Downloads · 30 days
0
lettucedetect-base-modernbert-onnx
maddyvicky001/lettucedetect-base-modernbert-onnx
lettucedetect-base-modernbert-onnx is a token classification model from maddyvicky001. Use it when you need labels on individual words, such as names. It is set up for onnx. The card lists the license as mit.
ONNX export of KRLabsOrg/lettucedect-base-modernbert-en-v1 (MIT, by KR Labs) for running hallucination detection without PyTorch, e.g. in serverless functions. Used by GroundCheck.
Access
Public
Updated Sep 24, 2026
Repo size
300 MB
Likes
0
Public
Repo makeup
Click a slice to open those files.
.onnx300 MB · 99%
At a glance
- Task
- Token Classification
- Library
- onnx
- License
- mit
- Access
- Public
- Created
- Sep 24, 2026
- Updated
- Sep 24, 2026
- SHA
- 24744c04
Try a prompt
Base models
Model card
From the Hugging Face model README
LettuceDetect base (ModernBERT) — ONNX
ONNX export of KRLabsOrg/lettucedect-base-modernbert-en-v1 (MIT, by KR Labs) for running hallucination detection without PyTorch, e.g. in serverless functions. Used by GroundCheck.
- Weights stored as fp16, computation in fp32 (each weight is cast to fp32 at load). 300 MB instead of 600 MB.
- Accuracy vs. the original PyTorch model: max absolute difference in P(hallucinated) of 0.0013 at 43, 691 and 4,066 tokens. (8-bit dynamic quantization was tested and rejected: errors up to 0.95.)
- Inputs:
input_ids,attention_mask(int64,[batch, seq], up to 4,096 tokens). Output:logits[batch, seq, 2]; class 1 = hallucinated. Input format is LettuceDetect's:[CLS] prompt [SEP] answer [SEP].
All credit for the model goes to the LettuceDetect authors — see the paper and code.
- Task
- Token Classification
- Library
- onnx
- License
- mit
- Languages
- en
- Created
- Sep 24, 2026
- Updated
- Sep 24, 2026