as machine learning models grow, data centers expand to keep up — but bigger isn’t always better.
the world's smallest data center
in the heart of san francisco, you can find the world's smallest data center.
our in-house localized data center reduces latency, enabling real-time processing and direct control. we're proving that the right setup even at a small scale can achieve powerful AI performance without a massive footprint.
about lil data center
the world's smallest data center was started to power Roboflow's testing and benchmarking infrastructure.
while small, our data center runs over 1,000 test jobs per week for open source projects.
lil data center runs:
Device Type:
Nvidia Jetson Orin Nano (Seeed)
Status:
Online
Running:
Benchmark Tests
Device Type:
Raspberry Pi 5
Status:
Online
Running:
Benchmark Tests
Device Type:
Mac Mini M4
Status:
Online
Running:
Benchmark Tests
Device Type:
Signed Jensen RTX 4080
Status:
Online
Running:
Benchmarking Tests
recent tests
these pull requests are for to open source inference repository, which runs tests and benchmarks via the lil data center. this includes benchmarking SOTA models like PaliGemma, Florence-2, OwlV2, and many more. see all supported models
Add Mage-VL codec-native video VLM to inference_models
Checks
CodeQL⚠️
torch_models:3.10⚠️
torch_models:3.12⚠️
torch_models:3.11⚠️
onnx_extras:3.11⚠️
onnx_extras:3.12
onnx_extras:3.10
build-dev-test (3.10, true)⚠️
build-dev-test (3.10, false)⚠️
build-dev-test (3.11, true)⚠️
build-dev-test (3.11, false)⚠️
build-dev-test (3.12, false)⚠️
build-dev-test (3.12, true)⚠️
Socket Security: Pull Request Alerts
build-dev-test (3.10)
build-dev-test (3.11)
build-dev-test (3.12)
build-dev-test (3.12)
build-dev-test (3.10)
build-dev-test (3.11)
build-dev-test (3.11)
build-dev-test (3.12)
build-dev-test (3.10)
call_is_mergeable / Check if branch is mergeable
build-dev-test (3.11)⚠️
build-dev-test (3.12)⚠️
build-dev-test (3.10)⚠️
integration-tests (3.11)
e2e-tests (3.12)
integration-tests (3.10)
Record model variant on usage-tracking rows
Triton kernel runtime error fallback
Add an optional S3-compatible cache layer for model files
Add visual prompts to SAM3 Video Tracker
models benchmarked @ lil data center
YOLOv11
CLIP
Owlv2
YOLO-NAS
LCS2S-Net
DocTR
YOLOv10
SAM2
Grounding DINO
YOLOv9
TrOCR
SAM
PaliGemma
YOLOv8
Florence-2
YOLOv7
YOLO-World
Request a Model
located at
9 AM - 5 PM
Monday through Friday
Tours available on request
166 Geary Street
San Francisco, CA 94108
United States