README.md

metadata

title: QAgent
colorFrom: purple
colorTo: indigo
sdk: docker
app_port: 7860
pinned: false
license: apache-2.0
hardware: cpu-basic
tags:
  - quantum-computing
  - qaoa
  - pennylane
  - llm-agents
  - tool-selection
  - auto-deploy

QAgent - quantum tool selection for LLM agents

This Space demonstrates QAgent: picking the best subset of k tools from N candidates for an agent task, solved with the Quantum Approximate Optimization Algorithm (QAOA). It is project 2 of 3 in the Quantum Co-Processor program (after QVerify, alongside QRoute).

What you can do here

• Explore the qagent-mini-50 tasks. Pick any of the 50 benchmark tasks; the 4x4 (or 4x2) tool grid highlights which tools the brute-force optimum, QAOA, and greedy each select, with score cards and approximation ratios.
• See the score landscape. A chart plots the scores of every size-k subset with markers showing exactly where QAOA, greedy, and the optimum fall - so you can see how close QAOA gets and how far greedy misses.
• Build an exploration history. As you browse tasks, the session tracks the QAOA vs greedy approximation ratio.
• Verify live. A button runs the pure-Python brute-force and greedy solvers live (no precomputed lookup) and confirms they match the served numbers.
• Read the leaderboard. The qagent-mini-50 summary: exact-match and mean approximation ratio for brute-force, greedy, and QAOA.

Design notes

• Precomputed QAOA. QAOA on 16 qubits is slow on CPU, so the QAOA results are precomputed with the documented v0.2 config (p=4, 160 steps, 1024 shots, seed 0, x mixer) and served from JSON for an instant experience. The classical solvers run live (they are pure Python and finish in milliseconds), which keeps the served numbers honest and the Docker image small (no torch / pennylane).
• Fully open. There is no IBM hardware path and no quota to protect, so the Space has no OAuth gating - anyone can use every feature.
• Honest scope. Simulator only; no quantum advantage is claimed. QAOA matches the optimum on every N=8 task and reaches a 0.915 mean approximation ratio at N=16, but exact-match at N=16 remains hard (the next step is Dicke-initialised XY-QAOA). Full benchmarks and source: the GitHub repo.

Auto-deployed from the qagent repository main branch via a GitHub Action (HfApi.upload_folder); see docs/deploy.md in the repo.