Hallucination evaluation
According to the model card, Lynx judges whether a RAG answer stays faithful to its document without adding or contradicting context.
Open Source Model Profile · PatronusAI
Llama-3-Patronus-Lynx-8B-Instruct
Llama-3-Patronus-Lynx-8B-Instruct is an 8.03B-parameter Llama hallucination-detection fine-tune from PatronusAI. According to the model card, it evaluates RAG answer faithfulness under CC-BY-NC-4.0 terms.
- Publisher
- PatronusAI
- Task
- text-generation
- Model type
- llama
- License
- cc-by-nc-4.0
- Library
- transformers
- Publication status
- Approved for indexing
Model overview
Llama-3-Patronus-Lynx-8B-Instruct is published by PatronusAI as a text-generation model. The captured configuration identifies LlamaForCausalLM with model type llama, and Safetensors metadata reports 8,030,261,248 parameters. According to the model card, it fine-tunes meta-llama/Meta-Llama-3-8B-Instruct into the Lynx hallucination evaluation model for RAG faithfulness checking in primarily English. The recorded license is cc-by-nc-4.0.
Recorded capabilities
Mixed QA training data
According to the model card, training mixed CovidQA, PubmedQA, DROP, and RAGTruth sources with hand-annotated and synthetic data.
8000-token sequences
According to the model card, the maximum sequence length is 8000 tokens.
PASS FAIL chat pattern
According to the model card, inference uses chat-formatted prompts requesting reasoning plus a strict PASS or FAIL verdict.
Use cases in the source record
- RAG hallucination checks where a document, question, and answer are assessed for faithfulness with a PASS or FAIL verdict.
- Chat-formatted evaluation pipelines that send the documented prompt as a user message with up to 600 new tokens per judgment.
Limitations and unknowns
- No evaluation results were extracted from this record.
- No context-window value beyond the card's 8000-token sequence statement was extracted.
- According to the model card, full training details are referred to the linked ArXiv paper rather than stated in the card.
- Provider state is historical snapshot data and should be refreshed before being presented as current.
Source and provenance
Source: PatronusAI/Llama-3-Patronus-Lynx-8B-Instruct
Captured: Unknown. Processed: 2026-09-07T19:34:35.402767+00:00.
Model Card for Model ID Lynx is an open-source hallucination evaluation model. Patronus-Lynx-8B-Instruct was trained on a mix of datasets including CovidQA, PubmedQA, DROP, RAGTruth. The datasets contain a mix of hand-annotated and synthetic data. The maximum sequence length is 8000 tokens. Model Details Model Type: Patronus-Lynx-8B-Instruct is a fine-tuned version of meta-llama/Meta-Llama-3-8B-Instruct model. Language: Primarily English Developed by: Patronus AI Paper: https://arxiv.org/abs/2407.08488 License: https://creativecommons.org/licenses/by-nc/4.0/ Model Sources Repository: https://github.com/patronus-ai/Lynx-hallucination…
F001F002F003F004F005F006F007F008F010F011F012F013F014F015F016F017F018F019