Carbon-X is a biology-focused model released by HuggingFaceBio that can be run fully on your own device using the Strata engine inside Stratat. It targets researchers who need to process biochemistry data while keeping it on their own hardware, which makes carbon-aware inference practical without sending inputs to remote servers.
Why Carbon-X exists and how it helps
Most large language and biology models are hosted only in cloud services, which means your data leaves your machine. Carbon-X was released to let users run such models locally so that input and output stay on the same device. This reduces the carbon footprint of inference because you only pay for the electricity your computer uses, and no data travels across data centers.
How to run Carbon-X on your machine
The Strata client installs the model and the Strata engine together, pulling only the weights that are required for inference. You need a laptop or desktop with a modern GPU or CPU that supports the ONNX runtime used by Carbon-X. A typical requirement is a processor with AVX2 support and at least 8 GB of RAM, plus a GPU that has its own memory bandwidth for the matrix math needed in biology models.
Measuring performance yourself
You can verify that the model runs on your own machine by timing a single inference pass and noting the power draw. Use the system monitor that comes with your OS to read instantaneous kWh usage during a short test, then multiply that reading by the elapsed seconds to get the energy used in joules (1 kWh = 3.6 MJ). If you know your machine's rated wattage, you can also convert watts directly to kilowatt‑hours.
| Component | Requirement | Notes |
|---|---|---|
| GPU | Any recent discrete GPU or integrated GPU with OpenCL/ CUDA support | Required for fast matrix ops |
| System RAM | Minimum 8 GB | More is better for large model buffers |
| Inference time (single pass) | Varies per machine | Run a small biology token sequence and note the time |
| Power (kW) | Approx. 100 W idle, 150 W under load for typical laptop | Check your device spec sheet for exact draw |
One sentence definition: Carbon-X is a locally runnable biology model that performs generation tasks on your own device without sending data to the cloud.
Checking the measurements on your own machine
Run the following steps on your own computer to confirm that the model is working locally and to record its energy use. First, open a terminal and start the Stratat launcher pointing at the installed Carbon-X model file. Then, open a simple script that calls the model once with a fixed input and prints the start and end timestamps. After the run, look at the OS power meter that shows instantaneous kW while the inference was active. Multiply the average kW by the elapsed seconds to obtain kWh, and divide by the number of kilowatt‑hours reported by your system’s power supply to see the real energy cost.
What the table tells you
The table lists the minimum hardware you must have, the kind of power draw you can expect, and the steps you need to reproduce the results yourself.
FAQ
Q: Is Carbon-X only for research labs? A: No, anyone with a capable PC can run it on Stratat, and it is useful for any user who wants private, on‑device biology generation.
Q: Where can I find the model files? A: The model is published on HuggingFace under the name HuggingFaceBio/Carbon-X, and the Stratat client will detect it when you launch the app.
Q: Does the model really lower my carbon impact? A: Because the inference stays on your own hardware, the only emissions come from the electricity your machine consumes while running, which is usually far less than the emissions of a cloud hosting a similar model.