes_to_count(self): | |
| """'how many cars' routes to CountingAgent.""" | |
| from agents.mother import classify_task | |
| from agents.base import EXPERT_COUNT | |
| result = classify_task("how many cars are in the parking lot") | |
| assert result == EXPERT_COUNT | |
| def test_detection_query_routes_to_detect(self): | |
| """'detect objects' routes to DetectAgent.""" | |
| from agents.mother import classify_task | |
| from agents.base import EXPERT_DETECT | |
| result = classify_task("detect all objects in the frame") | |
| assert result == EXPERT_DETECT | |
| def test_alert_query_routes_to_alert(self): | |
| """'alert for intruder' routes to AlertAgent.""" | |
| from agents.mother import classify_task | |
| from agents.base import EXPERT_ALERT | |
| result = classify_task("alert me if there is an intruder") | |
| assert result == EXPERT_ALERT | |
| def test_ocr_query_routes_to_ocr(self): | |
| """'read the text' routes to OCRAgent.""" | |
| from agents.mother import classify_task | |
| from agents.base import EXPERT_OCR | |
| result = classify_task("read the text on the sign") | |
| assert result == EXPERT_OCR | |
| def test_reasoning_query_routes_to_reason(self): | |
| """'explain why' routes to ReasoningAgent.""" | |
| from agents.mother import classify_task | |
| from agents.base import EXPERT_REASON | |
| result = classify_task("explain why the car stopped") | |
| assert result == EXPERT_REASON | |
| def test_vqa_fallback(self): | |
| """Generic question falls back to VQA.""" | |
| from agents.mother import classify_task | |
| from agents.base import EXPERT_VQA | |
| result = classify_task("what color is the building") | |
| assert result == EXPERT_VQA | |
| def test_orchestrator_classify_method(self): | |
| """MotherOrchestrator.classify() uses the task query for routing.""" | |
| from agents.mother import MotherOrchestrator | |
| from agents.base import Task, EXPERT_TRACK | |
| mother = MotherOrchestrator(max_agents_per_type=2) | |
| task = Task( | |
| type="track", | |
| payload={"image_ref": "/tmp/test.jpg", "query": "track the person"}, | |
| source="test", | |
| expert_hint="vqa", # hint says vqa but query says track | |
| ) | |
| # classify should use the query text, not the hint | |
| expert = mother.classify(task) | |
| assert expert == EXPERT_TRACK | |
| mother.shutdown() | |