Von

Name: Von
Author: wfzyx

wfzyxOpen weights

Jev's API over an NLI classifier: every option becomes a hypothesis and the distribution is a softmax over entailment logits. Its config matches tasksource/ModernBERT-large-nli entry for entry.

The PyPI package `von` is an unrelated stub; install from the GitHub repository. Its shipped temperature is only read from a local path, so it runs at T = 1.0 from the Hub.

Hugging Face GitHub

typed-decisions: 0.461
Metatext suite: 0.509
ECE untuned: 0.249
p50 per case: 597 ms

Specifications

Checkpoint: wfzyx/von-1.0
Publisher: wfzyx
Mechanism: NLI entailment head, one hypothesis per option
Backbone: ModernBERT-large NLI head
Parameters: 395M
License: Apache 2.0
Runs as: open weights, self-hosted
Library: von (GitHub)
Primitives: Choice, Score, Noul
Pricing: none — your compute

Metatext decisionsrank 14 of 21

Full leaderboard →

Per-question accuracy — default payload, systemone mode, 120 judgments

Task	Question	Type	Accuracy	Baseline	Beats?
agent_traces	outcome	choice	0.500	0.357	✅
agent_traces	needs_human	noul	0.286	0.714	❌
customer_service	department	choice	0.650	0.400	✅
customer_service	refund_requested	noul	0.950	0.700	✅
customer_service	frustration	score	0.300	0.500	❌
security_incidents	severity	choice	0.562	0.312	✅
security_incidents	actively_exploited	noul	0.438	0.688	❌

Fitted temperature per primitive (fold 0, judged on fold 1): choice 0.4, score 4, noul 5 — ECE 0.146 → 0.122, accuracy → 0.500.

typed-decisionsrank 4 of 9

Full leaderboard →

Per-question accuracy — default payload, systemone mode, 2,000 judgments

Task	Question	Type	Accuracy	Baseline	Beats?
td_agent_trace_observability	action	choice	0.320	0.360	❌
td_agent_trace_observability	needs_review	noul	0.450	0.550	❌
td_agent_trace_observability	outcome	choice	0.440	0.500	❌
td_agent_trace_observability	risk	score	0.370	0.340	✅
td_agent_trace_observability	urgency	score	0.330	0.370	❌
td_customer_service	action	choice	0.260	0.520	❌
td_customer_service	category	choice	0.720	0.380	✅
td_customer_service	churn_risk	score	0.680	0.470	✅
td_customer_service	needs_human	noul	0.670	0.670	❌
td_customer_service	urgency	score	0.610	0.420	✅
td_invoice_processing	discrepancy_severity	score	0.450	0.470	❌
td_invoice_processing	disposition	choice	0.470	0.450	✅
td_invoice_processing	duplicate	noul	0.110	0.890	❌
td_invoice_processing	matches_order	noul	0.540	0.540	❌
td_invoice_processing	urgency	score	0.490	0.510	❌
td_security_incidents	credential_compromise	noul	0.550	0.550	❌
td_security_incidents	disposition	choice	0.280	0.740	❌
td_security_incidents	severity	score	0.420	0.490	❌
td_security_incidents	true_positive	noul	0.730	0.730	❌
td_security_incidents	urgency	score	0.340	0.500	❌

Fitted temperature per primitive (fold 0, judged on fold 1): choice 4, score 7, noul 7 — ECE 0.248 → 0.027, accuracy → 0.413.

Published numbers

Reported by the authors or by third parties, on their own suites. Not comparable to the leaderboard above; attributed so you can check them.

Model cardwfzyx

Von README and model card

Measured on: NLI held-out mixture; jabr classifier-benchmark v1 (78 cases) and v2 (869 cases)

NLI held-out (ANLI/WANLI/MNLI/SNLI): 91.23%
jabr v1 macro (8 tasks): 93.5% vs Jev 97.2%, GLiNER2 78.5%

Author-reported. The "91.23%" is an NLI held-out set, not a Jev suite. The comparison script in the repository calls only Von and LAYA, so the Jev rows are not produced by its code.

An "88.30% Jev" table quoted in secondary sources could not be found in any current revision of the README or card.

Frequently asked questions

How does Von compare to Jev?

On typed-decisions Von scores 0.461 against Jev's 0.726, beating the majority baseline on 5 of 20 questions (Jev: 18/20).

Is Von open source?

Yes. wfzyx/von-1.0 is released under Apache 2.0 and runs on your own hardware with the von (GitHub) library.

Can Von extract entities?

No. Von answers questions about text and never returns spans; use a GLiNER-family model for extraction.

Jev vs Von

Jev against Von, the System One API over a ModernBERT-large NLI head.

LAYA vs Von

The two open models that take the Jev request unchanged, on the same suites.