DATABANK

GovSim: Emergence of Sustainable Cooperation in a Society of LLM Agents

Giorgio Piatti, Zhijing Jin, Max Kleiman-Weiner, Bernhard Schölkopf, Mrinmaya Sachan, Rada Mihalcea

Five LLM agents share a common-pool resource and must agree on limits to avoid collapse. A replication of Piatti et al. (NeurIPS 2024) across three scenarios.

papersourceREPS package

game theorytragedy of the commons

Runs

READMEdownload CSV
newest first785 runs
conditionrunmodelscenariotreatmentsurvival
fa73f6702a2e20261015t215743z-c1f7abe48694Qwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionoutsider + universalization5
fa73f6702a2e20261015t215212z-c7a3eaff769fQwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionoutsider + universalization6
fa73f6702a2e20261015t214501z-69b704450a47Qwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionoutsider + universalization2
fa73f6702a2e20261015t213850z-8f26831f0afcQwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionoutsider + universalization5
fa73f6702a2e20261015t213359z-a79f492e9da2Qwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionoutsider + universalization5
59a29cfc403c20261015t212237z-1d4a36bf1e10Qwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionoutsider3
59a29cfc403c20261015t211115z-4635e3619873Qwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionoutsider2
59a29cfc403c20261015t210313z-f49e641172adQwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionoutsider4
59a29cfc403c20261015t205652z-f918a0b4f896Qwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionoutsider2
59a29cfc403c20261015t205051z-cc2ffa596b51Qwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionoutsider1
0b2ea137dbd520261015t204940z-6476aaa1bc22Qwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionno discussion1
0b2ea137dbd520261015t204800z-c3b39d87ba79Qwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionno discussion1
0b2ea137dbd520261015t204630z-1916ec7bf942Qwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionno discussion1
0b2ea137dbd520261015t204519z-b8e9f2378798Qwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionno discussion1
0b2ea137dbd520261015t204359z-e9dd2ff8a95fQwen/Qwen1.5-72B-Chat-GPTQ-Int4pollutionno discussion1
0301c291753c20261015t204008z-3b1e6faf94faQwen/Qwen1.5-72B-Chat-GPTQ-Int4sheep farmingoutsider + universalization8
0301c291753c20261015t203147z-b2f571248a9aQwen/Qwen1.5-72B-Chat-GPTQ-Int4sheep farmingoutsider + universalization8
0301c291753c20261015t200943z-010e57ad26d5Qwen/Qwen1.5-72B-Chat-GPTQ-Int4sheep farmingoutsider + universalization6
0301c291753c20261015t195541z-593017cf208cQwen/Qwen1.5-72B-Chat-GPTQ-Int4sheep farmingoutsider + universalization5
0301c291753c20261015t195240z-7b6a712fbc82Qwen/Qwen1.5-72B-Chat-GPTQ-Int4sheep farmingoutsider + universalization6
5d8fd22be4dd20261015t194950z-026ee4bd4aa4Qwen/Qwen1.5-72B-Chat-GPTQ-Int4sheep farmingoutsider1
5d8fd22be4dd20261015t194538z-dcdd6338d0f1Qwen/Qwen1.5-72B-Chat-GPTQ-Int4sheep farmingoutsider1
5d8fd22be4dd20261015t194418z-92fba82390f1Qwen/Qwen1.5-72B-Chat-GPTQ-Int4sheep farmingoutsider1
5d8fd22be4dd20261015t194127z-cf932a8435c5Qwen/Qwen1.5-72B-Chat-GPTQ-Int4sheep farmingoutsider4
5d8fd22be4dd20261015t193857z-201aa6c0e5deQwen/Qwen1.5-72B-Chat-GPTQ-Int4sheep farmingoutsider1

Run it

scenario

enum

Scenario and treatment to run.

int

Round (month) cap; 0 = the upstream per-experiment value (12 baseline, 15 outsider).

model

llm

Model used by every agent. Use mock/model for an offline test.

enum

Memory embeddings: hash gives artificial similarities for offline testing; mxbai uses the paper’s embedding model at the revision pinned in the adapter and may download about 1.3 GB.

generation

float | null

Sampling temperature; null omits it from requests (upstream default 0.0).

float | null

Nucleus sampling top_p; null omits it from requests (upstream default 1.0).

int | null

Per-call completion budget sent as a cap on every request; null sends none, so the server applies its own limit (upstream asked for 8000, which never bound in practice).

enum | null

Reasoning effort for compatible models; null leaves the provider default. For Astra use low and set temperature/top_p to null.

compute

int

CPU threads per process for PyTorch and numeric libraries.

$(nix-build --no-out-link --tarball-ttl 0 https://github.com/antimemes/REPS/archive/main.tar.gz \
  -A exec.govsim) \
  --set experiment=fish_baseline_concurrent \
  --set max_rounds=0 \
  --set model=mock/model \
  --set embedder=hash \
  --set temperature=0 \
  --set top_p=1 \
  --set max_tokens=null \
  --set reasoning_effort=low \
  --set threads=2

Deposits

  • Raphael Francis Ltd785 runs, 39 excludedCC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
    page
    https://huggingface.co/buckets/raphaelfrancis/govsim
    license
    CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/)
    excluded

    4 runs. Azure grok-4.6 returned HTTP 200 responses with empty choices during the 2026-09-18 outages (03:04:49-03:07:30Z and 00:17:58-00:19:27Z) and scattered failures. Upstream GovSim replaced these responses with default values, including harvest decisions defaulted to 0 tons. All four affected published runs are excluded. Original records remain intact in the bucket; replacements were published on 2026-10-01.

    35 runs. Superseded original tier-1 azure/mistral-medium-3-5 runs, rerun with the same seeds and parameters after the assistant-prefill fix added message-level prefix: true. The 35 replacement runs were published and verified on 2026-09-25. Original records remain intact in the bucket; the site uses the replacements.