Ross ROSS = Recommend OSS · open-source software intelligence for agents

llm-as-a-verifier/llm-as-a-verifier

LLM-as-a-Verifier is a general-purpose framework that provides fine-grained feedback for any agent without requiring additional training. It achieves SOTA performance across coding, robotics, and medical agentic benchmarks. observed · 2026-08-28

github.com/llm-as-a-verifier/llm-as-a-verifier · homepage · Python · MIT (permissive) observed · 2026-08-28

Health v2 · maintenance only

58/100

  • Activity 98
  • Release rhythm 35
  • Longevity 10

Flags: no_releases young

How is this computed?

round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10) — computed 2026-09-02. Adoption (stars, forks) is never an input.

  • gap_med: n/a
  • age_days: 146
  • days_rel: n/a
  • days_push: 13
  • n_releases_24m: 0

Full methodology

Adoption not part of the score

2901 stars · 229 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

A Python framework that uses LLM token logprobs to produce fine-grained verification scores for agent trajectories without additional training. It provides best-of-N selection, pairwise comparison, and progress tracking APIs, and ships agentic benchmarks like Terminal-Bench and SWE-Bench Verified.

Use cases

  • rank candidate solutions from an LLM agent with best-of-N selection
  • score agent trajectories step by step for progress tracking
  • generate dense reward signals for reinforcement learning
  • verify coding agent outputs against custom criteria
  • evaluate robotics or medical agent rollouts with image inputs
  • run agentic benchmarks like Terminal-Bench and SWE-Bench Verified
  • compare two candidate answers pairwise with an LLM judge

When to choose

  • you need fine-grained, calibrated verification scores for agent outputs without training a reward model
  • you want test-time scaling via best-of-N selection with a cost-efficient tournament
  • you need dense rewards for RLHF-style training of agents
  • you want multimodal verification of image or video agent trajectories

When to avoid

  • you need a simple pass/fail test suite rather than probabilistic scoring
  • you cannot access an LLM backend that returns token logprobs
  • you need fully offline verification with no API or GPU costs
  • your evaluation requires deterministic, reproducible ground-truth checks only

Facets

library · maturity active

llm-inference machine-learning agent-framework benchmarking rag large-language-models machine-learning developer-tools python cross-platform cli llm-as-a-judge reward-model best-of-n-selection reinforcement-learning-rewards test-time-scaling pairwise-comparison progress-tracking multimodal-verification ai-agents

10 sources

Member repositories

RepositoryRoleHealth v2
llm-as-a-verifier/llm-as-a-verifiermain58

For agents

markdown · JSON · MCP: product_card(name="llm-as-a-verifier/llm-as-a-verifier")

Data as of 2026-08-30T08:39:29.467469+00:00 · Report a problem