vllm-sr/vela3-turbo

Name it. Vela finds it.
_{vLLM Semantic Router × KR Labs · Vela 3.0 family · vela3-turbo · 27B · base d3-turbo · Apache-2.0}

What it does · Quick start · Evaluation · The family · 📡 Collection

---

vela3-turbo is the 27B member of the Vela 3.0 family, built on d3-turbo. In one system_one call it answers routing and general decisions and finds the exact spans you name (entities, relations, personal data, unsupported claims, evidence), with probabilities and character offsets and no generated text.

Summary

vela3-turbo is the reference model of the Vela 3.0 family, on d3-turbo v1.0 (pinned). d3-turbo, frozen, answers every ordinary decision, bit for bit, and gated adapters add the spans you name, entities, relations and router signals. It has the family's top scores on classification and relations.

Highlights

• Zero-shot classification: 81.4 on the 13-set GLiFormer classification suite, above every model the paper reports (GPT-5-mini: 79.8, Gemma-4-31B-IT: 79.69; reported, GLiFormer paper Tab 5).
• Relations: 35.9 end to end (GLiFormer Table 4 protocol, avg of 4), above every model reported (best: Qwen3.5-9B, prompted, 25.7; reported, GLiFormer paper Tab 4); 51.7 on CoNLL04 with the head entity given (GLiFormer-base: 8.6, our run, same harness).
• Zero-shot NER: 64.6 on OOD-7 (CrossNER + MIT) from label names, above every GLiNER / GLiFormer number reported on it (best: gliner-multitask-large-v0.5, 62.76).
• Unsupported claims: 0.900 example-F1 on the 10,698-answer LettuceDetect v2 test set, above LettuceDetect v2 mmBERT-base: 0.869 (reported, its card) and Vela 2.0 9B: 0.885 (its card).
• Router signals: better than Vela 2.0 9B on 20 of the 24 test files of the public router signal suite, level on 4, worse on none (paired 95% CI).
• Decisions, bit for bit: every ordinary decision is answered by d3-turbo@da17b5d0 itself, adapters off (561 of 561 identical in our check).

_{vela3-turbo (27B) against reported numbers. Accent = vela3-turbo; grey = other models (reported).}

At a glance

What it does

Quick start

# pip install "transformers==5.17.0" torch torchvision pillow opencv-python-headless safetensors accelerate
from transformers import AutoModel

m = AutoModel.from_pretrained("vllm-sr/vela3-turbo", trust_remote_code=True)  # downloads d3-turbo at the pinned revision

# Name any labels; the spans come back with offsets and scores.
r = m.system_one(
    state="Geoffrey Hinton presented capsule networks at NeurIPS in Montreal.",
    questions={"ents": {"type": "span", "criteria": ["researcher", "algorithm", "conference", "location"]}})
# researcher "Geoffrey Hinton", algorithm "capsule networks", conference "NeurIPS", location "Montreal"

# Unsupported claims in an answer, against a context or an image.
r = m.system_one(
    state={"context": "The Eiffel Tower is 330 metres tall and was completed in 1889.",
           "answer": "The Eiffel Tower is 330 metres tall and was completed in 1925 by Gustave Eiffel's son."},
    questions={"halu": {"type": "span", "criteria": {"unsupported": None}}})
# flags "1925 by Gustave Eiffel's son"; add images=["photo.jpg"] to check an answer against a photo

# PII, relations and evidence use the same call.
r = m.system_one(
    state="Steve Jobs founded Apple in Cupertino together with Steve Wozniak.",
    questions={"pii": {"type": "span", "criteria": ["PERSON"]},
               "rel": {"type": "span", "instructions": "Which spans are the founder of Apple?", "criteria": {"founder": None}}})

Evaluation

All numbers on public test sets, measured through this package with system_one, one request at a time and the shipped thresholds. Other models' numbers are as reported by their authors (or scored in the same harness where marked). Full per-set tables: EVALUATION.md.

Named entities

Label names are the only input. Training holds no CrossNER or MIT text, though many generic CrossNER type names occur as training labels.

Relations

CrossRE 39.2 (in-domain) is above every model reported in the GLiFormer paper (best: GLiNER-Relex 18.1).

Evidence

ACL-Verbatim is in-domain: its silver training split is in the training data.

Unsupported claims (hallucination spans) Zero-shot classification

These questions run on d3-turbo with every adapter off.

Router signals (public router signal suite; above Vela 2.0 9B on 7 of 8)

vllm-sr/router-signal-suite (public), the test file of each signal, scored with the dataset's own scoring code. Each signal is a router question this package was trained on (supervised, as in Vela 2.0), asked one request at a time with the shipped thresholds.

Vela 2.0 9B was run by us through its public release on the same rows. Decision-1.0-Route-0.6B is scored from the suite's published predictions; it was trained on the suite's own train split, so these test files are in-distribution for it.

Bold: the paired 95% bootstrap CI against Vela 2.0 excludes a tie. The last column drops the test rows that share text (exact or a 13-gram) with this model's training data. Safety and jailbreak are scored on the rows whose text the suite publishes (2,230 of 3,003 and 1,878 of 3,000).

Domain and feedback are also trained on the suite's own train split (with the public router data Vela 2.0 used; the suite's test and hold rows, and anything sharing an exact text or a 13-gram with them, are kept out), as Decision-1.0-Route is, so their test files are in-distribution for both. The domain question has a 15th option, religion, which the suite maps to philosophy. Feedback has NO_FEEDBACK as a trained option.

Decisions (general decisions are the base model's, bit for bit)

General decisions are the base model's own answers, bit for bit, so a board entry of the base would be this package's decision score. Full score (the board's ranking): 60% private same-skill tests and 40% private new-domain tasks; public index: the 37 public benchmarks, chance-corrected. All values are board entries on the Decision Index 0.4.1 board (multimodalart/jev-decision-index, data generated 2026-10-28T09:02Z) unless noted; the rank is the position in the board's Full-score order, and tied rank marks entries the board shows as tied. d3-turbo v1.0, this package's base (revision da17b5d0), is not on the Decision Index 0.4.1 board; the public d3-turbo main is v1.1, which is on the board at 67.31 (rank 15, tied).

MASSIVE: three scenarios never used in our router training (iot, transport, cooking), 52 locales; accuracy on the 14,911 test utterances that match no training text of the model's lineage. The 342 excluded utterances occur verbatim in upstream training mixtures (tasksource MASSIVE / HWU64 rows). Vela 2.0 values are on the full 15,253, so these rows are not bolded. Super-NI 1568: 79 to 90 of its 162 ProPara passages share a 13-gram with other tasks' training rows.

Latency

Span questions with label names only.

Benchmarks

• OOD-7 (CrossNER AI, literature, music, politics, science; MIT movie, restaurant): find the entities of the named types in text from domains the model was not trained on (strict entity F1).
• NER, 26 datasets of the GLiFormer paper (IE_INSTRUCTIONS): entities of named types across 26 public NER sets, not all zero-shot (strict entity F1).
• CoNLL04, head entity given: given one entity of a sentence, find the entities it is related to by the named relation (F1 over those tail entities).
• Relations end to end, GLiFormer Table 4 protocol (CoNLL04, DocRED, FewRel, CrossRE): find the entity pairs and their relation from raw text (strict triplet F1).
• QASPER: find the sentences of a research paper that answer a question about it (word-F1).
• ACL-Verbatim (KRLabsOrg/acl-verbatim-spans): find the passage of an ACL paper that supports an answer (word-F1).
• LettuceDetect v2 test set (10,698 answers, as in the LettuceDetect v2 paper (GroundLM 2026 workshop at EMNLP 2026) and card): does an answer contain claims its context does not support (example-F1: the answer is flagged or not).
• RAGTruth: does a retrieval-augmented answer contain claims its retrieved context does not support (example-F1).
• MASSIVE, three unseen scenarios (iot, transport, cooking; 52 languages): pick the intent and the scenario of a user command the model never saw in training (accuracy).
• Super-NI task 1568 (ProPara): answer a multiple-choice question about how a process changes an entity (accuracy).
• Zero-shot classification, 13 sets (the GLiFormer paper's suite: SST-2, IMDB, AG News, Banking77, and more): pick the label of a text from label names only (mean macro-F1).
• Decision Index 0.4.1 (board, 2026-10-28): 37 public decision benchmarks plus private tests; the Full score ranks on the private same-skill and new-domain tests (chance-corrected skill, 0 = guessing).
• Router signal suite (vllm-sr/router-signal-suite): the vLLM Semantic Router's public test of its request signals (harmful requests, jailbreaks, personal data, unsupported answers, fact-check need, modality, domain, feedback); each label comes from public corpora (AUC or accuracy).

The family

How it works

• Frozen base. d3-turbo@da17b5d0 runs unchanged; ordinary questions are its own answers.
• Gated LoRA adapters. Each switches on only for its own questions, and every answer names the one it used (head_reason): decision (router signals), NER (entity labels with definitions), relation, image (hallucination spans with images), span (evidence, PII, unsupported claims, toxic spans, names-only labels).
• Spans as decisions. The labels are presented as options; a span head scores every word against every label, and agreeing neighbours form a span with character offsets.
• Label names in. Labels are read by name at this size; no definitions are written.
• Specification. Adds 5044 MB of adapters and heads; text or JSON state; requests with images, videos, assets, box or track questions, or long_context go to the base with every adapter off. Full evaluation: EVALUATION.md.

Trade-offs

• Unsupported claims: hallucination spans 0.900 against 0.921 for LettuceDetect v2 qwen-2b, RAGTruth 0.779 against 0.850 for HDM-2 (3B) (reported).
• NER from label names: 64.6 on OOD-7, above the best reported encoder (gliner-multitask-v1.0 63.25) and below B2NER-20B, standardised label names 72.1 (reported).
• End-to-end relations on generic types: 35.9 average on the four typed sets, above every model in GLiFormer Table 4 (reported), but 0.2 on SciERC and 0.5 on NYT11, where every entity has the generic type entity.

Citation

@misc{vela3_turbo,
  title        = {vela3-turbo (Vela 3.0, 27B): zero-shot extraction and router signals on Decision 3.0},
  author       = {{KR Labs} and {vLLM Semantic Router}},
  year         = {2026},
  howpublished = {\url{https://huggingface.co/vllm-sr/vela3-turbo}}
}