# cvs-health/uqlm

[JMLR 2026] "UQLM: A Python Package for Uncertainty Quantification in Large Language Models"

Repository: https://github.com/cvs-health/uqlm
Canonical: https://ross.abutalabs.com/products/uqlm
Homepage: https://cvs-health.github.io/uqlm/latest/index.html
Language: Python
License: Apache-2.0
License Family: permissive
Topics: ai-evaluation, ai-safety, hallucination, hallucination-detection, hallucination-evaluation, hallucination-mitigation, llm, llm-evaluation, llm-hallucination, llm-safety, uncertainty-estimation, uncertainty-quantification, confidence-estimation, confidence-score
Last push: 2026-09-02T23:09:26+00:00

## Health v2 (maintenance only)
Score: 86/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 100, release rhythm 97, longevity 35
- inputs: {"age_days": 503, "days_push": 0, "days_rel": 21, "gap_med": 8, "n_releases_24m": 44}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 1195, forks 131 (observed 2026-09-03T02:15:10.323236+00:00)

## What it is
UQLM is a Python library for detecting LLM hallucinations using uncertainty quantification techniques. It provides five categories of scorers (black-box, white-box, LLM-as-a-judge, ensemble, and long-text claim-level) that return confidence scores between 0 and 1 for LLM responses.

## Use cases
- detect hallucinations in LLM outputs
- score confidence of LLM responses
- quantify uncertainty in large language model generations
- evaluate LLM reliability
- flag unreliable claims in long-form LLM text
- add hallucination detection to an LLM pipeline

## When to choose
- you need response-level confidence scores for LLM outputs
- you want multiple scorer types including consistency-based and token-probability methods
- you need claim-level uncertainty for long-form responses
- you want a peer-reviewed, actively maintained Python package

## When to avoid
- you need training-time uncertainty methods rather than inference-time scoring
- you cannot tolerate extra LLM calls or latency for black-box or judge scorers
- you need model internals beyond token logprobs

## Facets
- artifact type: library
- maturity: active
- function: machine-learning, llm-inference, monitoring, testing
- domain: large-language-models, machine-learning, artificial-intelligence
- platform: python
- tags: hallucination-detection, uncertainty-quantification, confidence-scoring, llm-evaluation, ai-safety, llm-as-a-judge, natural-language-processing

## Member repositories
- cvs-health/uqlm (main) score 86

## Provenance
- Observed fields: from GitHub, fetched 2026-09-03T02:15:10.323236+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T06:22:28.613696+00:00, confidence not recorded.
  - readme: https://github.com/cvs-health/uqlm (fetched 2026-09-03T02:15:10.323236+00:00, sha fd4eefe4ca59)
  - homepage: https://cvs-health.github.io/uqlm/latest/index.html (fetched 2026-08-29T12:29:56.584612+00:00, sha 8d52ff4fcbcd)
  - registry_pypi: https://pypi.org/pypi/uqlm/json (fetched 2026-08-29T12:29:56.593997+00:00, sha 655b1f3350fe)
- Data as of 2026-08-30T08:39:29.467469+00:00.