# Farka model support

Catalog entries are options. Proof is earned per workflow.

Farka separates license review, hardware fit, live training, paired evaluation,
tuned inference, export, billing, and teardown. A model is not called verified
because it appeared in a list or because a quote was accepted.

## Verified configurations

Exact model, training method, passes over the data, accelerator-memory class,
and delivery combinations from collected production runs. Farka uses different
training approaches depending on the run: a light tune adds a small set of new
weights to the base model, while a full tune updates every weight. Provider-native
hardware remains attached to the audited run record. Results are configuration-specific. They do
not imply that every dataset, model size, or training recipe will improve.

| Model | Method | Result | Compute class |
| --- | --- | --- | --- |
| Qwen2.5 Coder 7B | Light tune, 1 pass | 15/30 base to 30/30 tuned | 16 GB accelerator |
| SmolLM2 360M | Full tune, 3 passes | 21/60 base to 58/60 tuned | 16 GB accelerator |
| Granite 4.1 8B | Light tune, 3 passes | Workflow verified | 16 GB accelerator |
| DeepSeek R1 Distill Qwen 7B | Light tune, 3 passes | Workflow verified | 16 GB accelerator |

"Workflow verified" means the full lifecycle completed for that configuration. It
is not a paired quality claim, which is stated separately where one exists.

## Current self-service boundary

Dense language models up to 9B parameters on the validated single-accelerator
path. Farka selects the cheapest verified capacity with sufficient memory.

## Enterprise capacity

Models over 9B, sparse and mixture-of-experts architectures, multi-accelerator plans,
private-network requirements, and customer-cloud workflows begin with a capacity
and security review. See <https://farka.ai/custom-training>.

## License review

Commercial use and weight modification must be supported by publisher evidence.
Unknown, non-commercial, or incompatible licenses fail closed. Catalog screening
supports, but does not replace, your own legal review.

## Contact

edi.hasaj@farka.ai