# d1-2B RAG Verifier decision model by landtml

> d1-2B RAG Verifier checks whether an answer is supported by its source. Read its RAGTruth results, threshold selection and comparison limits.

Updated 2026-11-12 (Australia/Adelaide)

## Quick answers

### What is d1-2B RAG Verifier by landtml?

[d1-2B RAG Verifier](<https://decisionmodels.ai/models/d1-rag-verifier>): A d1-2B LoRA for checking whether a supplied answer is supported by its source.

[d1-2B RAG Verifier facts](<https://decisionmodels.ai/models/d1-rag-verifier>)

[Source](<https://huggingface.co/landtml/d1-2B-RAG-Verifier>)

### Who makes d1-2B RAG Verifier?

[d1-2B RAG Verifier](<https://decisionmodels.ai/models/d1-rag-verifier>): landtml

[d1-2B RAG Verifier facts](<https://decisionmodels.ai/models/d1-rag-verifier>)

[Source](<https://huggingface.co/landtml/d1-2B-RAG-Verifier>)

### Are d1-2B RAG Verifier's weights open?

[d1-2B RAG Verifier](<https://decisionmodels.ai/models/d1-rag-verifier>): Merged BF16 weights, LoRA adapter, Q8\_0 GGUF and scoring records Licence: LFM Open License v1.0

[d1-2B RAG Verifier facts](<https://decisionmodels.ai/models/d1-rag-verifier>)

[Source](<https://huggingface.co/landtml/d1-2B-RAG-Verifier>)

By landtml · Weights upload: 2026-11-08 · OPEN

A d1-2B LoRA for checking whether a supplied answer is supported by its source. It returns one response-level support probability, without identifying unsupported spans. Use the recorded source/request/answer fields and the exact question from the card.

## Facts

- Size: 1.9B
- Inputs: English source, request and generated answer
- Licence: LFM Open License v1.0
- Access: Merged BF16 weights, LoRA adapter, Q8\_0 GGUF and scoring records
- Evaluation scope: RAGTruth; training sources capped at 12,000 characters, with longer inputs unevaluated

## Claims and sources

- **VENDOR-REPORTED**: The maker reports F1 81.5, AUROC 0.936 and ECE 0.030 on 2,700 RAGTruth responses. Checkpoint and threshold 0.56 were selected on document-grouped development data. The base comparison reports F1 70.9 but uses Q8\_0 llama.cpp while the fine-tune uses BF16. Base-training overlap is unknown; third-party table rows use their own papers and code. [source](<https://huggingface.co/landtml/d1-2B-RAG-Verifier>)

## Links

- [Source: Card, weights and evaluation](<https://huggingface.co/landtml/d1-2B-RAG-Verifier>)
- [Source: Scoring code](<https://huggingface.co/landtml/d1-2B-RAG-Verifier/blob/main/eval/evaluate.py>)
- [Source: Licence](<https://huggingface.co/landtml/d1-2B-RAG-Verifier/blob/main/LICENSE>)

## Related projects

No related projects are listed in the registry.