ahammad115566/qwen-smeft
ahammad115566/qwen-smeft is a domain-adapted large language model fine-tuned from Qwen3-8B by Ahmed Hammad and Veronica Sanz. Optimized for Standard Model Effective Field Theory (SMEFT), it excels at SMEFT operator reasoning, EFT basis translation, and physics-aware scientific dialogue. This model is specifically designed for structured theoretical question answering within the particle physics domain.
Loading preview...
Overview
This model, ahammad115566/qwen-smeft, is a domain-adapted large language model (LLM) specifically fine-tuned for Standard Model Effective Field Theory (SMEFT). Developed by Ahmed Hammad and Veronica Sanz, it is based on the Qwen3-8B architecture and was trained using LoRA on a curated corpus of SMEFT and high-energy physics preprints. The model is designed for instruction-following scientific question answering within its specialized domain.
Key Capabilities
- SMEFT operator reasoning: Understands and processes complex SMEFT operators.
- EFT basis translation: Facilitates translation between different Effective Field Theory bases.
- Physics-aware scientific dialogue: Engages in technical discussions with an understanding of particle physics concepts.
- Literature-style technical explanation: Generates explanations in a style consistent with scientific literature.
- Structured theoretical question answering: Provides precise answers to theoretical questions within the SMEFT domain.
Important Considerations
- Domain-locked: This model is intentionally specialized and not suitable for general-purpose tasks.
- Potential for hallucination: May occasionally hallucinate operator identities.
- Uneven coverage: Due to 3620 training examples, coverage of the SMEFT operator space might be uneven, potentially leading to less reliable answers for rare operators or non-Warsaw bases.
Usage Note
When using tokenizer.apply_chat_template(...) with this model, ensure enable_thinking=False to prevent conflicts with the model's custom structured prompt format for reasoning, as it was not tuned with Qwen3's native thinking blocks.