vllm-sr/d3
Zero-Shot Classification • 26B • Updated • 2.75k • 181
AI & ML interests
An open, programmable decision layer for models and compute.
Recent Activity
Xunzhuo
updated 6
models about 6 hours ago
vllm-sr/d3-flash
Zero-Shot Classification • 8B • Updated • 2.66k • 62
vllm-sr/d3-mini
Zero-Shot Classification • 5B • Updated • 2.22k • 29
vllm-sr/d3-nano
Zero-Shot Classification • 2B • Updated • 1.97k • 32
vllm-sr/d3-lite
Zero-Shot Classification • 0.9B • Updated • 1.74k • 26
vllm-sr/d3-edge
Zero-Shot Classification • 0.6B • Updated • 1.83k • 37
Xunzhuo
updated a
Space about 6 hours ago
Xunzhuo
updated a
collection about 7 hours ago
Xunzhuo
updated 7
Spaces about 7 hours ago
d3-mini: Decision 3.0 live demo
🔮
1
Live demo of d3-mini on the free Decision API
d3-lite: Decision 3.0 live demo
🔮
Live demo of d3-lite on the free Decision API
d3-edge: Decision 3.0 live demo
🔮
Live demo of d3-edge on the free Decision API
d3: Decision 3.0 live demo
🔮
1
Live demo of d3 on the free Decision API
d3-flash: Decision 3.0 live demo
🔮
Live demo of d3-flash on the free Decision API
d3: Decision 3.0 live demo
🔮
3
Live demo of d3 on the free Decision API
Decision 3.0
🔮
Live demo of the d3 decision models: decision.vllm-sr.ai
Xunzhuo
published a
Space about 7 hours ago
The package is broken
3
#4 opened about 21 hours ago
by 
mazurkin
Post
138
🥬 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
adaamko
submitted a
paper to Daily Papers 3 months ago