| --- |
| title: AfroBR-LangBench Tiny Eval |
| emoji: 🌍 |
| colorFrom: blue |
| colorTo: indigo |
| sdk: gradio |
| sdk_version: "4.44.1" |
| python_version: "3.10" |
| app_file: app.py |
| pinned: false |
| --- |
| |
| # AfroBR-LangBench Tiny · Model Comparison |
|
|
| Base `meta-llama/Llama-3.2-3B-Instruct` vs the AfroBR-LangBench **Tiny AutoScientist** LoRA |
| (`pt_afro_brasileiro_qa`, r=16, alpha=32, 1 epoch, all-linear). |
|
|
| Same blind A/B evaluation harness as |
| [afrobr-langbench-eval](https://huggingface.co/spaces/Fernandosr85/afrobr-langbench-eval) |
| (7-dimension sociolinguistic safety rubric, judged blind by Claude), run on the |
| same 20-example held-out set, for a direct 2x2 comparison against the 109B |
| Llama-4-Scout run. |
|
|
| Runs on **ZeroGPU**. |
|
|
| Check out the configuration reference at https://huggingface.co/docs/hub/spaces-config-reference |
