AnyJev: Turning LLMs into Decision Models

• Fact AnyJev turns any LLM into a Jev-style decision model with typed decisions and real probabilities.[1]AITurn any LLM into a Jev-style decision model: typed decisions, real probabilities, no training.
• Fact It can serve models via vLLM or transformers backend, achieving higher decisions per second.[1]AI2.2× the decisions per second on vLLM and 2.3×–2.7× on the transformers backend, accuracy unchanged
• Fact Rotation budget reduces the number of option rotations needed for K-option choices.[1]AI7.2 rotations instead of 18 at a certified 1% disagreement rate, 2.2× the decisions per second on vLLM
• Fact A pooling server plus a few kilobytes of head constitutes an L2 deployment, requiring no logits or parsing.[1]AIAn L2 deployment is a pooling server plus a few kilobytes of head. No logits, no parsing, no patched engine, and nothing generated.
• Fact AnyJev can be used with Qwen3-8B on BANKING77, achieving high accuracy and improved calibration.[1]AIQwen3-8B on a real BANKING77 item. Every number is a model output.
• Fact The GitHub repository nokia-applied-research/anyjev was created on 21 Sept 2026 and had 1,012 stars on 2 Oct 2026.[1]
• Fact Its description reads: “Turn any LLM into a Jev-style decision model: typed decisions, real probabilities, no training. (continue updating, welcome any issue and PR request)”[1]
• Fact That is 1,012 stars in 12 days, an average of 87.6 per day.[1]

• Analysis Use AnyJev to wrap an LLM, calibrate decisions with rotation budget, and deploy efficient decision endpoints via vLLM or transformers.AI
• Analysis Automatically detected from public activity data. Not yet reviewed by an analyst.

• Fact AnyJev turns any LLM into a Jev-style decision model with typed decisions and real probabilities.[1]AITurn any LLM into a Jev-style decision model: typed decisions, real probabilities, no training.
• Fact It can serve models via vLLM or transformers backend, achieving higher decisions per second.[1]AI2.2× the decisions per second on vLLM and 2.3×–2.7× on the transformers backend, accuracy unchanged
• Fact Rotation budget reduces the number of option rotations needed for K-option choices.[1]AI7.2 rotations instead of 18 at a certified 1% disagreement rate, 2.2× the decisions per second on vLLM
• Fact A pooling server plus a few kilobytes of head constitutes an L2 deployment, requiring no logits or parsing.[1]AIAn L2 deployment is a pooling server plus a few kilobytes of head. No logits, no parsing, no patched engine, and nothing generated.
• Fact AnyJev can be used with Qwen3-8B on BANKING77, achieving high accuracy and improved calibration.[1]AIQwen3-8B on a real BANKING77 item. Every number is a model output.

Architecture

Insufficient evidenceNo architecture has been documented for this signal.

Insufficient evidenceDeveloper implications have not been assessed yet.

Insufficient evidenceNo adoption indicators recorded.

Timeline

1. 2 Oct 2026

   First seenFirst seen by the Astrava pipeline[1]

• Fact Quantization to fp8 is not recommended as it reduces accuracy.[1]AI--quantization fp8 is available and not recommended — it buys single-question latency and costs accuracy.
• Analysis Generated from activity metrics only; verify the linked sources before acting on it.

Insufficient evidenceNo competing approaches documented yet.

Insufficient evidenceNo independent projects or organisations are documented as building on this yet.