| --- |
| language: en |
| license: apache-2.0 |
| tags: |
| - slm |
| - rag |
| - hallucination-guard |
| - logits-processor |
| - pytorch |
| - transformers |
| --- |
| |
| # 🛡️ De-Hallucinator |
|
|
| An inline token-probability uncertainty guard and semantic fact-checking engine for Small Language Models (SLMs). |
|
|
| `De-Hallucinator` extends the Hugging Face `LogitsProcessor` pipeline to intercept text generation token-by-token. The moment an SLM drops an uncertain token below a configured logprob threshold, the generation sequence halts instantly, triggers a quantized NLI cross-encoder factual pass against your grounding context, and forces an immediate End-of-Sentence (EOS) cutoff if the assertion fails. |
|
|
| ## 🚀 Installation |
|
|
| You can install the compiled wheel asset directly from this Hugging Face repository: |
|
|
| ```bash |
| pip install [https://huggingface.co/YOUR_HF_USERNAME/YOUR_REPO_NAME/resolve/main/de_hallucinator-0.1.0-py3-none-any.whl](https://huggingface.co/YOUR_HF_USERNAME/YOUR_REPO_NAME/resolve/main/de_hallucinator-0.1.0-py3-none-any.whl) |
