llm-semantic-router/Vela-1.0-Encoder-307M-Halu
Token Classification • 0.3B • Updated • 2
Adam Kovacs PRO
adaamko
AI & ML interests
Explainable AI, rule-based models, hallucination detection, fact-checking LLMs
Recent Activity
liked a model about 17 hours ago
llm-semantic-router/Vela-1.0-Encoder-307M-Halu upvoted a collection 5 days ago
⛵ Vela 1.0 upvoted an article 23 days ago
Training and Finetuning Multi-Vector Embedding Models with Sentence TransformersOrganizations
liked a
model about 17 hours ago
upvoted a collection 5 days ago
upvoted an article 23 days ago
Article
Training and Finetuning Multi-Vector Embedding Models with Sentence Transformers
tomaarsen
• • 134 upvoted a collection 24 days ago
reacted to tomaarsen's post with 🔥 30 days ago
Post
3933
🚨 I've just published Sentence Transformers v6.0, introducing MultiVectorEncoder: ColBERT-style late interaction models are now a fourth model type, for training, inference, and interpretation, alongside the dense, sparse, and reranker models! Details:
Where a regular embedding model compresses a whole text into one vector, a multi-vector model keeps one vector per token and scores query against document with the MaxSim operator. That preserves token-level matching information that a single vector has to average away. It is also the state of the art for visual document retrieval, where a text query is matched against page images directly, charts and tables included, with no OCR step in between.
Any PyLate, Stanford ColBERT, or ColPali checkpoint loads straight into the same familiar API: model.encode_query(), model.encode_document(), and model.similarity() just work, whether the documents are texts or page images.
Does it help? LightOn trained LateOn (multi-vector) and DenseOn (dense) on the same data with the same 149M ModernBERT backbone, and the multi-vector model wins on 9 of the 13 NanoBEIR datasets: 0.6868 vs 0.6764 mean NDCG@10. The price is a bigger index, and the new HierarchicalTokenPooling module halves it at roughly no retrieval cost.
Antoine Chaffin, Raphaël Sourty, and I wrote a blog post walking through multi-vector models in practice: loading the various checkpoint formats, encoding and scoring, plugging them into a search stack, running them on page images, and keeping the index affordable. Check it out if you want to get started, or just point your Agent to the URL: https://huggingface.co/blog/multi-vector-encoder
pip install sentence-transformers==6.0.0
Release notes: https://github.com/huggingface/sentence-transformers/releases/tag/v6.0.0
Where a regular embedding model compresses a whole text into one vector, a multi-vector model keeps one vector per token and scores query against document with the MaxSim operator. That preserves token-level matching information that a single vector has to average away. It is also the state of the art for visual document retrieval, where a text query is matched against page images directly, charts and tables included, with no OCR step in between.
Any PyLate, Stanford ColBERT, or ColPali checkpoint loads straight into the same familiar API: model.encode_query(), model.encode_document(), and model.similarity() just work, whether the documents are texts or page images.
Does it help? LightOn trained LateOn (multi-vector) and DenseOn (dense) on the same data with the same 149M ModernBERT backbone, and the multi-vector model wins on 9 of the 13 NanoBEIR datasets: 0.6868 vs 0.6764 mean NDCG@10. The price is a bigger index, and the new HierarchicalTokenPooling module halves it at roughly no retrieval cost.
Antoine Chaffin, Raphaël Sourty, and I wrote a blog post walking through multi-vector models in practice: loading the various checkpoint formats, encoding and scoring, plugging them into a search stack, running them on page images, and keeping the index affordable. Check it out if you want to get started, or just point your Agent to the URL: https://huggingface.co/blog/multi-vector-encoder
pip install sentence-transformers==6.0.0
Release notes: https://github.com/huggingface/sentence-transformers/releases/tag/v6.0.0
upvoted an article about 1 month ago
Article
Multi-Vector (Late Interaction) Embedding Models with Sentence Transformers
- +1
tomaarsen, NohTow, raphaelsty
• • 114 liked 2
models about 1 month ago
published a
model about 1 month ago
updated 4
models about 1 month ago
KRLabsOrg/LFM2.5-Encoder-350M-hallucination-detector
Token Classification • 0.4B • Updated • 33 • 1
KRLabsOrg/lettucedect-v2-qwen-2b
Text Generation • 2B • Updated • 1.11k • 4
KRLabsOrg/lettucedect-v2-taxonomy-head
0.3B • Updated • 94 • 2
KRLabsOrg/lettucedect-v2-mmbert-base
Token Classification • 0.3B • Updated • 4.91k • 4
upvoted an article about 1 month ago
Article
A 0.6B model at #2 on MTEB(Law)
Hanno-Labs
• • 5 liked a
model about 1 month ago
upvoted an article about 1 month ago
Article
Lattice: an 8 MB static retriever that embeds Wikipedia in 7 minutes
erikkaum
• • 19 updated a
dataset about 1 month ago
replied to their post about 1 month ago
the parquet (what load_dataset serves) is canonical — 2,015 rows at 50.3%, the verified split that annotations/REPORT.md describes. The root JSONLs are an earlier, pre-verification generation; they're now moved to legacy/ to avoid confustion. The usage snippet tag is fixed too. Thanks for noticing :)
posted an update about 1 month ago
Post
121
🥬 LettuceDetect v2 — span-level hallucination detection for code, tool output, and structured documents.
Hallucination detectors are trained on document QA, but agents ground their answers in source code, tool output and markdown. On code-agent answers, existing detectors reach 0.17 span-F1 and even 550B zero-shot judges at most 0.22.
We built a unified span-level benchmark — 74,285 newly constructed examples (145K+ with RAGTruth and 14-language PsiloQA folded in), every span typed and character-labeled — and trained two detectors on it:
🤖 KRLabsOrg/lettucedect-v2-qwen-2b — generative, typed spans + explanations in one pass, 32K context, **0.689 span-F1** (0.60 on code-agent)
⚡ KRLabsOrg/lettucedect-v2-mmbert-base — 307M multilingual encoder for high-throughput setups
🏷️ KRLabsOrg/lettucedect-v2-taxonomy-head — types the spans of any binary detector
It also reaches the best reported English PsiloQA IoU (0.724) and 81.8 RAGTruth example-F1, so specializing on code didn't cost general RAG performance.
📚 Dataset: KRLabsOrg/lettucedetect-code-hallucination
📄 Paper: https://arxiv.org/abs/2607.00895
The models are now integrated natively into vLLM Semantic Router — joint blog post on how it works: https://vllm-sr.ai/blog/lettucedetect-v2-generative-hallucination-detection
Hallucination detectors are trained on document QA, but agents ground their answers in source code, tool output and markdown. On code-agent answers, existing detectors reach 0.17 span-F1 and even 550B zero-shot judges at most 0.22.
We built a unified span-level benchmark — 74,285 newly constructed examples (145K+ with RAGTruth and 14-language PsiloQA folded in), every span typed and character-labeled — and trained two detectors on it:
🤖 KRLabsOrg/lettucedect-v2-qwen-2b — generative, typed spans + explanations in one pass, 32K context, **0.689 span-F1** (0.60 on code-agent)
⚡ KRLabsOrg/lettucedect-v2-mmbert-base — 307M multilingual encoder for high-throughput setups
🏷️ KRLabsOrg/lettucedect-v2-taxonomy-head — types the spans of any binary detector
It also reaches the best reported English PsiloQA IoU (0.724) and 81.8 RAGTruth example-F1, so specializing on code didn't cost general RAG performance.
📚 Dataset: KRLabsOrg/lettucedetect-code-hallucination
📄 Paper: https://arxiv.org/abs/2607.00895
The models are now integrated natively into vLLM Semantic Router — joint blog post on how it works: https://vllm-sr.ai/blog/lettucedetect-v2-generative-hallucination-detection
- 3 replies
updated a
collection about 1 month ago