General AI models are good at a bit of everything. Farka trains an open model on examples of your task, shows how much it improved on examples it never saw, and gives you the model to download and keep.
New workspaces start with $10 of credit while launch seats remain. No card needed for the first run.
Cell 2 log: axis 3 torque spiked to 118 percent for 400 ms.joint_overtorque
Example rows from robot-faults-eval.jsonl, each resolving to one supported fault label.An example: free-text logs from factory equipment go in, and the trained model answers with exactly one of four fault labels, a fixed string a downstream system can act on.joint_overtorquevision_dropoutgripper_slipsafety_stop
A specialist learns your task from your own examples, so it answers in exactly the format you need, can run on smaller hardware, and belongs to you.
Upload examples of the task and the answers you want, chat files made for OpenAI or Foundry, or your manuals. Farka checks format, quality and duplicates before training.
Pick a supported model and review the quote. Train from labeled examples, or define outcomes that can be checked automatically.
Test the result on examples kept out of training before delivery. Download the model files or a package ready to run on your own infrastructure.
Smaller models are fast and cheap to run. Larger ones handle harder reasoning, images and long documents. Farka checks which hardware fits your training setup before quoting the run.
80 GB accelerator memory class. Fit depends on the full training configuration and available capacity.
Reasoning, code and agents
Images, text and assistants
Many languages and longer documents
16 GB accelerator memory class. Fit depends on the full training configuration and available capacity.
Before / after: 25/30 → 30/30
Before / after: 15/30 → 30/30
Before / after: 21/60 → 58/60
Larger models are catalog options, not measured results. The before-and-after scores come from recorded internal tests. Every new run must pass its own quality checks.
See all models and boundariesCompare before and after on the same examples kept out of training. These are internal classification tests, not a guarantee of results on your data.
Correct answers before and after training
Light tune · 1 pass · 16 GB accelerator
270 training examples and 30 test examples. Internal classification task.
Base and trained model tested on the same 30 examples kept out of training.
Read the evidenceCorrect answers before and after training
Full tune · 3 passes · 16 GB accelerator
600 training, 60 development and 60 blind test examples. Internal classification task.
Base and trained model tested on the same 60 examples kept out of training.
Read the evidenceYour quote covers training, data storage, quality checks and your chosen delivery option. Credits are reserved only when training starts.
See how your quote is built Create workspace to estimatePin a training run to the EU or the US and it runs nowhere else. A run that cannot be placed in-region fails rather than moving.
Download standard safetensors files that load in common open-weight tools, or a package ready to run on infrastructure you control.
Read the docsBring a larger model, unusual objective, private infrastructure boundary, or tailored deployment requirement. We will scope a custom training and evaluation path around the outcome your company needs.
Create a workspace, upload a dataset, and quote your first run before any GPU starts.
Create your workspace