--- license: apache-2.0 tags: - unsloth language: - en base_model: - unsloth/Meta-Llama-3.1-8B pipeline_tag: text-generation datasets: - Varadrajan/informal_formal_tweet --- # Uploaded adapter - **Developed by:** Varadrajan - **License:** apache-2.0 - **Finetuned from model :** unsloth/Meta-Llama-3.1-8B This llama adapter was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library. [](https://github.com/unslothai/unsloth) ## πŸ“– Overview / Purpose This adapter enables transforming **informal / casual** English sentences into **formal / polished** prose. It’s ideal for tone standardization, content polishing, and elevating everyday speech into refined writing. --- ## πŸš€ How to Use You have two main options: - **On-the-fly adapter usage**: Load base model + adapter during inference (keeps highest fidelity). - **Merged model inference**: Merge the adapter into a model (16-bit or 4-bit) so inference needs only one model artifact (simpler deployment). --- ## πŸ§ͺ Merging Options & Tradeoffs - **16-bit merge** β€” preferred for better output quality while simplifying inference - **4-bit merge (experimental)** β€” smaller footprint but may degrade quality due to quantization / rounding errors Use the method that fits your resource constraints and quality requirements. --- ## πŸ“Œ Intended Applications & Use Cases - Customer support / chat interfaces, rewriting user text more professionally - Content polishing tools, style editors - Internal documentation, standardization of tone across teams - Writers / non-native English users seeking more formal expression --- ## ⚠️ Limitations & Risks - The adapter might over-formalize or introduce vocabulary changes - In edge cases (slang, idioms, sarcasm), output may misinterpret meaning - Merged low-bit models can degrade quality - Always review output in sensitive or public contexts ---