Jeeves. Reasoning improves Jev-like decision models
A reasoning Jev-style classifier with a diffusion drafter, trained with SFT and CISPO.
Jev-like models give calibrated decision probabilities, but at low accuracy. A lot of pipelines therefore rely on a reasoning model as a fallback. Jeeves trains a Jev-like Qwen3.5-9B (LoRA and a pointer head) using CISPO to reason before it decides.
This results in better performance on out of domain tasks, and outperforms Jev in JevBench hard (public).
Accuracy with thinking, greedy, 2,560-token cap. The Kev-9B and Jev columns are the numbers Kev publishes.
- No Kev-9B JevBench result is published. These are Kev-8B (Qwen3).
All JevBench numbers are on the public easy, standard and hard tiers (231 items). The sealed judge tier is not included, and the Jev and Kev numbers are restricted to the same public items.
Without thinking the same checkpoint scores 0.804 on our test split (2,962 items), against 0.840 with it.
Requirements: Python 3.12 and a CUDA GPU.
Download the released weights and serve them:
Or fuse your own trained checkpoint into a standalone model and serve it with a drafter:
Then send a request in Jev’s format:
Response on one H100 (FP8), with the three questions thinking in parallel:
sdk/ is a drop-in replacement for Jev’s Python SDK (typesafe-sdk):
The client connects to http://127.0.0.1:8009 by default (or JEEVES_BASE_URL), needs no API key, and waits up to 120s.
options is optional and ignored by Jev clients that don’t send it. Server-wide defaults are set with the matching serve flags.
On 325 dev questions:
Questions, states and answers are loaded into the Qwen chat template like
The model then rolls out its reasoning chain, and after the token we append
A pointer head scores each option with a scaled dot product between a query projection of the hidden state at
These are rare, largely unused tokens in the Qwen tokenizer. Ablations found that using plain text like “State” in the prompt instead worsened performance. Likewise, not repeating the questions after the reasoning block also decreases performance. The final probabilities are a softmax over the option scores, divided by a temperature fitted on the dev set.
Stopping at step 402 keeps the best calibration and dev score. Past it, the head over-sharpens on the saturated RL pool.
A diffusion view of the frozen model (drafter/), inspired by Orthrus.
Unlike Orthrus, which supports attention-only models, it supports Qwen3.5’s Gated DeltaNet layers by letting mask tokens cross-attend to those layers’ post-convolution keys and values.
Block 4 is the default because it stays cheap when several questions are batched.
You can build the datasets locally using the prep scripts. This downloads the public datasets from Hugging Face at the revisions pinned in prep/public.py:
Each public dataset stays under its own license.
On 8 GPUs, with the data in data/, bash run.sh runs the whole pipeline:
If you use Jeeves, its training recipe or its drafter, please cite:
Jeeves – Reasoning improves Jev-like decision models