Bharat24B by Leo Labs
Copyright2026 Leo Labs for the adaptation and original release helper code.

Base: Mistral Small3.224B Instruct by Mistral AI, Apache License2.0.
Base weights are obtained separately from the upstream repository.
Changes: text attention/MLP LoRA adapters trained on5,000 synthetic math/code responses; optional Bharat serving identity policy; release metadata and inference helper.

Question source attribution and modifications are provided in DATA_ATTRIBUTION.md.