Configuration Parsing Warning:In adapter_config.json: "peft.task_type" must be a string

jeff-adapter-ground

Passage selection and answer verification. Selects the passage that best answers a question and checks whether a generated answer is supported by its sources. These are separate functions.

A LoRA adapter for jeff-base v1.4, Jeff's small open decision model (a fine-tune of Qwen3.5-0.8B). You give it a situation and a question with named options; it returns a calibrated probability for every option from one forward pass, with no generated text to parse.

When to use it

  • Your retrieval pipeline needs to select the best passage or determine that none answers the question.
  • You need to verify a generated answer against its sources before displaying it.

When not to use it

  • You need to generate an answer or identify the specific unsupported claim; this adapter returns a selection or verdict.
  • You need an assessment based on outside knowledge; this adapter uses only the supplied sources.

What it decides

What a call sends:

Field Required What it holds
sources (context) yes The source passages the answer was written from, each with a short label such as "Document A: Title" or "Source 1".
answer (input, fields) yes The answer to check.

jeff.retrieval.ground.grounding (is_supported, choice)

Instructions: Is the answer supported by the sources? Judge only from the sources, not from outside knowledge.

Labels:

  • supported: Everything the answer claims is stated in or follows directly from the sources
  • partly_supported: Some claims are supported, but at least one is not in the sources
  • contradicted: The sources state something that conflicts with the answer
  • unsupported: The answer's main claim is not in the sources at all

jeff.retrieval.ground.rerank (classify, choice)

Instructions: Which passage answers the question? If none of them does, choose none.

This function takes its own fields:

Field Required What it holds
question (input, fields) yes The question to find an answering passage for; the passages are the options, each as p1, p2, ... with its text.

Labels:

  • The caller gives the labels (at most 41), each a key and its text.

How it was measured

Base: Jeff v1.4; adapter trained 2026-10-25.

Test Rows Accuracy
passages from public data sets (SQuAD 2.0, HotpotQA, FEVER) and LLM-written documents, held out by source document, with LLM-made changes for the partly supported, contradicted and unsupported cases (mixed) 4,160 96.6%

Accuracy is the share of test rows whose top answer was right. Kinds: public, a public benchmark; real, held-out real documents; generator, an in-house test from the same generator as the training data (it measures less than a public or real test does); mixed, several kinds; unknown, not recorded.

When it is confident (from the calibration set): the threshold is the confidence from which an answer counts as confident, precision the share of confident answers that were right, coverage the share of answers that were confident.

Function Threshold Precision Coverage Calibration rows
grounding 0.634 98.0% 98.1% 720
rerank 0.797 98.0% 91.0% 720

How to call it

Install the Python SDK and the engine (see the install guide; until the packages are on PyPI, install them from github.com/firelex/jeff).

The example below is "Heat pump warranty without a service". Of three warranty clauses, only the one on missed annual services says what happens to the warranty, so the pick is p2.

On your machine (Jeff.local downloads this adapter and the base from Hugging Face the first time):

import jeff

client = jeff.Jeff.local(["ground"])

decide = client.function("jeff.retrieval.ground.rerank", labels={
    "p1": "4.1 Aster 2 heat pumps installed on or after 1 March 2026 carry a seven-year parts and labour warranty from the date of installation, subject to clause 4.3.",
    "p2": "4.3 The warranty remains valid only if the unit is serviced at least once every 12 months by a Fenwick-approved engineer. If a service is missed, the warranty ends on the date the service was due and cannot be reinstated.",
    "p3": "6.2 To book a service, call 0161 496 0123 or use the customer portal. Services are available Monday to Saturday, 8:00 to 18:00.",
    "none": "None of the passages answers the question.",
})

decision = decide("Does the warranty on my Aster 2 heat pump still apply if I skip the annual service?")

print(decision.value, decision.confidence)

Through the Jeff API (an API key from jeffhub.ai):

import os

import jeff

with jeff.Jeff.remote("https://api.jeffhub.ai", token=os.environ["JEFF_API_KEY"], timeout_s=30) as client:
    decide = client.function("jeff.retrieval.ground.rerank", labels={
        "p1": "4.1 Aster 2 heat pumps installed on or after 1 March 2026 carry a seven-year parts and labour warranty from the date of installation, subject to clause 4.3.",
        "p2": "4.3 The warranty remains valid only if the unit is serviced at least once every 12 months by a Fenwick-approved engineer. If a service is missed, the warranty ends on the date the service was due and cannot be reinstated.",
        "p3": "6.2 To book a service, call 0161 496 0123 or use the customer portal. Services are available Monday to Saturday, 8:00 to 18:00.",
        "none": "None of the passages answers the question.",
    })

    decision = decide("Does the warranty on my Aster 2 heat pump still apply if I skip the annual service?")

    print(decision.value, decision.confidence)
curl -s https://api.jeffhub.ai/v1/decide \
  -H "Authorization: Bearer $JEFF_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"calls": [{"function": "jeff.retrieval.ground.rerank", "input": "Does the warranty on my Aster 2 heat pump still apply if I skip the annual service?", "labels": {"p1": "4.1 Aster 2 heat pumps installed on or after 1 March 2026 carry a seven-year parts and labour warranty from the date of installation, subject to clause 4.3.", "p2": "4.3 The warranty remains valid only if the unit is serviced at least once every 12 months by a Fenwick-approved engineer. If a service is missed, the warranty ends on the date the service was due and cannot be reinstated.", "p3": "6.2 To book a service, call 0161 496 0123 or use the customer portal. Services are available Monday to Saturday, 8:00 to 18:00.", "none": "None of the passages answers the question."}}]}'

Recorded answer (2026-11-03): p2 with confidence 1.0.

To download the weights yourself:

hf download mstrasser/jeff-base --revision v1.4 --local-dir jeff-base
hf download mstrasser/jeff-adapter-ground --revision v1.4 --local-dir adapters/ground

Files

File What it is
adapter_model.safetensors, adapter_config.json The LoRA weights and their configuration
readout.safetensors The adapter's readout over the answer codes
decision_config.json Answer codes, temperature, prompt layout and the base it was trained on (the SHA-256 of the base's weights, which the engine checks)
manifest.yaml The adapter's manifest: its functions, labels, fields and calibration, as the SDKs read it
test.jsonl The held-out test set the numbers above come from
calibration.jsonl The calibration set the thresholds come from
LICENSE The Apache License 2.0, for the Qwen3.5-0.8B notice below

Licence

Adapter licence: Apache-2.0

Qwen3.5-0.8B notice: these weights were modified from Qwen3.5-0.8B by the Jeff project: jeff-base is a fine-tune of Qwen3.5-0.8B, and this adapter was trained on top of it. Qwen3.5-0.8B is Copyright 2026 Alibaba Cloud and licensed under the Apache License, Version 2.0; a copy of that licence is in LICENSE.

Jeff is an independent project, not affiliated with or endorsed by the makers of Qwen.

Downloads last month
118
Inference Providers NEW
This model isn't deployed by any Inference Provider. πŸ™‹ Ask for provider support

Model tree for mstrasser/jeff-adapter-ground

Adapter
(34)
this model